TOPICS IN THE RECONSTRUCTION AND DEVELOPMENT OF INDO-EUROPEAN ACCENT Ronald I. Kim A DISSERTATION in
Linguistics
Presented to the Faculties of the University of Pennsylvania in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy 2002
UMI Number: 3054962
UMI Microform 3054962 Copyright 2002 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code.
ProQuest Information and Learning Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, Ml 48106-1346
To my family and friends
ii
Foreword It's rare for someone to enter a Ph.D. program with a clear idea of what he or she wants to write his doctoral thesis on, and I'm no exception. Yet when the time came in the fall of 1999 for me to choose a dissertation topic, I found myself strongly attracted to a problem that had first caught my attention four years before. During my senior undergraduate year at Princeton, as I was stumbling my way towards (barely) finishing a less-than-successful major in astrophysics, I took a graduate seminar on the historical grammar of Slavic. I had been interested in Eastern Europe and the Slavic world ever since studying some Russian in the summer of 1991 before and during a trip to the Soviet Union, but even I wasn't prepared for how eagerly I would take to Slavic historical linguistics. What especially caught my attention in Russian grammar was the complex system of stress alternations. This interest has only increased the more I learn about languages such as Serbo-Croatian, which for both linguistic and extralinguistic reasons has become my favorite. That same fall, I successfully avoided my thesis research and improved my German by reading Manfred Mayrhofer's article on the prehistory of the Iranian languages in the volume Compendium Linguarum Iranicarum. Here too, I took to the material with a passion, one that has been maintained under the surface over the years and finally culminated earlier this spring with an article on the prehistory of the stress and case system of Ossetic. Once again, however, it was Mayrhofer's brief discussion of the evidence for stress in Iranian that intrigued me, in particular his comment that the difficult task of reconstructing Proto-Iranian stress on the basis of a few scattered and/or not entirely understood phenomena in Avestan and Middle Iranian, plus data from Pashto and some other Modern East Iranian languages I had never heard of, was "perhaps comparable to that in Balto-Slavic accentology." These ideas remained in the back of my mind during my first two years of graduate school, and resurfaced when my advisors, Don Ringe and Rolf Noyer, teamed up in the fall of 1998 to teach a seminar on Indo-European accent within the context of the latest advances in metrical theory. Don, a classically trained Indo-Europeanist, and Rolf, a former student of Morris Halle, were both interested in the application of the footless "brackets-and-edges" model of stress computation to the prosodic systems of Sanskrit, Greek, the Balto-Slavic languages, and Proto-Indo-European, the reconstructible ancestor in
of the Indo-European languages. Needless to say, this seminar did a lot to stimulate my dormant interest in questions of accent and stress in Indo-European in general, and in Greek and Balto-Slavic in particular. The following fall, I wrote up a dissertation proposal. After one year on leave, two part-time jobs, two car accidents, two summers of traveling, and thousands of pages of reading, this is the result. Portions of the research in this thesis have been presented to a number of audiences. A preliminary version of chapter 3 was presented at the 24th annual Penn Linguistics Colloquium at the University of Pennsylvania, 26-7 February 2000. After extensive revision, it appeared in volume 7.2 (2001) of the University of Pennsylvania Working Papers in Linguistics (PWPL). Chapter 3 has been adapted and only slightly revised from that article, which I hope to submit to an Indo-European journal in the near future. Extremely early, poorly informed, and error-ridden versions of the Balto-Slavic chapters were read at the 19th East Coast Indo-European Conference at the University of Georgia, 3-6 June 2000 and at an invited talk sponsored by the Program in Linguistics of Princeton University on 11 April 2001. I had the opportunity to give a less embarrassing presentation at the Department of Linguistics of Swarthmore University on 21 February 2002. An embryonic version of chapter 4, particularly section 4.3, was read at the 26th Penn Linguistics Colloquium at the University of Pennsylvania, 1-3 March 2002. Finally, I had the great pleasure of presenting section 6.6, on the relationship of accentuation to present and aorist morphology in Slavic, to the 21st East Coast Indo-European Conference right here at Penn, 13-16 June 2002. To all of these institutions and conferences, I offer my sincere thanks for allowing me the opportunity to present my work as it evolved over the last two and a half years. Except for chapter 3, almost all of this thesis was written in just over three months, between early March and late June 2002. I was told in late February that I had been awarded a one-year Mellon postdoctoral fellowship at Cornell University, which was welcome and entirely expected news, but meant that I had to finish a nonexistent dissertation in time to graduate in August. I've had to cut a lot of corners and avoid too much detail (especially in chapter 1), and the thorough research I've tried to make a hallmark of my few publications so far has had to be sacrificed. Still, I hope the final product hasn't suffered too much as a result — but that's not for me to judge.
IV
The last six years have been interesting and tiresome, exciting and boring, memorable and forgettable, but the last two and a half years have been one unending wild ride. I'd like to thank some of the people who've made this thesis and degree possible and helped me to end my graduate career on a positive note. My principal advisor, Don Ringe, has been a mentor for my studies in IndoEuropean, and in historical and general linguistics, since we made contact in July 1994. I sometimes wonder how differently my life might have turned out if I had thrown out the piece of paper on which his e-mail address was written, instead of deciding to give it a shot and introduce myself. Don was largely responsible for securing a fellowship for me in 1996, and since then he has done more than I can say to encourage my growth as a scholar. Every page of this thesis indirectly reflects his standards of precision in research and argumentation and attests to the influence of his example. The person who gave me that piece of paper with Don's e-mail address was none other than Rolf Noyer, my other advisor. I was already anticipating coming to Penn when I learned that he would also be joining the Department of Linguistics there in the fall of 1996. Rolf taught my first linguistics class, an introduction to phonetics and phonology, in the spring of 1994. I'm sure he remembers some of our heated discussions after class in which I gave him an earful of my ignorance of modern linguistic science. Rolfs brilliant ability to think on an abstract level, combined with his fervent interest in "real data" and historical questions, have been a source of inspiration for me, and I hope he finds the following pages not totally lacking in theoretical interest. The third member of my dissertation committee, Gene Buckley, has always welcomed my not always intelligent questions on phonology. His class notes and advice were invaluable in the writing of chapter 1, which I hope does justice to his outstanding teaching. In addition, I'd like to express a special debt of thanks to two other professors, Gillian Sankoff and Bill Labov. Although neither has much to say about the content of this thesis, both of them have contributed enormously to my love for linguistics. A lot of my thinking about linguistic variation and change from social, historical, and acquisitional perspectives has been motivated by their seminars, whether on pidgin and creole languges, language contact, dialectology, or sound change in progress. I hope some of their influence will be apparent in the pages below.
v
I have been incredibly fortunate, particularly in the last two years, to be able to count on my family and closest friends for support. Despite the definite risk of leaving someone out, they deserve mention in print: • • •
•
• •
•
•
•
my mother, father, and sister H(eidi), who are, I think, prouder of me now than they've ever been. I love you no matter what. the rest of the Kim family, including all the ones I've never met. my best friend Kevin, who's helped me in so many ways, and shared so many good (and bad) experiences, for over 10 years and is now on his way to achieving his dreams. Mark, my closest friend in Philadelphia, who has done more than his share of commiserating with me over the ups and downs of graduate school and academia, and knows (and has) what it takes to make it. my high school friend Gerard, who's probably relieved and bemused to see me join him as an academic. my fellow grad students or "classmates" (I don't think I'll ever get used to the word "colleague") in the Department of Linguistics. To Tom, Justin, Kieran, Masato, Uri, Cassie, Na-Rae, Alan, Elsi, Sergio, Suzanne, and whoever else, thanks for putting up with me, giving me plenty to laugh about, and teaching me more than a thing or two. all the professors and older "colleagues" who've helped me on the way, from my old mentor and role model Franklin Odo to the Indo-Europeanists Craig Melchert, Miles Beckwith, David Testen, Jay Jasanoff, Alan Nussbaum, Jens Rasmussen, Josh Katz, Michael Weiss, Joe Eska, and Olav Hackstein. the folks who did the little things and don't usually get the credit: Jim Lavine and Stan Cheek, who checked out books from Princeton's Firestone Library for me; Doug and Marcus at the Bucks Co. coffeehouse on Sansom St., who shared many a work-filled afternoon with me; my roommate Veronika, who answered with patience all my annoying questions about Czech vowel length; and my students in Introduction to Formal Linguistics, Introduction to Sociolinguistics, and the the Princeton spring 2001 writing class on language, who taught me almost as much as (I hope) I taught them. my friends, Penn undergraduates and graduate students and local West Philly folks, who played such a huge role in helping me to turn my life around in the spring of 2000 with PSAS and everything that came after. To Mike, Tina, Matt, Reshma, Brian,
vi
• • •
Annie, and others, plus Alexi and Wendy, Christine and Joe, Ryan, Brandon, Maya, and all the folks at the sorely missed Comet, thanks in more ways than you know. my Aussie Internet friends Kate and Ezza. Danijela, whose support and encouragement were so critical during the dark days, when there was nothing I could do for her in return. Hvala za sve! finally, my dearest Anna, who makes me so happy.
To all of you, I express my deepest thanks, and hope that even if you don't understand more than a small fraction of this thesis, you do understand what it means. ¡Hasta la victoria siempre! Pobedicemo! Ronald I. Kim Philadelphia, 2 July 2002
vii
ABSTRACT TOPICS IN THE RECONSTRUCTION AND DEVELOPMENT OF INDO-EUROPEAN ACCENT Ronald I. Kim Supervisor: Donald A. Ringe, Jr.
Recent advances in metrical theory have opened the possibility of applying the resulting insights to the reconstruction of the Proto-Indo-European (PIE) accentual system and its development in the individual Indo-European (IE) languages. After introducing the principles of the "brackets-and-edges" framework of Idsardi (1992) and Halle and Idsardi (1995), the accentual categories of the PIE noun and verb are examined and analyzed in terms of underlying accent of roots, suffixes, and endings. As Halle (1997) has proposed, the parameter settings for determining surface stress in PIE are the same as those of IE languages such as Russian: stress falls on the first accented syllable head of the (phonological) word; if the word contains no accented morphemes, it receives default initial stress. The properties of the archaic ablaut- and stress-alternating paradigms, as well as of thematic nouns and verbs, may thus be represented as the realization of different combinations of accentual specifications: aerostatic roots, for instance, are accented and surface with constant root-stress, whereas in amphikinetic paradigms the root and suffix are unaccented, with default initial stress if the ending is unaccented and ending-stress if the ending is accented. The main outlines of this system, which is largely reconstructed on the basis of Sanskrit, have been preserved in ancient Greek: despite numerous innovations, both the parameter settings and the accentuation of particular morphemes survive in the noun and the non-finite forms of the verb. In the Balto-Slavic languages, the complex alternations of stress, intonation, and length may be derived without serious difficulty from PIE. In place of a rightward stress shift from a non-acute syllable, Proto-Balto-Slavic has undergone stress retraction to acute syllables; this hypothesis accounts for a number of previously unexplained peculiarities. The accentuation of the Balto-Slavic nominal case endings continues the inherited PIE distinction between "strong" (unaccented) and "weak" (accented) endings in the non-thematic stem classes, which has spread to the thematic
Vlll
stems independently in the two branches. Finally, the stress alternations in the simple thematic present receive a plausible historical explanation and shed new light on the accentuation of this category in (post-)PIE; the relationship among present and aorist formations and accentual paradigms in Slavic is revealed to be the direct continuation of a classical IE pattern.
IX
Table of Contents Foreword
in
Abstract
viii
Chapter 1. Introduction and Theoretical Background 1.1. 1.2. 1.2.1. 1.2.2. 1.2.3. 1.2.4. 1.3.
The reconstruction of PIE stress A theoretical framework for lexical stress Earlier treatments The evolution of "brackets-and-edges" Principles and applications of metrical bracket theory Metrical bracket theory and Indo-European Stress and accent in the IE languages
1 4 4 5 6 11 14
Chapter 2. The Proto-Indo-European Accentual System 2.1. 2.2. 2.2.1. 2.2.1.1. 2.2.1.2. 2.2.1.3. 2.2.1.4. 2.2.1.5. 2.2.2. 2.2.3. 2.3. 2.3.1. 2.3.1.1. 2.3.1.2. 2.3.1.3. 2.3.2. 2.3.2.1. 2.3.2.2. 2.3.2.2a. 2.3.2.2b. 2.3.2.2c.
Introduction: coexisting systems in Proto-Indo-European grammar The noun Archaic ablauting stems Root nouns with alternating stress Aerostatic root and suffixed nouns Amphikinetic stems Proterokinetic stems Hysterokinetic stems Thematic stems Collectives The verb Athematic stems Root presents and aorists with alternating stress Narten presents Hysterokinetic alternations: the optative and nasal-infixed presents Thematic stems Thematized presents and aorists Suffixed thematic presents: *-yé/ó-, *-s£é/ó-, *-éy e / 0 Zero-grade presents in *-yé/
x
19 21 21 24 26 31 37 43 46 49 54 54 55 58 61 63 64 66 66 69 71
Chapter 3. Continuity amid Innovation: The Development of Proto-Indo-European Accent in Greek 3.1. 3.2. 3.3. 3.3.1. 3.3.2. 3.3.3. 3.4. 3.4.1. 3.4.2. 3.5. 3.6.
Introduction Was PIE accent preserved in ancient Greek? Relics of PEE accent in the Greek verb Exceptions to recessive verbal accent Survivals of lexical accent: the thematic aorist Survivals of lexical accent: the aorist passive and the perfect PIE nominal accentuation in Greek o- and a.7ε:-stems Consonant- and semivowel-stems Implications for the phonological prehistory of Greek More innovations in Greek nominal accent: continuity in disguise?
75 76 79 79 83 87 89 90 92 94 96
Chapter 4. From Proto-Indo-European to Balto-Slavic: Dybo's and Hirt's Laws Reconsidered 4.1. 4.2. 4.2.1. 4.2.2. 4.2.3. 4.3. 4.3.1. 4.3.2. 4.3.3. 4.4. 4.4.1. 4.4.2. 4.4.3.
Introduction: some methodological considerations Background The evidence The prosodic inventories of Lithuanian and Proto-Slavic Accentual classes and accentual paradigms "Dybo's Law" and the origin of accented and postaccenting stems Hirt's Law: Dybo's Law reversed? A restatement of Hirt's Law within the brackets-andedges framework Nominal compounds and the rise of accented stems From b and c to a and b: the prehistory of Baltic accentuation Proto-Baltic or Pre-Lithuanian? The evidence of Old Prussian How did postaccenting nouns become unaccented?
XI
101 103 104 110 120 129 130 138 143 145 145 146 147
Chapter 5. Making Historical Sense of Synchronic Complexity: The Proto-Indo-European Origins of Balto-Slavic Nominal Accentuation 5.1. 5.1.1. 5.1.2. 5.1.3. 5.1.4. 5.2. 5.3.
Reconstructing the accentuation of Proto-Slavic nominal inflection Oblique plural endings, and the accentuation of o-stems in Proto-Slavic Unstressability and reanalysis in a-stems Proto-Slavic consonant-stems: a lack of evidence? Synthesis Lithuanian, Old Prussian, and Proto-Balto-Slavic nominal accentuation Conclusions: PIE survival and Balto-Slavic innovation
150 154 162 166 168 170 176
Chapter 6. More Surprises in Store: The Evidence of Balto-Slavic Accentuation for the Proto-Indo-European Verb 6.1. 6.2. 6.3. 6.3.1. 6.3.2. 6.3.3. 6.4. 6.5. 6.6.
Conclusion.
Introduction: simple thematic presents in (post-)PIE Extra-Balto-Slavic evidence for the accentuation of the simple and characterized thematic present The accentuation of the simple thematic present in Proto-Slavic The evidence A postaccenting thematic vowel in Slavic? The evolution of the alternation in Proto-Slavic and the Slavic languages: evidence for the brackets-and-edges model Baltic evidence for the alternation? The PIE origins of the Balto-Slavic postaccenting thematic vowel: athematic verbs and primary endings The distribution of presents and aorists in Proto-Slavic: an unexpected Indo-European inheritance The importance of theory for historical linguistics
Bibliography
184 186 188 188 194 196 199 201 209
228 229
xii
Abbreviations Grammatical Terms sg. = singular pl(ural) fem(inine) masc(uline) neut(er) acc(usative) dat(ive) gen(itive) instr(umental) loc(ative) nom(inative) voc(ative) act(ive) aor(ist) fut(ure) impf. = imperfect(ive) inf(initive) iptv. = imperative mid(dle) mp. = mediopassive opt(ative) pass(ive) pf.=perfect(ive) pres(ent) ptcp. = participle subj(unctive) Languages Aiol(ic) Anat(olian) Att(ic) Bulg(arian) Cz(ech) Dor(ic) Gmc.=Germanic Goth(ic)
Gr(eek) H(ieroglyphic) Luv(ian) Hitt(ite) Hom(eric) Ion(ic) Lat(in) Luv(ian) Lyc(ian) Myc(enaean) O(ld) C(hurch) S(lavonic) O(ld) E(nglish) O(ld) H(igh) G(erman) O(ld) Ir(ish) O(ld) N(orse) O(ld) P(ersian) O(ld) Pr(ussian) P(roto-) P(roper) N(ame) Russ(ian) Skt. = Sanskrit S(erbo-)C(roatian) Sl(avic) Slov(ene) T(ocharian) A/B Ukrainian) Ved(ic)
Chapter 1 Introduction and Theoretical Background 1.1. The reconstruction of PIE stress Until the 1960s, the main outlines of Indo-European (IE) historical linguistics and the reconstruction of Proto-Indo-European (PIE) would have remained recognizable to a late 19th-century scholar of the Neogrammarian generation. Although the first decades of the 20th century witnessed an enormous amount of research on the philology and historical phonology and morphology of individual IE languages, the student of Indo-European could still acquire a solid foundation, and in many cases a correct impression of the communis opinio, from Brugmann's Grundriß der vergleichenden Grammatik der indogermanischen Sprachen, the standard Neogrammarian reference. In the past generation, the situation has changed drastically. Not only has the evidence of the two subgroups of IE languages discovered at the turn of the 20th century, Anatolian and Tocharian, begun to be incorporated more fully and consistently into IE linguistics, but the incorporation of new theoretical approaches to linguistics — above all the generative revolution, beginning in the late 1950s — and an increasingly improving understanding of the nature, causes, and diffusion of linguistic change, based upon research in first- and second-language acquisition and sociolinguistic studies of change in progress today, have entirely reshaped the wider intellectual context of the field. Since the 1970s, the influence of these and other "modern" areas of linguistics on IE studies has been steadily increasing, and Indo-Europeanists have increasingly engaged in phonetics and phonological theory, historical syntax from a generative perspective (e.g. Garrett 1990, Taylor 1990, 1994, Eska 1998), and computational approaches to genetic relationship (e.g. Ringe et al. 1998), to name only a small sample of recent, theoretically informed publications. Few aspects of IE linguistics are in a more advantageous position to benefit from these developments than the reconstruction of the accent and stress system of PIE and its evolution in the various IE languages. In recent decades, the "kinetic revolution" has contributed enormously to our understanding of the stress system of PIE, particularly the intricate relationship between alternations of stress and those of vocalism, or "ablaut". Meanwhile, progress in the interpretation of the synchronic and historical phonology of Hittite and the other Anatolian languages (cf. especially Melchert 1994a and now Kimball
1
1999) has permitted the sometimes startlingly archaic evidence from these languages, particularly Hittite, to be compared on a secure footing with the rest of IE. The remarkable efflorescence of Balto-Slavic accentology since the appearance of Stang's 1957 monograph, dominated by the names of Dybo, Illič-Svityč, and Garde, has resulted in a greatly improved understanding of the reconstructed Proto-Slavic and Proto-Balto-Slavic accentual systems and their evolution in the various Baltic and Slavic languages. Last but certainly not least, the rise of generative phonology has contributed to accentology as well: the past 25 years have witnessed enormous advances in the development of theoretical approaches to the analysis of accent and stress in human language. Unfortunately, the integration of these disparate fields of research into the problem of PIE accent has been limited, belated, and not altogether successful. For one, the results of recent research in Balto-Slavic accentology have not, in general, become widely known to nonspecialist Indo-Europeanists, even those interested in problems of accent. As a result, the comparative basis for the reconstruction of PIE accent has in practice remained limited to the "classical" trinity of Sanskrit, Greek, and (where Verner's Law reflexes permit) Proto-Germanic (see § 1.3). Moreover, theoretically oriented studies of accent and stress in IE languages have been few and far between: with the exception of a few articles by Paul Kiparsky (e.g. 1967, 1973) on IE, Greek, and Slavic and by Morris Halle (1973) on Russian, attempts to interpret the accentual system of PIE and BE languages with lexical accent, especially Russian and Sanskrit, did not make much of an impact until the 1990s. To be sure, the application of traditional methods of historical reconstruction has led to significant progress in reconstructing the general principles of accent and stress in PIE and the older IE languages, from Brugmann's Grundriss and Hirt's 1895 monograph to recent handbooks such as Rix 1976 or Melchert 1994a. To take only the most obvious example, the discovery of the entire system of ablaut-alternating classes in the noun and verb was a spectacularly successful result of the classical comparative method, in which analogical innovations and levelings were identified and separated from archaic retentions. Such basic principles of historical linguistics as the irreversibility of mergers will play a role more than once below: cf. the Vedic-Greek stress correspondences in §3.1, which are simply incompatible with any hypothesis which proposes that PIE lexical accent was lost in the prehistory of Greek. Yet the historical reconstruction and development of stress patterns in IE raise a number of other important questions which have not been satisfactorily addressed in previous studies. When one hypothesizes that different patterns reconstructed for PIE or a 2
prehistoric stage of an IE language have "influenced each other", exactly what does that mean? If, for instance, Balto-Slavic has "generalized mobile stress", say from the proterokinetic or amphikinetic inflection, to thematic stems, exactly how is that generalization implemented in the grammar, and how does it affect the underlying representation of the suffixes and endings? Most importantly, how are the various accentual classes and alternations reconstructed for PIE represented in the synchronic grammar of the protolanguage, and how are the accentual classes and alternations actually observed in languages such as ancient Greek, Russian, or Lithuanian to be analyzed synchronically? Only through a more precise investigation of the underlying prosodic structures can we hope to refine our reconstruction of the PIE system and improve our understanding of the historical developments which have taken place from PIE to the attested languages. The present study attempts to address some of these questions in the reconstruction and development of IE accent, within what appears to be the most promising theoretical framework to date: the metrical bracket (or "brackets-and-edges") theory of Idsardi 1992 and Halle and Idsardi 1995, building upon earlier work such as Halle and Vergnaud 1987. In the rest of this chapter, I introduce the precursors and basic principles of this theory (§ 1.2) and review the empirical evidence from IE languages for the reconstruction of PIE accent (§1.3). The accentual classes of the noun and verb in PIE are presented and analyzed according to the brackets-and-edges theory in chapter 2. The subsequent chapters investigate the evolution of this reconstructed PIE accentual system in two IE branches, Greek and Balto-Slavic. In chapter 3,1 demonstrate, contrary to some recent studies, that PIE lexical accent has been preserved in Greek, although altered and not infrequently reinterpreted as a result of significant innovations. Finally, chapters 4 through 6 investigate several topics in the accentual history of BaltoSlavic. namely the question of rightward vs. leftward shifts of stress (Dybo's and Hirt's Laws) and the paradigmatic alternations of the noun and verb. I reformulate the major findings of the last generation of scholars in terms of metrical bracket theory, and review the various developments which have produced the accentual systems of Lithuanian, Old and Modern Russian, Serbo-Croatian, and other Baltic and Slavic languages which have preserved lexical stress or reflexes thereof.
3
1.2. A theoretical framework for lexical stress 1.2.1. Earlier treatments From the birth of generative phonology in the late 1950s (cf. Halle 1959, 1962, Chomsky and Halle 1968) until the mid-1970s, it was customary for phonologists to treat stress as a property of particular vowels or other stress-bearing segments. Just as distinctively long or back vowels were marked with the phonological features [+long] or [+back], vowels were analyzed as bearing a multivalent feature [stress], i.e. one which could assume three or more values (e.g. 1 to 5, with 0 = unstressed). For instance, the English word extraterrestrial, with main stress on the third e and secondary stress on the first e, could be represented as [230130] or the like. This interpretation of stress was successfully challenged by Liberman in his 1975 dissertation, The Intonational System of English, which heralded the birth of metrical theory as a distinct subfield of phonology. Liberman argued that unlike, say, vowel length or consonantal place of articulation, stress does not correspond to a feature on a particular segment. In favor of this view, he noted the following idiosyncrasies of the distribution and behavior of stress: •
1. culminativity: in general, words contain only one main stress. This property is alien to normal segmental features: one does not expect to find e.g. one long vowel per word, or one back vowel. In fact, one commonly finds words with e.g. multiple back vowels, or composed entirely of vowels of a particular feature (vowel harmony).
•
2. rhymicity: stresses do not cluster together, but tend to be evenly distributed. True segmental features, on the other hand, tend to spread and/or be found in neighboring segments, e.g. [back] or [voiced] (cf. 1).
•
3. lack of assimilation: stress, unlike the other features just mentioned, does not spread to adjacent syllables; when a stress shift occurs, as commonly in the world's languages, the previously stressed syllable becomes unstressed, e.g. Engl, thirteen —> thirteen tables (see below).
•
4. multivalency: if stress is a segmental feature, it would be the only feature which assumes more than two values, in contrast to the binarity of e.g. [high], [voiced].
Based on these facts, Liberman proposed instead that stressed vowels are not themselves marked in distinction to unstressed vowels. Rather, stress is the phonetic manifestation of phonological grouping of segments into larger units, usually called "feet".1 'Depending on the language, there may be several criteria for determining which syllables are stressed and which unstressed. Hayes (1995:9-21) summarizes the diagnostics for English, e.g.
4
The subsequent development of metrical theory and rise of competing formalisms are reviewed in Kager 1995. In brief, the first theories (Liberman and Prince 1977, Hayes 1981) made use of binary trees, which consisted of strong and weak branches representing relative prominence, and metrical grids, which depicted the stress information encoded in the trees. Alternative proposals made sole use of trees (e.g. Kiparsky 1979) or grids (Prince 1983), but these encountered complications in evaluating rhythm based on trees alone and in accounting for various manifestations of constituency (feet), respectively. In two groundbreaking studies, Hayes (1987, 1995) successfully accounted for a wide range of accentual patterns in the world's languages by appealing to different types of binary feet. Hayes (1981) had earlier proposed a symmetrical investory of feet, which could be either left- or right-headed and quantity-sensitive or quantity-insensitive. In his later studies, this system was abandoned in favor of an asymmetric inventory, consisting of moraic trochees, syllabic trochees, and iambs (1995:62-85); each of these possibilities is illustrated with data drawn from a wide variety of languages (pp. 125-269). One possible drawback of Hayes's theory, however, is its distinction between the roles of syllable weight and syllable prominence in stress patterns (pp. 270-306): alongside or in place of quantity-sensitive trochees and iambs, languages may distinguish one or more levels of relatively more "prominent" syllables which tend to attract stress, e.g. "superlong" syllables of the shape CV:C, syllables containing certain segments or marked for other features, etc. Yet it is difficult to make a clear crosslinguistic separation between traditional mora-based notions of syllable "weight", i.e. of "heavy" and "light" syllables, and prominence: in many languages the two are identical, whereas in others prominence is defined largely or entirely without respect to moras or syllable codas. 1.2.2. The evolution of "brackets-and-edges" In 1987, Halle and Vergnaud published An Essay on Stress, which outlined an approach to metrical theory by means of bracketed grids. Although their theory gave great importance to constituency, feet of the sort espoused by Hayes are not invoked; instead it is the constituent markers, or brackets, which play a central role. In order to represent binary effects, extrametricality, or syllable prominence, the two authors make use of various operations involving the projection or insertion of brackets into the grid.
reduction of unstressed vowels or aspiration of voiceless stops or the pattern of different intonational tones.
5
The next step in the evolution of this line of research was marked by Idsardi's 1992 dissertation, The Computation of Prosody. The most salient innovation of this approach is the complete elimination of feet or any other independent constituents of prosodic computation.
Stress is determined on the basis of metrical "grids" composed of
potentially stress-bearing marks or elements, denoted by asterisks, which represent all those segments capable of bearing surface stress. These elements are grouped into larger units by various left or right brackets, whose placement is specified in a variety of ways (see below, § 1.2.3). These groups of stress-bearing units, called constituents, outwardly resemble metrical feet, but their extent and position are entirely determined by the brackets: thus feet play no independent role of their own in this model. Idsardi's theory, which will be referred to below as metrical bracket theory, has since been introduced into the mainstream of phonological theory: thanks to Halle and Idsardi's 1995 article in The Handbook of Phonological Theory (Goldsmith 1995), it is widely known even to phonologists not specializing in prosodic issues. As we shall discover, the specific principles of metrical bracket theory make it eminently applicable to the peculiarities of stress patterns in IE languages and the reconstructed prosodic system of PIE. 1.2.3. Principles and applications of metrical bracket theory As noted above in § 1.2.2, stress in the "brackets-and-edges" model is computed through a series of constructions on a metrical grid. Just as much of the observable variation in syntactic behavior across human language may be represented as the instantiation of different choices for particular parameters (one of the major premises of classical generative syntax, or the theory of "principles and parameters"), the stress patterns of languages are distinguished by the selection of a relatively small set of parameters. The steps in the determination of stress on the metrical grid are listed below, with parameters in italics; cf. Halle and Idsardi 1995:406ff. 1) Line 0 Projection Grid marks, or elements, are projected from certain segments in the phono logical structure. Most languages limit Line 0 Projection to syllable heads — typically vowels — but other phonemes may also project a grid mark, e.g. sonorants in Lithuanian syllables containing vowel + sonorant. 2) Syllable Boundary Projection (optional) Project the left/right (L/R) boundary of certain syllables onto line 0. 6
The relevant feature may be phonetic, e.g. length, or lexical specification of underlying accent, as in Russian, ancient Greek, and PIE (see below). 3 a) Edge-Marking Parameter Insert a L/R bracket to the L/R of the L/R-mosl element in line 0. The brackets define the line 0 grid marks into metrical constituents, similar to the types of feet specified in earlier theories. Here, however, the constituents are not themselves the parameter, but are determined by the placement of brackets within the grid. 3b) Iterative Constituent Construction (ICC; optional) Beginning from the R/L-most L/R bracket, insert a L or R bracket, respectively, after each pair of elements. 3c) Head Location Parameter Project the Z/R-most element of each constituent onto line 1. 4a) Edge-Marking Parameter Insert a L/R bracket to the L/R of the Z/R-most element in line 1. 4b) Head Location Parameter Project the L/R-most element of each constituent onto line 2. 5) Conflation (optional): In languages lacking secondary stress, conflate lines 0 and 1. The segment whose corresponding element is marked with an asterisk on line 2 then receives primary phonetic stress. In languages lacking conflation, one or more of the elements with line 1 projections may receive secondary stress. The great strength of metrical bracket theory is that it allows the simple computation of many stress patterns found in the world's languages, based on a just handful of parameter settings. Equally important is that the theory does not appear to overgenerate, i.e. predict stress patterns unattested in human language. In particular, the parameter of Syllable Boundary Projection allows elements with a particular phonological property to be distinguished for metrical purposes, e.g. syllables containing a long vowel, or ending in a particular segment, or lexically specified with underlying accent. All of these possibilities, along with others, are found in the languages of the world: some of them are illustrated in the case studies below. As a first example, let us consider the computation of stress in Latin, an IndoEuropean language which has lost lexical stress (see §1.3). In Latin, the position of stress is entirely predictable from the phonological properties of the stress domain. Stress is 7
confined to the final three syllables: if the penultimate syllable is "light", i.e. contains a short vowel with no coda consonant, stress falls on the antepenult; if it is "heavy", i.e. contains either a long vowel or a short vowel followed by a coda consonant, stress falls on the penult. dú.kimus amá.wimus virgtnita.s perpendiculairissimus
'we lead' 'we (have) loved' 'virginity' 'most perpendicular'
du.ké.mus ama:wisti: virginitá.tis perpendiculá.ris
'we will lead' 'you (have) loved' 'virginity (gen.)' 'perpendicular'
Under a foot-based theory of stress, such a pattern may be accounted for with final-syllable extrametricality and the construction of a moraic trochee: hence du: (ké:) mus, perpendicu (la:) ris but (dú: hi) mus, vir (gi ni) ta:s (Hayes 1995:91-2). Both of these devices may be captured in metrical bracket theory by means of parameter settings: RLR Edge Marking to exclude the final element of line 0 of the grid,2 and Boundary Projection of heavy syllables (CV:, CVC) combined with right-to-left Iterative Constituent Construction. The full settings for Latin stress computation are hence the following: 1) Project syllable heads (vowels only) 2) Project L boundary of heavy syllables, i.e. those containing a long vowel and/or coda consonant 3a) Edge Marking: RLR 3b) Iterative Constituent Construction: right-to-left 3c) Head Marking: L 4a) Edge Marking: LLL 4b) Head Marking: L In Latin, heavy syllables, but not light syllables, project a left bracket onto line 0 of the metrical grid. The following stress-computing operations are sensitive to the presence of this bracket on the element corresponding to the penultimate syllable: if no bracket is present, leftward ICC builds a left bracket to the left of the antepenultimate syllable head, which according to the parameter settings for Line Projection receives the stress; if the penultimate syllable is long, it will receive the stress. The metrical grids for virginita.s and perpendiculd:ris are hence constructed as follows: 2
Certain types of extrametricality do not seem to lend themselves readily to an analysis with edge- and head-marking; see Buckley 2000 for details. Modern English presents an especially complex stress system in which syllable weight and the ICC interact with categorially conditioned extrametricality: cf. such minimal pairs as retard (verb) vs. retard (noun), . For an analysis of English stress in the framework of Optimality Theory, see Hammond 1999:149ff.
8
1) Project syllable heads (vowels only)
vir gi ni ta:s
per pen di cu la: ris
2) Project L boundary of heavy syllables vir gi ni ta:s3
per pen di cu la: ris
3a) Edge Marking: RLR vir gi ni ta:s
per pen di cu la: ris
3b) Iterative Constituent Construction: right-to-left vir gi ni ta:s
per pen di cu la: ris
3c) Head Marking: L
vir gi ni ta:s
per pen di cu la: ris
4a) Edge Marking: RRR
vir gi ni ta:s
per pen di cu la: ris
4b) Head Marking: R
vir gi ni ta:s
per pen di cu lá: ris
3
In order to prevent a word-final heavy syllable from projecting a left boundary. Halle and Idsardi (1995:424-30) propose a constraint "Avoid * ( * #". This has the additional benefit of allowing monosyllabic words — all of which contain a heavy syllable as a result of the minimal word requirement of Latin — to project a boundary and so receive stress.
9
In addition to Syllable Boundary Projection and the construction of iterative binary "feet", this analysis involves another crucial component of metrical bracket theory, namely the capacity for marking extrametricality. The parameter settings given above for Latin may be altered to account for the stress placement of any other language in which stress is confined to certain subsets of the phonological word and/or is determined entirely by syllable "weight", i.e. the marking of certain syllables. A simple example of the former possibility is the Slavic Macedonian literary language, where stress uniformly falls on the initial or only syllable of words of fewer than three syllables, and on the antepenult of words of three or more syllables. Hayes (1995:205) analyzes this pattern with final-syllable extrametricality, as in Latin, and a syllable trochee.
Under the brackets-and-edges approach, the stress pattern of
Macedonian may be captured by means of the following parameter settings: 1) Project syllable heads (vowels only) 3a) Edge Marking: RLR 3b) Iterative Constituent Construction: right-to-left 3c) Head Marking: L 4a) Edge Marking: RRR 4b) Head Marking: R The metrical grids for kniga 'book', vodéničar 'miller', vodeničari
'millers', and
vodeničári-te 'the millers' are thus *
(* *\
*
* *\
* * *•) *\ *
*
vo dé ni čar
kni ga * *
*\ * *\
*
*)
*
vo de ní ča ri
*
*
#\
#
*~\ *\
*
*-v *
vo de ni čá ri te
Let us now examine a language whose stress pattern demonstrates the need for a powerful Syllable Boundary Projection parameter. In Kwak'wala, a Wakashan language of British Columbia, Canada, stress falls on the first "heavy" syllable — i.e., on the first syllable which is specified by Syllable Boundary Projection as projecting a left bracket onto line 0 of the grid. If there are no heavy syllables, the final syllable is stressed: unlike Latin and Macedonian, Kwak'wala is an "unbounded-stress" language, i.e. it does not limit 10
stress to a particular subset of the word. But whereas all coda consonants count for "heaviness" in Latin (i.e. both V: and VC project a syllable boundary), while other languages do not count coda consonants (so that only long vowels project a syllable boundary; see §1.2.4 on Khalkha Mongolian), Kwak'wala syllables are "heavy", and project a boundary onto line 0, if the vowel is long or if the coda consonant is a nonglottalized sonorant. Obstruents and glottalized sonorants, on the other hand, do not count for weight: hence mánsa is stressed on first syllable, but taX'ćá on the second (cf. Bach 1975, Zee 1988 for details). qá:sa ?axwsó:le makw9lá mánsa maxánxand taX'ćá
'to walk' 'hellebore (a plant)' 'moon' 'to measure' 'to strike edge' 'to warm oneself
That sonorants in coda position, but not obstruents, count for syllable weight is crosslinguistically unexceptional, but the contrasting behavior of glottalized vs. nonglottalized sonorants is not obviously compatible with the traditional definition of syllable weight. Syllable Boundary Projection, however, readily allows for such cases as Kwak'wala, which raises the possibility that even more idiosyncratically defined sets of syllables may also be specified by this parameter; we shall discover that this is exactly the situation in Indo-European. 1.2.4. Metrical bracket theory and Indo-European We have seen in the examples above that Syllable Boundary Projection readily accounts for the empirical fact that certain syllables are treated differently from others in the computation of stress. Such a powerful parameter is necessary to account for the wide range of definitions of syllable "weight" or "heaviness" across languages: closed syllables of the shape CVC project a bracket onto line 0 in some languages but not others; and within one and the same language, the exact nature of the coda consonant can be crucial, cf. Kwak'wala mánsa vs. taX'ćá. Nevertheless, the stress patterns examined above all share the property that the syllables which project a bracket onto line 0 of the grid are defined by (auto)segmental properties, e.g. long vowel and/or coda consonant in Latin, long vowel or non-glottalized coda sonorant in Kwak'wala. Another example of such a language is Khalkha Mongolian,
11
in which syllables containing a long vowel, and no others, project a left boundary onto line 0. The parameters for stress computation in Khalkha Mongolian are listed below, along with grids for xoltabara 'leadership' and gará.sa: 'from one's own hand': 1) Project syllable heads (vowels only) 2) Project L boundary of long vowels 3a) Edge Marking: RRR 3c) Head Marking: L 4a) Edge Marking: LLL 4b) Head Marking: L * r* *
xol
*
*
*)
*
ta
ba
ra
ga
* (* (*
* (*)
rá:
sa:
But if Syllable Boundary Projection allows a particular subset of elements to project an additional bracket, it should be possible for that subset to be determined not only by overt segmental features such as vowel length or coda consonant, but by more abstract properties, such as membership in a lexically specfied class. In fact, a number of Indo-European languages exhibit stress patterns whose analysis depends on the assumption that certain syllable heads are lexically specified for Syllabic Boundary Projection. Consider the paradigms of the three Russian nouns dúm-a 'thought, Duma', borod-á 'beard', and gospož-á 'lady', all feminine and belonging to the "a-stem" class. nom. sg. gen. dat. acc. instr. loc.
dúm-a dúm-y dúm-e dúm-u dúm-oj(u) dúm-e
gospož-á gospož-í gospož-é gospož-ú gospož-éj(u) gospož-é
borod-á borod-ý borod-é bórod-u borod-ój(u) borod-é
nom. pl. gen. dat. acc. instr. loc.
dúm-y dúm-0 dúm-am dúm-y dúm-ami dúm-ax
gospož-í gospož-éj gospož-ám gospož-í gospož-ámi gospož-áx
bórod-y boród-0 borod-ám bórod-y borod-ámi borod-áx
How can a theory of stress account for the coexistence of these accentual paradigms? The anwer is simple: Halle and Idsardi (1995) and Halle (1997) postulate that certain 12
morphemes, including roots, affixes, and endings, are underlyingly marked for projecting a left bracket onto line 0. These are henceforth referred to as accented morphemes. Surface stress in Russian is determined according to the same parameters as in Khalkha Mongolian: stress falls on the first underlyingly accented syllabic head, rather than the first long vowel; if the phonological word contains only unaccented elements on line 0, it receives default initial stress. This pattern is illustrated below with derivations of the instr. pi. and ace. sg. of the three nouns above. 1) Project syllable heads (vowels only) 2) Project L boundary of heads lexically specified for accent 3a) Edge Marking: RRR 3c) Head Marking: L 4a) Edge Marking: LLL 4b) Head Marking: L unaccented borod-(ámi bórod-u
accented (dúm-(ami (dúm-u
Instr. pi. Ace. sg.
* *
(dum a mi * (* (*
*
*
* (* (*
postaccenting gospož(-(ámi gospož(-ú
*
(* /*
(*
*\
bo rod á mi
*
*/•
dum u
(* *
*\
*
* *)
i*
gos pož á mi
*
*)
*
*(
(* *\
gos pož ú
bó rod u
Because the stem dum- is accented and is the leftmost morpheme, it always surfaces with stress. Similarly, the forms of the postaccenting stem gospož-, with a left bracket to the right of the second and last stem syllable, exhibit uniform stress on the post-stem syllable, i.e. the first or only syllable of the ending. In contrast, the stem borod- is unaccented, so that the position of stress depends on the accentual specification of the ending. If the ending is accented, the stress falls on the ending, e.g. in nom. sg. borod-(a > borodá. If, however, the ending is unaccented, like the stem, then the entire form is unaccented, and surfaces with default stress on the leftmost syllable as a result of the Head Marking parameter: hence e.g. ace. sg. borod-u > bórodu.
13
This analysis of Russian stress may be extended to other Balto-Slavic languages such as Serbo-Croatian and Lithuanian, as Halle demonstrates (1997:286-90; cf. §4.2.3 and chapter 5), as well as other ancient IE languages with lexical stress such as Sanskrit (ibid., 291-5) and Greek. As it is extremely unlikely that all these related languages could have independently evolved such a system, we may deduce that the PIE accentual system was characterized by underlying lexical specification of morphemes: all roots, prefixes, suffixes, and inflectional endings were either accented or unaccented, with different combinations producing different paradigmatic alternations.4 Furthermore, the parameter settings given above for Russian also apply to the other languages mentioned, so that one may tentatively conclude that they were also valid in the protolanguage. In all, the theoretical framework of Idsardi and Halle in my opinion provides the crucial insight into lexical stress in Indo-European and best captures the phenomena described above. The following chapters investigate the reconstruction and development of IE accent and stress within that framework, and also to what extent the stress phenomena of PIE and the IE languages may be accomodated under a bracketed grid approach. 1.3. Stress and accent in the IE languages Before proceeding to the analysis of the PIE accentual system, let us review the evidence from the various IE languages for the reconstruction of PIE accent. As noted in § 1.2.4, the prosodic systems of a number of IE languages are characterized by lexical accent, in which individual morphemes may be specified with an underlying, inherent accent. Other languages have lost lexical accent and generalized fixed stress; among them, some (e.g. Germanic) provide indirect reflexes of the original placement of stress, whereas others (e.g. Latin) are without value for purposes of accentual reconstruction. Traditionally, the reconstructed system of PIE stress has been based on the evidence of Sanskrit, more specifically of the preclassical stages of the language from the Vedas up through the grammarian Panini. Of all the IE languages, Sanskrit preserves by far the most synchronically transparent accentual system and the widest range of stress alternations in both noun and verb, as well as other probably archaic phenomena such as postlexical destressing (enclisis) of finite verb forms in non-clause-initial (or non-verse4
From here onward, stress with respect to the IE languages refers to an observable (surface) phonetic phenomenon: a stressed segment (usually a vowel) is distinguished by higher pitch, greater intensity, etc. On the other hand, accent is an underlying property of particular segments or morphemes. Thus in Russian instr. pi. dúm-ami 'with, by means of thoughts', both the stem (dum- and ending -ami are accented (see above), but only the stem vowel u is stressed.
14
initial) position in main clauses. The grammatical tradition clearly distinguishes between vowels which are uddtta 'raised' and which are an-uddtta 'not raised'; the svarita, or circumflex, results from an originally accented i or u which has become a glide before a following a or at, hence e.g. ya < ia. For further details, see Whitney 1924:28-34. Alongside Sanskrit, ancient Greek has also figured in discussions of PIE accent, but usually in an auxiliary role. The reason for this is not far to seek: although Greek has preserved a number of stress alternations reflected in Sanskrit, e.g. in root nouns (§2.2.1.1), the restriction of stress to the final three syllables of the (phonological) word, and the near-total generalization of recessive stress in finite verbal forms (§3.3.1), have destroyed much of the lexical information and accentual alternations preserved in Sanskrit. It is no surprise, then, that Greek data has typically been cited in support of PIE reconstructions which have been projected backwards from Sanskrit. We shall see in chapter 3 that despite some recent claims, stress in Greek may be computed by means of the same parameters as in other IE languages with lexical stress, and that the Greek accentual system does continue that of PIE; moreover, several important innovations, particularly in the noun, may be explained starting from the relevant PIE preforms. The discovery at the turn of the 20th century of the IE languages of 2nd- and 1 stmillennium BC Anatolia added a third comparandum of great antiquity to Indo-Aryan and Greek, but the proper evaluation of stress in Hittite and Luvian was hindered for decades by difficulty in interpreting the cuneiform writing system. It is now generally accepted that the "plene" writing of certain vowels by means of an additional cuneiform sign, e.g. Hitt. e-eš-zi 'is', denotes length; because short vowels are in many cases lengthened under stress, and long unstressed vowels are rare in both Hittite and Luvian, it is often possible to determine the position of stress from such plene spellings, e.g. e-eš-zi [ístsi], underlyingly /és-tsi/. To be sure, Hittite (and even more so Palaic and Luvian) has leveled many of the inherited stress alternations, especially in the noun. Nevertheless, a significant number of Hittite stress patterns agree with those of Sanskrit, and the language not infrequently preserves precious archaisms. More importantly, a growing number of scholars now believe that Anatolian was the first branch to diverge or "branch off from the rest of the PIE speech community, which makes it necessary to investigate how much of the accentual system reconstructed from Sanskrit, Greek, and other "classical" IE languages is reflected in Anatolian and so may be assumed for PIE (see also §6.1 on the verb). For all of these reasons, then, Anatolian data willfigureregularly throughout chapter 2. 15
Finally, a number of Baltic and Slavic languages, namely Lithuanian and Old Prussian, and East Slavic, most of South Slavic, and Slovincian, famously preserve a complex system of stress and intonation, replete with intraparadigmatic alternations, derivational patterns, and a whole series of rightward and leftward stress shifts in the individual languages. The amount of data involved in the study of historical Balto-Slavic accentology has usually had the effect of frightening away all non-specialists, including most Indo-Europeanists and phonologists. As chapters 4 through 6 are devoted to the development of PIE accent in these language, I defer description of their accentual systems to §4.2. Other IE branches contribute indirectly to the reconstruction of PIE stress. The Germanic languages, as far as we can tell, had generalized word-initial stress already by Proto-Germanic, but traces of an earlier stage survive in the treatment of PIE voiceless stops. In pre-Proto-Germanic, the voiceless stops *p, *t, *k (< PIE *Jc, *k), *kw developed to voiceless fricatives *f, *þ, *χ, *χw in word-initial position and after a stressed vowel, but otherwise became voiced *β, *ð, *γ, *γ w (cf. Streitberg 1896:116ff.). Thus PIE *t develops to PGmc. *þ in *bhréh2te(r) > PGmc. *brðþer- 'brother', but to voiced *ð in PIE *ph2té(r) > PGmc. *faðer- 'father'. This conditioning has given rise to new grammatical alternations, especially in the preterite of so-called "strong" verbs (those which continue PIE unsuffixed thematic presents in *-e/0-), e.g. post-PIE perf. 3sg. *we-wórt-e, 3pl. *wewi-t-ér 's/he has, they have turned' (cf. Skt. vavárta, vavtfúr) —> *wórte, *wrtr|t > *wórþe, *wurðúnd > PGmc. pret. *warþ, *wurðun > OE pret. wearþ, wurdon, OHG ward, wurtun 'became'.5 Although Iranian records date from the 1st millennium BC, none of the Old or Middle Iranian languages marked the place of stress in the same way as the classical Indian grammarians or the Hellenistic scholars of Alexandria. Our knowledge of the accentual systems of these languages, then, is limited to indirect reflexes in Avestan (the language of the Avesta, the holy text of Zoroastrianism) and to possible accent-conditioned sound changes in Avestan and in Middle Iranian languages, e.g. Middle Persian or Sogdian.6 In 5
Traces of this alternation survive in modern Germanic languages today: cf. English sg. was vs. pi. were < OE wees, wsron < PGmc. *was, *wsrun, lose vs. adj. (< pret. ptcp.) for-lorn < OE (for)leosan,for-loren (Ger. ver-lieren, ver-loren, with generalization of -r- from the pret. pi. and ptcp.). 6 Thus pre-Sogdian appears to have syncopated at least some vowels which were unstressed in Proto-Iranian, prior to the loss of lexical stress and generalization of the "Rhythmic Law" (stress on the first "heavy" syllable, including long vowels and tautosyllabic sequences of vowel + sonorant, otherwise on the final syllable); cf. Sims-Williams 1989:181-2. On Middle Persian evidence for the stress of PIr. presents in *-áya-, see §2.3.2.2c, fn. 63
16
many cases, the indirect testimony of these languages concurs with the attested stress of Sanskrit: cf. Av. vahrka- = Skt. vfka- 'wolf (Mayrhofer 1989:12-3 with refs.). For the most part, however, Iranian offers little for the reconstruction of the accentual system of PEE. The real comparative value of Old Iranian, particularly Avestan, lies in its complex inflectional morphology in the noun and verb, which frequently preserves archaic features already lost in the earliest Vedic Sanskrit.7 Finally, the position of stress in Tocharian B is frequently revealed by the writing of underlying /a/, /a/ as ā resp. a in stressed syllables and as a resp. ā in unstressed syllables (in central and eastern dialect texts). The indirectly attested stressed system of TB has been the subject of a dissertation (Marggraf 1970) and several subsequent studies, e.g. Winter 1990, 1993. Although Tocharian appears to have at least partially preserved PIE lexical accent (Ringe 1987), several factors combine to reduce its probative value: several major accentual innovations, including the widespread generalization of underlying secondsyllable stress in the verb and at least two post-Proto-Tocharian shifts in TB; the loss of most inherited PIE nominal inflectional morphology; the apparent elimination of contrastive stress in TA, the only other Tocharian language; and the restricted nature of the few accentually-conditioned sound changes in the prehistory of Tocharian. In view of these difficulties, I have generally left the Tocharian evidence out of the discussion in chapter 2. The remaining IE languages provide no reliable evidence for the placement of stress in PIE. In Italic and Celtic, initial stress has been generalized, perhaps from underlyingly unaccented forms as has happened in the West Slavic languages (see §4.2.1). Although stress in classical Latin falls on the antepenult or penult, as described above in §1.2.3, the regular syncope and weakening of vowels in medial and final syllables (e.g. past pass. ptcp. mōtus 'moved' < *mowetos, cf. pres. mov-et; pres. 3sg. ac-cipit 'accepts' < *ad-capit, cf. capit 'takes') Meiser 1998:66-74), which is not found in the oldest Old Latin inscriptions, strongly suggest that Old Latin had initial stress up to at least the 5th c. BC; similarly for Oscan and Umbrian, which exhibit widespread syncope of short vowels, A number of modern East Iranian languages also appear to preserve reflexes of Proto-Iranian lexical stress, including Wax!, Yidgha-MunjI, and most notably Pashto (cf. the references in Mayrhofer 1989:13). These languages have almost entirely lost the complex Proto-Iranian inflectional system of nominal cases and classses and verbal moods, voices, etc., so that the vlaue of their evidence is largely confined to demonstrating that PIr. did indeed inherited the PIE contrast of barytone vs. oxytone o- and d-stems (see §2.2.2). 7 The close correspondence in ablaut-grade alternations of root, suffix, and ending between Avestan and Sanskrit makes it likely that Proto-Iranian originally did possess the same stress alternations as Vedic; see §2.2.1, passim.
17
e.g. Osc. iptv. 3sg. actud, Umbr. aitu < *aktōd < *agitōd (Lat. agitōd) 'may s/he do'. The Celtic languages have likewise eliminated lexical stress contrasts, with stress falling uniformly on the initial syllable in Irish and the other Goidelic languages and on the penultimate in British Celtic. Armenian, on the other hand, has stress on the underlyingly word-final syllable (i.e. excluding epenthetic [a]; Schmitt 1977:32). Finally, Albanian also does not appear to preserve any clear reflexes of contrastive stress from PIE. Among the older IE languages, then, only Sanskrit, Greek, Anatolian (especially Hittitte), and Balto-Slavic directly continue the PIE system of mobile stress and underlying accentual contrasts. Given the large-scale innovations and restrictions of lexical stress in Greek (see chapter 3) and the intricate and complicated nature of the BS1. data, it is no surprise or secret that the archaic and relatively transparent Sanskrit system are projected backwards to PIE, with corroborating evidence cited from Verner's Law reflexes in Germanic and from Hittite when available. Let us evaluate the PIE accentual system in the light of recent scholarship and determine to what extent the reconstructed alternations of stress may be accomodated under Idsardi and Halle's metrical framework of brackets and grids.
18
Chapter 2. The Proto-Indo-European Accentual System 2.1. Introduction: coexisting systems in Proto-Indo-European grammar Before describing the accentual system reconstructive for Proto-Indo-European, a few words on the nature of the reconstructed grammar are in order. All human languages are characterized by a certain number of phenomena — phonological, morphological, and syntactic — which do not conform to the patterns, rules, or constraints governing the rest of the grammar. Synchronically, these exceptions are irregularities and must be learned as such, by children acquiring it as a first language or by adult second-language learners. Recent studies of language acquisition and child errors have supported the view that children abstract productive phonological, morphological, and syntactic rules or processes on the basis of the default patterns in their language(s). In many cases, they overgeneralize these rules to irregularities, which must be learned as less productive or lexically restricted rules or processes or as lexical exceptions. Hence the commonly heard stories of two-, three-, and even four-year-olds who, having learned that the past tense in English is most commonly expressed with the suffix -d (in its various allomorphs, [-d], [-t], [-ad]), incorrectly produce "goed", "maked", "singed", and "bringed"; for these children, as well as for second-language learners, of course!), the irregular past forms went, made, sang, brought must be learned as individual lexical exceptions.1 From a diachronic perspective, however, synchronic exceptions are often historical relics, left over from an earlier stage of the language and preserving patterns which were productive in the grammar of that stage. The past tense of English provides an especially clear example: whereas the regular past in -(e)d is common to "weak" verbs in all Germanic languages and has been inherited from Proto-Germanic, the alternations found in "semi-weak" verbs such as bring, brought or seek, sought reflect an archaic and unproductive PGmc. pattern, and the PGmc. vowel alternations which survive in "strong" verbs such as sing, sang, sung (and nominal song) or ride, rode, ridden (and nominal road) reflect ablaut grades in the present, perfect, *-no- participle, and deverbal noun in
'On the acquisition of the English past tense as a model of how children learn rules and exceptions, see Pinker 1999.
19
(post-)PIE.2 For historical linguists, such irregular forms, distributions, or patterns are therefore of great interest, as survivals of an earlier stage of the language concerned. The reconstructed grammar of PIE is no exception in this regard. Like some of its more archaic descendants, e.g. Hittite, Vedic Sanskrit, or Homeric Greek, PIE exhibits a contrast between a more archaic system of nominal and verbal inflection, in which alternations of ablaut and accent are intimately intertwined, and relatively younger categories, which become increasingly productive in post-PIE and the daughter languages, in which those alternations are reduced in complexity and importance. In the noun, these categories are the archaic ablauting stems, in which surface stress is reconstructed as falling on the root or stem in certain case forms (traditionally called "strong") and on the root, stem, or ending in the other so-called "weak" cases. Depending on the position of stress, each of these morphemes may take a different ablaut grade, i.e. *e, *o, or zero (*0): in general, e-grade appears in the stressed syllable, and 0- or o-grade in unstressed syllables. Similarly, the verb exhibits a number of archaic formations in the present, aorist, and perfect, all characterized by an alternation of stress among root, affix, and ending between "strong" and "weak" forms of the paradigm. The reconstruction of this most archaic PIE morphophonological system is based principally on the evidence of Vedic and classical Sanskrit, Homeric Greek, and Hittite, as well as relics in these and other IE languages. Even in these three languages, the ablautaccent alternations have mostly become unproductive: within the individual histories of Indo-Aryan and Greek, for instance, we may observe a progressive elimination of ablauting consonant-stem nouns or athematic verbs in favor of thematic stems, marked by a near-lack of ablaut alternation and columnar stress (cf. §2.2). The latter formations, which likewise have become productive and dominant in most IE languages, are obviously easier for children, as well as second-language learners, to acquire; consequently their expansion, and the replacement of ablauting roots with o-stem equivalents or derivatives, would have been favored over time.
2
Cf. also bear, bore, borne (the latter now orthographically distinct from originally identical adjective born, e.g. in / was born vs. / have borne), with the associated nouns barrow, burrow, burden, etc. Such examples can of course be multiplied for English.
20
2.2. The noun 2.2.1. Archaic ablauting stems Until approximately a generation ago, the reconstruction of the PIE noun was presented as a set of discrete paradigms, each with its own idiosyncrasies. With some exceptions (notably Pedersen 1926, Kuiper 1942), few attempts were made to integrate the various reconstructed stress patterns, ablaut grades, etc. into a coherent system of nominal inflection and derivation. This traditional approach may still be encountered in recent handbooks of IE, e.g. Sihler 1995. The 1960s and '70s witnessed the gradual acceptance of a new view of the inflectional morphology of the PIE noun, one in which the ablaut and accent alternations of non-thematic stems, i.e. nouns not formed with the thematic vowel *-o- or *-e-h2 (cf. §2.2.2), were combined into a unitary picture. Among the leading names associated with this "kinetic revolution" are the scholars of the "Erlangen school", in particular Johanna Narten, Heiner Eichner, and Jochem Schindler, who in a series of now classic articles (Narten 1968, Eichner 1972, Schindler 1975a, 1975b) worked out most of the details of this sytem. The Neogrammarians were already aware that alternations of ablaut and accent appeared to be intimately intertwined in a number of old-looking consonant-stem nouns. Cf. the following paradigms of 'father' in Sanskrit, Avestan, and ancient Greek (the Avestan column has been supplemented by forms from nar- 'man' and mātar- 'mother'): Sanskrit pitā pitar pitár-am pitúr pitr-é pitr-ā pitár-i
Avestan ptā (G), pita nara patarām (G), pitaram OP piga (<*pi0rah) faSroi (G), piβre
Greek patéir páter patér-a patr-ós
nairi
patr-í (d;
nom./voc./acc. du. instr ./dat./abl. gen. loc.
pitár-ā(u) pitŗ-bhyām pitr-ós
nara narabiia nará
patér-e [patér-oi
nom./voc. Piace.
pitár-as pitfn
patér-es patér-as
gen.
pitŗņām
patarō, voc. mātarō (G) mātarqš-ía (G) ([-tŗs] < *-tŗns) narqm
nom. sg. voc. ace. gen. dat./abl. instr. loc.
21
»?
patr-5:n
dat./abl. instr. loc.
pitŗ-bhyas pitŗ-bhis pitŗ-su
pterabiiō (dat.) patrá-si (dat.)
On the basis of these three inflectionally archaic languages, one may reconstruct a contrast between stressed, ¿-grade stem *-ér- on the one hand, and 0-grade stem *-r- with stressed ending on the other; in the Romantic-influenced terminology of the Neogrammarians and latter-day scholars, these variants may be labeled "strong" and "weak" (German stark and schwach). The former is found in the voc, ace, and and loc. sg., nom.-acc. du., nom. pi. and, in Greek, ace. pi.; the latter variant occurs in the oblique cases (gen., dat., instr. of all three numbers; loc. du. and pi.) and, in Indo-Iranian, the ace. pi. Both Greek and IndoIranian agree in exhibiting nom. sg. reflexes of an accented suffix *-ē(r) with lengthened grade, which Szemerenyi (1962:12-3, 21fn.l 1, 1970:109), following a line of scholars going back to the 19th century, has plausibly explained as the reflex of pre-PIE **-ér-s; this phonological rule has now been generally accepted and is commonly referred to as "Szemerenyi's Law".3 Assuming that the ace. pi. in PIE was a "strong" case form, like Greek but unlike Indo-Iranian, we may therefore reconstruct the following paradigm of 'father' for PIE as follows: nom. voc. ace. gen. dat./abl. instr. loc.
sg. *ph2ter *ph2tér *ph2tér-iņ *ph.2tr-és *pb.2tr-éy *pb.2tr-éhi *ph2tér-(i)
du. *ph2tér-hļe4
pl. *ph2tér-es «
*ph2tér-ņs *ph2tr-on *ph2tŗ-bhós *ph2tŗ-bhís *ph2tŗ-sú
The ablaut alternation between nom. sg. *-ér (< **-ér-s) strong *-ér- and weak *-r-' has left numerous traces elsewhere in IE. Celtic has preserved the original distribution to a 3
Jasanoff (1988b:71fn.3) has adduced support for this sound change in the reconstructed 3pl. pert", endings *-€r (Hitt. -ēr, Lat. -ēre < *-ēr + i) and *-i;s, attested in Skt. 3pl. -ur (cf. r-stem gen./abl. sg. -ur< *-ŗs\ originally proper to aerostatic inflection; §2.2.1.2) and the isolated GAv. ðikōitaraš 'they have understood' (or 'had understood'; Jasanoff 1997a) alongside usual -arS: if *-e"r continues earlier **-ers, the variant endings are instantly recognizable as pre-PIE ablaut variants **-ers - *-ŗs. Szemerenyi's Law may also have applied to word-final **-VRd: cf. 'heart' below in §2.2.1.1. On the consequences of Szemerenyi's Law for the root aorist of a sonorant-final root, cf. R. Kim 2001b. The third environment for Szemerenyi's Law, **-VRh2 > *-V:R, is discussed below in §2.2.1.3. 4 Nom./acc. du. *-hie (> Gr. -e in pód-e) has been apocopated in o-stem *-o-hie > *-oh[ (> Ved. -cf, Gr. -o.\ Lat. du-ō, Lith. -«, -MO-, OCS -a, TA wu < *dwō); cf. Jasanoff 1988:73-4fn.l0, Melchert 1994a:51-2 for the sound change.
22
considerable extent: cf. Gaulish matir, duxtir < *-tēr, Old Breton brotr, Middle Breton breuzr, pi. bre(u)d-e(u)r, Middle Welsh brawt, pi. brod-er, -yr < *brātīr, *brāteres, Olr. athair < *atlr, ace. athair < *ateren, gen. athar < *atros (McCone 1994). In Latin, the weak variant has been generalized to all cases other than the nom. sg.: hence nom. sg. pater < *-ēr vs. ace. pat-r-em < *-r-iņ (for *-ŗ-m), gen. pat-r-is < *-r-es, nom. pi. pat-r-ēs <— *-r-es (with /-stem -ēs, as typically for Latin), ace. pi. pat-r-ēs < *-r-ņs, gen. pat-r-um < *-r-ōm. Similarly, Tocharian has replaced the inherited alternation with a simple contrast of nom. sg. vs. other, e.g. TB pācer, TA pācar < PT *pacer < *-ēr, obi. TB pātār < *-ŗ-m, gen. TB pātri < dat. *-r-ey. Although PIE 'father' has not been preserved in Balto-Slavic, other r-stem relationship nouns (cf. below, no. 5) have generalized full-grade *-er- from the strong forms to the oblique cases, resulting in an opposition of nom. sg. *-ēr vs. *-erelsewhere: cf. Lith. (archaic) mót-é, OCS mat-i 'mother' < *-ē, 5 acc. mót-er-ja, mat-er-ía < *-er-iņ, gen. mót-er-(e)s, mat-er-e < *-er-es, nom. pi. mót-er-(e)s, mat-er-e < *-er-es, acc. mót-er-is, mat-er-i < *-er-ņs, gen. mót-er-u, mat-er-ú < *-er-ōn. Relics of these alternations have also survived in Germanic (cf. Streitberg 1896:249-52) and Armenian (e.g. nom./acc. hayr 'father' < *hayir < *patēr vs. gen./dat/Ioc. hawr < *patr-ós, *-éy, *-í; Schmitt 1977:100). The alternation of ablaut and stress in 'father' represents only one type of archaic ablauting stems reconstructible for PIE. Let us consider these types in turn and determine to what extent each of them, and the system of ablaut- and stress-alternating paradigms established by the Erlangen scholars, may be modeled according to the metrical bracket theory of Idsardi and Halle.
In the following, I treat the three component morphemes,
root (R), suffix (S), and ending (E), as three underlying syllable heads. At an early stage of PIE grammar, unstressed syllables were realized with zero-grade ablaut, i.e. with syncope of the underlying vowel *e. This phoneitc surface realization rule had already ceased to be productive by the reconstructible stage of PIE, having long since become morphologized as a component of particular (i.e., athematic) nominal and verbal classes. Over time, stress-conditioned ablaut alternations were increasingly limited to these archaic formations and gradually eliminated by "analogy", i.e. the generalization of one or another ablaut grade; this is the case in Hittite and Sanskrit. Eventually ablaut alternations became dissociated altogether from stress, either via the near-complete generalization of nonalternating surface forms, e.g. in Balto-Slavic, or as a result of extensive or complete 5
For PIE *-ēr by analogy to «-less amphikinetic n-stem nom. sg. *-ð < *-ōn (vs. hysterokinetic *-e:n; cf. Jasanoff 1989:138, fn.9). On PS1. *-i < PBS1. *-ē:, see §4.2.2, fn. 15.
23
elimination of lexical stress, in ancient Greek, Germanic, Italic, Celtic, Tocharian, Iranian, and Armenian. In these latter IE branches, alternations of ablaut have become a fossilized peculiarity of the grammar and, like the strong verbs of modern English, must be learned as such. 2.2.1.1. Root nouns with alternating stress This class of nominal stems consists entirely of a monosyllabic root, without suffix. In the strong cases, i.e. the nom. and ace, this root is stressed, whereas in the weak, i.e. oblique cases it is the case endings which receive the stress. Both Sanskrit and Greek provide numerous examples:6 nom./voc. sg. ace. genVabl. dat. instr. loc. nom./voc. pi. ace. gen. dat./abl. instr. loc.
Sanskrit pāt 'foot' pād-am pad-ás pad-é
Greek pó:s 'foot' pód-a pod-ós
aíks 'goat' atg-a aig-ós
dat. pod-í
aig-í aíg-es aig-as aig-3:n
pad-í
vāk 'voice' vāc-am vac-ás vac-é vac-āa vac-í
pād-as pad-ás pad-ām pad-bhyás pad-bhís
vāc-as vac-ás vac-ām vag-bhyás vag-bhís
pód-es pód-as pod-5:n
pat-sú
vak-sú
d&i.po-sí
pad-ā
aik-sí
In Greek, this stress alternation is not only universal in monosyllabic stems, but actually productive: nominal stems which become monosyllabic as a result of contractions, even after the period of the Homeric epics, generally acquire ending-stress in the gen. and dat., e.g. PGr. neut. *ows, *owh-at- > Horn, ous 'ear', gen. oúatos, nom./acc. pl. oúata, but Att. contracted 5:t-a, whence gen. sg. o:t-ós (see further §3.4.2). To be sure, not all such root nouns may be traced back to PIE: a number lack convincing etymologies, while others have been remodeled from root-accented aerostatic root nouns (see 2 below), including *pód- ~ *péd > Gr. ace. pód-a, gen. pod-ós, Ved. pá:d-am, gen. pad-ás and *woyk'- ~ *wéyk > Ved. vit, ace. vis-am, gen. viš-ás. 6
Contra Szemerenyi (1962:13, 1970:109), the lengthened-grade nom. sg. of obstruent-final stems, e.g. Skt. pāt, Gr. TΓCÓS (Hesych.), Lat. pes ors-stem Ved. su-mánās, Gr. eu-menε:s 'well-minded, kind', cannot be by sound change, but must instead be analogical to that of sonorant-stems, which has arisen by Szemerenyi's Law (cf. above, §2.2.1).
24
Nevertheless, in some cases it is possible to establish solid word (or paradigm) equations involving Anatolian cognates, which increases the likelihood that such root nouns belonged to the alternating type in PEE. •
•
•
•
PIE *dyéw-s, ace. **dyewm > *dyēm (by Stang's Law, whence analogical nom. *dyēws; Szemerenyi 1956:184ff.), gen. *diw-és 'sky, god': Skt. dyāus, dyām, div-ás, -é, -ā, loc. dyávi, Gr. Zeús, Zε:n-a (with regular consonant-stem ace. ending -a), Di(w)-ós, -í, cf. Hitt. šivna- 'god', Siunali-, Lyd. ciwvali- 'divine' < *dyéw-, Luv. tiwat- 'Sun-god' < *diw-otPIE *h2ént- ~ *h2ņt- 'face': Hitt. hanz(a) 'front', adv. hanz(a) 'in front' < loc. *h2énti,7 dat.-loc. hantl 'apart' < 'in front, opposing' < dat. *h2ņt-éy; thematized *h2ént-os > PT *ante >~TB ante, TA ānt 'visage, brow'; cf. loc. *h2ént-i > Skt. anti 'opposite, near', Gr. anti 'opposite, against, (in return) for', Lat. ante 'before'; dat. pi. (du.?) *h.2ņt-bhí 'to/on both sides' > Gr. amphi 'on both sides, around, about', Lat. amb(i)-, Olr. imb-, im(m)-, PGmc. *umbi (OE ymbe, OHG umbi) 'around', whence masc. nom./acc. du. *h2ņt-bhóhi 'both' > Gr. ámpho:, Lat. ambō, —> PT masc. nom./obl. du. *antp-3y > TB āntpi, TA āmpi; see in general Jasanoff 1976 PIE *h2éhi/3S- - *h2hi/3S-' 'ash': Hitt. hāš 'ashes, soap', ace. hāššan; derived *h2éhi/3S-eh2- ~ *h2hj/3S-' in hāšš-aš 'hearth', ace. hāšš-an, gen. hašš-āš, dat.-loc. haSS-ī, abl. hašš-āz, whence thematized *h2ehi/3S-eh2- > Lat. āra (Melchert 1994a:78, 1994c:235-6; extended from root *h2ehi/3- > Pal. hā- 'be warm'?) PIE nomVacc. *lcēr 'heart' (< **£erd by Szemerenyi's Law: Hitt. SA-/r/kēr/, Horn. Gr. k£:r, Skr. hārd-i, hard-, Arm. sirt with *-d from oblique cases) ~ obi. *krd'heart': Hitt. obi. /kard-/ (gen. kard-iyaS, dat. karti, all. karta, abl. kartaz, instr. kardit);8 on CLuv. zār-za, HLuv. za+ra/i-za /zārza/ = /zārd/ + neut. sg. particle /-sa/ see ibid., 233, 272: /zārd/ <— *zōr < *lεēr+d or *¿ŗd-), Skt. hŗd-, Av. zarad-, Lat. cor, gen. cordis, cf. Lith. Sirdls (/-stem), PS1. *sírdíce, Olr. cride (yo-stems)
To this list one may also add the word for 'mouth': although the details are complicated and still unclear, Hitt. aiš 'mouth', gen. iššāš, dat.-loc. iššī, abl. iššāz, pl. dat.-loc. iššāš, along with CLuv. āšš-, Pal. āš(kummāwa) 'sacralized meats' (i.e. 'mouth-sacralized') and Latin ōs, probably continue a root noun *h]eh3S-s (> Lat. ōs, Skt. ās-, Av. āh-), *hļli3S-és (contrast Melchert [1994a: 115-6], who posits an s-stem *h]éh3-s- ~ *h]h3-és-). Finally, the old-looking Hitt. noun for 'house', E-ir /per/, gen. parnaša, dat. parni, all. parna, instr. parnaz (secondarily root-stressed, cf. Melchert 1994a: 125-6; thematized in Lyd. bira-, Luv. neut. nom./acc. sg. parnan-za), which finds no cognates elsewhere in IE,
7
On the stress of the loc. sg., cf. §2.2.1.3. Probably with generalized root-stress; similarly for the oblique forms of 'house' below (Melchert 1994a: 115-6). Mobile stress has apparently also been lost in Palaic: cf. dat. kārti, ēšhana'blood', gen. [h]āranaš 'eagle' (ibid., 226). 8
25
probably continues a PIE nom./acc. *pēr (< **pern by Stang's Law? D. Testen, p.c, fall 1996) ~ obi. *prn-'.9 The alternation of root- and ending-stress in root nouns follows straightforwardly according to the brackets-and-edges model if one assumes that the root and strong case endings are underlyingly unaccented, whereas the weak case endings are accented, i.e. are prespecified as projecting a left bracket onto line 0 of the metrical grid. Thus the strong case forms are unaccented and receive default initial stress according to the parameters posited in § 1.2.4 for PIE, and the weak case forms stress their one underlyingly accented morpheme, namely the ending. The derivation of stress is illustrated below for ace. *h2ént-iņ and gen. *h2#t-és: *
*
(*
(*
* *\ *h2ent m > *h2ént-iņ
* r*\ *ti2ent es > *h2ņt-és10
We may therefore formulate the preliminary hypothesis that the PIE nom. and ace. case endings were unaccented, while the oblique endings were accented; since the underlying accentual specification of morphemes should not vary according to the forms in which they occur or depend on the accentuation of neighboring morphemes, this distinction should hold not only for root nouns, but for all consonant stems which take these endings. 2.2.1.2. Aerostatic root and suffixed nouns Although the alternation of root- and stem-stress between the direct and oblique cases is well attested in Sanskrit and Greek, there are indications that PIE originally possessed another type of root noun. It has long been known that a number of root nouns exhibit both o- and e-grade ablaut in the daughter languages. To take the most famous example of this type, the root noun for 'foot' appears with o-grade in Greek (ace. pód-a, gen. pod-ós; the long vowel in nom. pous is of unclear origin, but must be analogical), Armenian (otn < *pod-iņ), and Germanic (OE fōt, pl.fēt < *fōt-s (?), *fōt-iz < PGmc. root noun *fōt-, with analogical long vowel extended from the nom. sg. throughout the paradigm), but with e-grade vocalism in Italic (Lat. ace. ped-em, gen. ped-is; nom. pes once again with analogical long vowel). 9
See Schindler 1972a for a survey of root nouns in Indo-Iranian and Greek. On the surface realization of the unstressed root with zero-grade ablaut, see the discussion in §2.2.1.3. ,0
26
In other cases, the ablaut alternation has survived only in a form which has become dissociated from the rest of the productive paradigm, perhaps already in (late) PIE, and survives only in fixed expressions. For instance, 'night' is traditionally reconstructed as *nokwt-, given its reflexes in Gr. núx, gen. nukt-ós, Lat. nox, gen. noct-is, Skt. náktam 'by night' (fossilized ace. in adverbial usage), TB nekclye, TA n[o]ktim 'in the evening', TA nokte 'last night' (R. Kim 2000b: 155), but the discovery of Hitt. gen. nekuz /negwts/ in the expression nekuz mehur 'evening time' indicates that this noun was originally aerostatic *nók w t- ~ *nékwt- (Schindler 1967). Likewise, the root noun *dom- 'house, home' exhibits o-grade throughout IE, e.g. Lat. domus, OCS domu («-stems; cf. Sk. dámúnas 'dear to the home'?), Skt. dámas, Gr. dómos (also Lith. nāmas with nasal assimilation?), Arm. tun, but the oblique e-grade of the original aerostatic paradigm *dóm- ~ *démsurvives in the expression *dém-s pótis 'lord of the house' > Ved. dám-pati-, GAv. gen. dāņg patōiš, —> Gr. des-pótε:s 'lord, master'. Whereas most aerostatic root nouns alternate between o-grade in the nom. and ace. and e-grade in the oblique, a small number contrast full-grade and lengthened-grade reflexes, e.g. *méms- ~ *méms- 'meat': cf. Skt. mās, thematized māmsám < *mēms- vs. TB pi. tant. rriīsa < PT *mysnsa < *mems- (for loss of the nasal in TB, cf. piš 'five' < PT *pyanša < PIE *pénkwe, Ringe 1996:70-lfn.3).11 These were clearly less numerous than the *o ~ *e nouns, but the existence of parallel cases among r/n-stems (see immediately below) strongly suggests that one must reconstruct two subtypes of aerostatic inflection in the noun, i.e. the "o-subtype" and the "ē-subtype". Alongside root nouns, one also finds a number of r/n-stems with divergent ablaut reflexes in the daughter languages. In these archaic "heteroclitic" nouns, the nominal stem consists of a root and a suffix, which contains *-r- in the nom. and ace. but *-n- in the oblique cases. The type is robustly attested in Hittite, which attests underived r/n-stems as well as productive suffixes in *-r- ~ *-n- (e.g. inf. -war, gen. -waS< *-wŗ, *-wen-s), but have left only scattered relics in other IE branches (e.g. Indo-Iranian, Greek, Germanic). A number of r/n-stems exhibit o-grade reflexes, e.g. PIE *wod-r 'water', directly continued by Hitt. wātar and indirectly by OE water, OHG wazzar (cf. Goth, wat-ō, gen. wat-in-s, ON vatn), OCS voda. In addition, one also finds r/n-stems with *ē and *e-grade root vocalism, parallel to root nouns: cf. Gr. hε:par, gen. hεipatos, Av. yākara vs. Skt. yák-ŗt (with secondary -t), gen. yak-n-ás 'liver'. 1
'Goth, mimz could continue either *mems- or *mēms-, in the latter case by shortening of a long vowel before a tautosyllabic sonorant ("Osthoffs Law").
27
In a landmark article, Schindler (1975a:4-6; cf. also 1972b:32-6) proposed that these two subclasses of root nouns and r/h-stems were originally aerostatic in PIE, i.e. the stress fell on the root in all cases. The first type contrasted *o in the nom. and ace. with *e in the oblique cases; similarly, the second type had *f in the nom. and ace. and *e elsewhere. The nom., ace. and gen. sg. of 'foot', 'meat', 'water', and 'liver' were thus nom. sg. ace. gen.
*pód-s *pód-iņ *péd-s
*mēms-s *mēms-m *méms-s
nom./acc. sg. gen.
*wód-ŗ
*Hyēkw-ŗ
*Hyékw-ņ-s
*wéd-ņ-s
It should be noted that this postulated distribution of ablaut grades is directly attested only in a single paradigm, Hitt. nom./acc. wātar 'water', gen. witen-aš, and even the latter probably continues *wedén-os <— *wéd-jj-s, with full-grade stressed suffix from proterokinetic paradigms (cf. below) and regular gen. sg. ending -aš < PA *-os.12 In all other IE languages, and in all other such root nouns, one or the other ablaut grade has been generalized throughout the paradigm.13 Additionally, the root-stressed oblique forms of r/n-stems tend to be replaced by forms with stress on the suffix or ending, on the model of proterokinetic or hysterokinetic nouns, respectively (Schindler 1975a:6-7; cf. below, 3 and 4): hence *wed-én-s or *wed-n-és (or *wéd-n-os, *wed-én-os, etc. In addition to Hitt. witen-aš, which must have been remodeled twice (in the suffix *-en- [with or without accent shift] and in *-os, the productive Hitt. gen. sg. ending), cf. dat.-loc. sg. lamnl 'hour' < *lomn-éy <— *lémn-i to *lóm-ŗ ~ *lém-n-. Clearly, then, aerostatic nouns were a recessive class already in PDE, and have largely disappeared even in the most archaic IE languages. Finally, PIE appears to have had at least a handful of nouns of other suffixed classes which likewise were marked by constant root stress.
12
In the absence of spellings such as 'witēnaš", however, it is also possible that the / of witenaS results from raising of PA *e between *w and a coronal, cf. wītt- 'year' < PA *wét- (Melchert 1994a: 101). 13 Indo-Iranian alternations such as Skt. pad- -pad- 'foot', vāc- - vac- 'voice' (cf. above. §2.2.1.1) probably do not directly continue the aerostatic alternation of *o - *e. If Brugmann's Law operated as a sound change only before sonorants (the "Kleinhans restriction": cf. Mayrhofer 1986:146-8. Ringe 1996:4-5 and in detail Hajnal 1994a), it is possible that nom./acc. -a:- has been generalized from roots of the shape *CoR- - *CeR-, but I am not aware of any relevant examples.
28
•
• • •
• •
•
•
•
*h20w-i- ~ *h2éw-i- 'sheep': CLuv. hāwīš, HLuv. ha-wa/i-i-sa /hāwīš/, Lyc. acc. sg. xawā 'sheep' (-5 from wawā 'cow'?) <— *Hów-ī- (with "motion"-suffix *-I- < *ih 2 .) < _ PA /-stem *Hówi- (cf. Hitt. UDU-/S) < PIE *h2ów-i- (the Lycian form points decisively to PA *H- < PIE *h2-; cf. Kimball 1987:189, Melchert 1994a:72, R. Kim 2000a:38fn.2 with refs.); likewise Skt. ávis, Gr. Horn, óis, Att. ois (Argolic ówis), Lat. ovis, Olr. oi, Lith. avis, OCS ovī-ca 'ewe' < PIE *h20w-i-, Arm. hoviw 'shepherd' < *h20wi-peh2- (i.e. *h20wi- 'sheep' + *peh2- 'protect'), Goth, awi-str* 'sheepfold', awēþi 'flock/herd of sheep' (ON cer, OHG ou, ouwi, Dutch ooi, OE ēowu, ēwe, Mod. Engl, ewe) vs. TB āuw 'ewe' (pi. awi) < PIE *h2éw-i- (Schindler 1994:397, cf. already 1969:153fn.60; for discussion cf. R. Kim 2000a) *h2ó£-r-i- ~ *h2ék"-r-i- 'point': Gr. ókris, ákris, Ved. ášrih from either; *wðst-u- ~ *wást-u- 'settlement': Skt. vāstu 'residence, home', TB ost, TA wast 'house' < PT *wosta < *wāstu vs. Myc. wa-tu-, Horn, ástu 'fortification, city' < *wást-u*h]ós-u- ~ *hjés-u- 'good': cf. Hitt. āššu- 'good', the result of paradigmatic crossing of *āsu- ~ *ássu- < *h2ósu- ~ *h2ésu- (Melchert 1994a:63, 1994b:300-3); Horn, eus, neut. ev (Att. adv. eu < *éu with archaic recessive stress, cf. Hoenigswald 1998) < *ehu- < *h]esu- (Skt. su-, Av. hu- 'good-, well-', PS1. su- in sú-mīrtī 'death', sudorvv 'healthy' < zero-grade *hisu- in compounds) *wēs-u-/*wós-u- ~ *wés-u- 'good': Olr./w < *wēsu, Luv. wāšu- < *wósu (*wésu would give "waššu" by Cop's Law), Skt. vásu-, GAv. vohv, vaņhu-14 < *wós-u- or *wés-u*bhréh2tēr, *-tŗs 'brother': Skt. bhrātā, Gr. phrá:tε:r, PGmc. *brōþēr (Goth, brōþar [< voc. *-er, cf. Stiles 1988:115-8], OE brōðor, OHG bruoder, Germ. Bruder), all attesting or continuing root stress; gen. Skt. bhrdtur, ON brōðor < PGmc. *-ur < *-rs, extended to other r-stems similarly *méh2tēr, *-tŗs 'mother'? The full-grade of the root (vs. *ph2tēr 'father') indicates an aerostatic or proterokinetic paradigm, as does the initial stress of Greek and Balto-Slavic (Gr. mε:tε:r, OLith. mate, gen. móteres [AP 1; Senn 1966:141], Cak. máti, mátere [references]). The stem stress of Skt. mātāa, PGmc. *mōðēr (> OE mōdor, OHG muoter) may be original, but it could of course also be analogical to that of 'father': cf. TB mācer, TA mācar < PT *macer with the vocalism of *pacer < *ph2tēr, for expected TB "mocer", TA "macer" < *mocer < *meri2tēr. *h[nēh3m-ņ ~ *hļnéh3m-n- 'name': TB hem, TA horn < PT *nema < *h]nēli3m-ņ; remade elsewhere to proterokinetic *hinéh3m-ņ ~ *hjiļh3m-én- from phonetic resemblance to nomina actionis in *'-mņ - *-mén-: weak stem generalized in Olr. ainm, OPr. emmens, PS1. *jime. (OCS imq; cf. Cz. jméno), HLuv. á-ta4/s-man-za /ad(a)man-za/ < *hjņh3m-ņ, Gr. ónoma (see §3.6); Skt. nāman-, Lat. nōmen, Hitt. lāman < *hinéh3m- with either weak aerostatic or strong proterokinetic stem (on the Anatolian reflexes cf. Melchert 1994:67, 83) some ¿-stems: *sf d-(o)s, *séd-s > *sēd-os, *séd-es- (?), cf. Olr. síd 'peace, mound' vs. Welsh hed; *nēb h -(o)s, *nébh-s —> *nēb h -os, *nébh-es- (?) > HLuv. /tīpas/ vs. CLuv. tappaš- (assimilated to normal proterokinetic 5-stem type in Hitt. nēpiš, etc.; cf. 4 below) 14
On the orthographic alternations of the Av. form and their consequences for the prehistory of PIr. w-stem inflection, cf. Testen 1994.
29
Once again, we see that the root takes different ablaut grades in the strong and weak cases: *o vs. *e in 'sheep' and 'good', *ā vs. *a in 'fortification'. In the case of 'brother', no such ablaut alternation is recoverable, as all the reflexes of this word in the IE languages point to an e-grade: its assignment to aerostatic inflection is based on the unanimous evidence of Sanskrit, Greek, and Germanic for root stress and on the argument that the Sanskrit gen. sg. in -ur < *-ŗs, which has become universal among r-stems in that language, must have originated in nouns which stressed the root, and hence had zero-grade of the suffix *-r- and ending *-s, in a weak case.15 Note finally that the otherwise unmotivated o-grade vocalism of such securely reconstructed PIE /-stems as *ghóstis 'host, guest', *pótis 'master' may well betray an original aerostatic paradigm in (pre-)PIE. Unlike 'ewe', for which an e-grade reflex has recently been discovered in Tocharian (see above), all known reflexes of these two nouns happen to have generalized the *o of the nom. and ace: Lat. hostis 'host, guest', PGmc. *gastiz 'guest' (ON gestr, OE ļiest), PS1. *gostI (i-stem, cf. §5.1.1); Hitt. emphatic particle -pat, Gr. pótis (cf. the reflexes of *dém-s pótis above), *potis 'able' in possum, potes, potest, etc. 'can' (< *potis sum, es, est), Lith. viēš-pats 'master', OPr. ace. waispattin 'mistress'. Similarly, the u-stems *dóru 'wood' (Ved. dāru, gen. drós, Av. dāuru, gen. draoš, Gr. Horn, dóru '(wooden) spear', do.rós < *dorw-ós, Att. dóratos), *gónu 'knee' (Ved.janu, gen. jnós, Gr. Horn, gónu, go:nós < *gonw-ós, Att. gónatos), *h20yu 'life' (Ved. āyu, gen. yós, GAv. āyú, gen. yaoš; Gr. ouk(i), Arm. oð' < *(ne) *h2<3yu kwid 'not ever', cf. Cowgill 1960), usually reconstructed as proterokinetic on the basis of the Indo-Iranian paradigms, could have been aerostatic *dór-u- ~ *dér-u-, *gón-u- ~ *gén-u-, *h20y-u- ~ *h2éy-u-, with the e-grade clearly underlying Lat. genu (genu) and Hitt. genu, gen. gēnuwaš (but abl. ganut /gnud/ < *gnu- with zero-grade). As Schindler (1975a:7) suggests, these could have been remade to proterokinetic *dór-u ~ *dr-éw-s, *gón-u ~ *gn-éw-s, *h2<5y-u ~ *h2y-éw-s at an early date; in addition to the Indo-Iranian forms cited above, note YAv. ace. sg. ¿num (as if < *gn-u-), probably backformed to gen. Znaos*. The new weak stem, with full grade *-ew- of the suffix, has been generalized in e.g.
15
SchindIer (1975a:4-5, 1975b:262) and Hoffmann (1976:597-8; cf. Mayrhofer 1989:15) have plausibly derived the Indo-Iranian i- and «-stem instr. sg. in *-ī, *-ú (Ved. utl to utis 'aid'. GAv. aSJ. xratu, vo/iu to a Si S 'praise', xratuS '(mental) strength', vaņhu- 'good') from PIE * '-i-hi. *'-u-h, in aerostatic paradigms, with zero-grade of both stem and ending.
30
PGmc. neut. *knew-%, *trew-a. (Goth, kniu*, triu* [pi. nom./acc. kniwa, dat. pi. triwam 'with wooden clubs'], OE cnēo, trēo(w) 'tree').16 The accentual difference between aerostatic and alternating root nouns may easily be represented in the brackets-and-edges hypothesis. If the root is underlyingly accented, then all forms of the paradigm will surface with stress on the root syllable, thanks to the leftmost head marking on line 1. The metrical grids for the ace. and gen. of 'foot', 'water', and 'liver' would thus have been * /*
* /*
*pód-iņ * (* í*
* r*
*Hyēkw-ŗ
*wód-ŗ * (*
í*\
*péd es > *péd-s
í*
* r* í*
*\
*wéd en > *wéd-ņ-s
s
(*
Í *
*\
w
*yék en s > *Hyékw-ņ-s
As noted above, aerostatic root and suffixed nouns were in retreat from PIE onwards. In terms of underlying root accentuation, then, it appears that accented roots were disfavored and came to be replaced by unaccented roots in stress-alternating paradigms: in other words, columnar stress on the root was gradually abandoned in favor of accentual mobility, much as has taken place within the brief recorded history of the Aukstaitian dialects of Lithuanian which underlie the modern literary language (references; cf. §4.4.1). 2.2.1.3. Amphikinetic stems Let us now turn to suffixed nouns in which the stress of the strong and weak case forms is not identical. These may be divided into three classes, depending on the alternation of stress: amphikinetic, with root-stress in strong forms and ending-stress in weak forms (hence "amphi-kinetic", moving between the first and last morphemes of the word); proterokinetic, with strong root-stress and weak suffix-stress (hence "proterokinetic", moving between the first two morphemes); and hysterokinetic, with strong suffix-stress and weak ending-stress (hence "hystero-kinetic", moving between the last two morphemes). We shall consider each of these in turn. 16
The locatives to these aerostatic «-stems would probably have been *dér-u-, *gén-u-, but as far as I am aware these have not survived in any IE language.
31
The class of amphikinetic stems has generally been considered the most archaic of the three mobile types of suffixed nouns. Only a few members of this class may be reconstructed with complete confidence for PIE, or have left unambiguous reflexes in one or more daughter languages.17 •
• • •
• • •
*dhégh-ōm 'earth', ace. *dnéán-ōm < **dhédh-om-m (by Stang's Law, cf. Stang 1965, Schindler 1977:31), gen. *dhán-m-és, loc. *dhán-ém: Hitt. tēkan (neuter due to merger of nom. and ace; Kimball, cited in Ringe 1996:42, contra Schindler 1967:195-6), dat. takriī (with -n- for *-m- from nom./acc. and loc), fossilized loc. tagān (tāgan) 'on/to the ground' < *dhdh-óm <— *dnáh-ém (Melchert 1994a: 108, 135).18 ^ *éh]t-mōn 'breath, soul', ace. *éhit-mon-iņ, gen. *hit-iņn-és > Skt. ātmā, ātmān-am, ātman-as 'soul, self alongside tmā, tmánas 'self (Ved. instr. tmánā; cf. PGmc. *«ēþm- ~ *5ēðm- > OE ¿dm, OHG ātum 'breath') *h2élc-mōn 'anvil, heaven', ace. *h2élc-mon-iņ, gen. *h2lc-mn-és > Ved. ášm-ā, ace. ášmān-am, gen. ášman-as, Gr. ákmo:n, ákmon-os (with generalized strong stem), Lith. ašmuō, gen. aSmefis 'cutting edge' *h2éws-ōs 'dawn', ace. *h2éws-os-iņ, gen. *h2us-s-és > Ved. usās, acc. usāsam, gen. usás (alongside usásam, usásas), GAv. ušá, YAv. acc. ušaijhsm, PGr. *awhōs > Doric awos (presumably for *a:wo:s), Horn. é:o:s, Att. heo:s (with unexplained stress; initial h- by "Hauchumsprung", from word-internal *-h-), *ausōs —> Lat. aurōr-a *pént-oh2-s 'path',19 gen. *pjg.t-h.2-és > GAv. pantá, gen. paβō (borrowed into PGmc. as *pa0-az, cf. Mayrhofer 1970:224-6), Ved. pánthās (with -th- from weak cases), gen. path-ás; cf. Mayrhofer 1981:432-3 *péyh2-wōn 'fat', gen. *pih2-un-és > Skt. pívā, pívan-as, Gr. pí:(w)o:n, pí:on-os *swé-sōr 'sister', acc. *swé-sor-iņ, gen. *su-sr-és > Skt. svásā, Av. xvahā, Lith. sesuō, gen. sesefs (seserés?), Lat. soror, gen. sorōr-is (with generalized nom. stem), Gr. (Hesykh.) ¿ores 'female relatives' < nom. pi. *swé-sor-es, TB ser, TA sar < PT *swv3sēr3 < acc. sg. *swesorrņ. Cf. PGmc. *swestar, OCS sestra: the -t- of the
1
' N o t surprisingly, differences of root ablaut have generally been leveled in the attested languages, usually in favor of the full-grade variant of the nom./acc. Forms such as Hitt. tēkan, loc. tagān or Skt. āttnā, una are precious relics of the original system of stress-conditioned root ablaut alternation. 18 The acc. sg. has been remodeled with the root ablaut of weak cases, as *dhg'h-ōm > PInlr. h *j sām > Ved. ksām, GAv. zg (whence nom. sg. Ved. ksās, GAv. z&\ Rasmussen 1978:105, Hoffmann and Forssman 1996:141; contra Schindler 1967:204) or as *dhgh-óm-m (with regular ending *-iņ) > Gr. /tV'dna (similarly nom. *dhgh-ōm > kWo'.-n), PT *tkena > TB kem, TA tkam; cf. R. Kim 2000b: 125-6 for discussion and refs. Schindler (1977:31-2; cf. Mayrhofer 1986:152-3) has proposed a sound change by which *dh was lost in the cluster *dhghm- in zero-grade forms such as gen. *dhghm- > *g h m-, whence Ved. jmás, gmás, Gr. adv. kf'amai 'to the ground', Lat. humus 'earth, ground', PBS1. *Zem-(i)jā (OCS zeml'a, Lith. iēmé). 19 Often reconstructed as *pónt-eh2-s, based on the root ablaut of Lat. pons 'bridge', OCS pqh 'path, way'; neither the origin of this o-grade nor the original function of the suffix *-(o)h2- is clear to me.
32
Gmc. and SI. forms can only result from epenthesis in the cluster *-sr-, i.e. zero-grade of the suffix, which implies either an aerostatic or amphikinetic paradigm.20 To this list one may add the small class of stems in *-oy- ~ *-i-, continued in Indo-Iranian as the type of Skt. sákh-ā 'companion, friend', ace. sákhāy-am, gen. sákhyur (<— *sákhyas), GAv. huš-haxā 'good companion', ace. huš-haxāim, dat. haš-e < *sékw-ō (for expected *sékw-oy-s with *-ō from «-stems), ace. *sékw-oy-iņ, gen. *s(e)kw-y-és (to the root *sekw- 'follow', cf. Lat. socius),21 and in Greek as the type of peithd: 'persuasion', ace. -5: < *-óa, gen. -d:s < *-óos.22 In addition, aerostatic inflection appears to have characterized the collectives of ablauting neuter nouns, including heteroclitic r/n- and //n-stems as well as n- and s-stems (Schindler 1975a:3-4).
Over a century ago, Johannes Schmidt (1889:191-218)
demonstrated the existence of a collective in nom.-acc. *-ōr, with the same o-grade as the animate nouns listed above: the long vowel of *-ōr has almost certainly developed from pre-PBE *-or-h2 by Szemerenyi's Law, as in *wéd-ōr and *kwétwōr below (cf. Szemerenyi 1970:155, Nussbaum 1986:129-30).
Although the paradigmatic ablaut and stress
variations have not surprisingly been largely leveled in the daughter languages, sufficient traces survive to confirm the postulation of an aerostatic paradigm for PEE: • •
• • •
*wéd-ōr 'the waters' —> *wedōr > Hitt. widār, —> *udōr > Ved. udāa (2x), Gr. sg. húdo:r 'water'; gen. *ud-n-és, loc. *ud-én > Ved. ud-n-ás, ud-án *séh2-wōl 'sun' —> PGr. *hāwel-iyos > Aiol. āwélios, Horn, ε.élios, Att. hε:lios, PGmc. *sāwil- > Goth. n. sauil, ON f. sól, —> *sh2Wōl > *s(u)wōl > Lat. sōl (proterokinetic gen. *sh2-wén-s > *suh2éns > PIr. *huwanh > GAv. xvāņg, whence backformed nom./acc. Ved. súvar, GAv. huuar* from neut. sg. *séh2-wļ? cf. R. Kim 2001b:29fn.26) *(hj)édwōl 'evil' > PAnat. *édwol > CLuv. ādduwāl (cf. derived adj. ādduwāli-), —> *edwól > Hitt. idālu- (derived from *(hļ)édwo- > CLuv. ādduwa- 'evil (adj.)'; cf. the references in ibid.) *kwétwōr 'four' —> *kwetwōr > Goth, fidwōr (Stiles 1985:85-5, 1988:116-8), Ved. catvār-i (cf. non-neut. catvār-as; accent shift already in PIE? cf. Klingenschmitt, cited in Rix 1990:44 on the "*kwetwóres rule": *C0éCioC|V > *C0eCióC)V) s-stems, cf. GAv. man-a, YAv. man-as-ca '(and) thoughts', GAv. raoc-a, raoc-ás-cā '(and) lights' < PInlr. *-ās < PIE -ōs (remodeled in Ved. mánāmsi, rócāmsi, cf. GAv. 20
For a different interpretation of the Germanic cognate, see Stiles 1984. 'Not hysterokinetic, contra Beekes 1995:180-1. 22 Perhaps also in Tocharian in the small class of nouns including TB reki. TA rake "word' < PT *rēké to the root *rek- (cf. OCS reSti, rekq 'say'), TB leki, TA lake 'bed, couch' < PT *léké to the root *legh- 'lie (down)' (Gr. lékto 'lay', OE licgan 'be lying (down)'); cf. Ringe 1996:82-3 and already Pedersen 1925:39-40fn.2. 2
33
•
varacā.hī-cā 'and energies'; J. Schmidt 1889:135-60; Hoffmann and Forssman 1996:155) «-stems, cf. GAv. nam-qm 'names', dām-qn, -qm < PInlr. *-an < PIE *-on (Ved. -a, -ān-i, cf. GAv. afšmānī 'drawbacks'; J. Schmidt 1889:98-106, Hoffmann and Forssman 1996:144)
Finally, athematic participles to root presents and aorists appear to have alternated between stressed o-grade suffix *-ónt- in the strong cases and stressed ending and zerograde suffix *-nt- in the weak cases, e.g. masc. nom. sg. *his-ōn (Ved. son, GAv. hqs, Gr. eá:n), acc. *hjs-ónt-iņ (Ved. sánt-am, Av. hantam, Gr. eónt-d), gen. *his-ņt-és (Ved. sat-ás; Av. hatō, Gr. eónt-os with nomVacc. stem and regular ending) to *h]es- 'be'. 23 The suffixal o-grade of the strong form, however, suggests that *his-ónt- may be a replacement for earlier *hiés-ont-, and that the original inflection was amphikinetic. Such was undoubtedly the case for perfect participles, whose suffix occurs in the ablaut variants *-wos- ~ *-us-. Traces of the original amphkinetic inflection have survived best in the archaic unreduplicated ptcp. of *woyd- ~ *wid- 'know': masc. nom. sg. *wéyd-wōs, acc. *wéyd-wos-iņ > Gr. eid-ó:s, eid-ót-a (<— *weid-wá:s, *weid-wóh-a, see §3.3.3; on the replacement of *-h- < *-s- by *-t- in post-Myc. Gr., cf. Szemerenyi 1967), Skt. vid-ván, vid-vāms-am (influenced by adjs. in -vant-, cf. Jamison 1991), GAv. vīduuá, YAv. vJSuuagham; gen. *wid-us-és —> Skt. vid-ús-as, GAv. vīdušō. The masc. and neut. sg. paradigms of these participles in PEE were thus
nom. sg. acc. gen. dat. loc.
masc. *his-ōn *his-ónt-iņ
neut. *his-ón (?) " *h]S-ņt-és *his-ņt-éy *hjs-ént-0
masc. neut. *wéyd-wōs *wéyd-us (*-wos?) *wéyd-wos-iņ *wid-us-és *wid-us-éy *wid-wés-0 24
The phonological interpretation of this type follows straightforwardly from the analysis of alternating root nouns above in 1. Just as e.g. PIE *h2ént-iņ, *h2ņt-és are the surface realization of unaccented nom.-acc. and accented oblique endings added to an underlyingly unaccented root, so the amphikinetic stress alternation in 'earth' reconstructed for PIE (and attested in Hittite) follows readily if both root and suffix are unaccented: the nom. and acc, being made up of three unaccented morphemes, are entirely unaccented and 23 24
0n Herakleian nom. pi. ént-es, see §3.4.2, fn. 26. 0n the fem., see §2.2.1.5.
34
surface with default initial stress, whereas the oblique cases are stressed on their first and only accented morpheme, the ending. The resulting metrical grids are given below for nom. *dhégh-ōm, *pént-oh2-s and gen. *dhgh-m-és, *pņt-h2-és: *
*
(* * h
* h
*d ég om
*
*)
*d h e^ h om
s > *dnétfh-ōm
*
es
* (* *
*pént
(* (*)
> *dndh-m-és *
*
*)
oh2
s
*pent
*
*
ofi2
es
(* (*)
> *pņt-h2-és
The loc. sg. may also be analyzed in a similar manner, but its peculiarities of ablaut and stress call for special comment. The oldest loc. sg. ending reconstructible for PIE is *0 , the "endingless" locative which has left clear traces especially in Anatolian (cf. Neu 1980) and Indo-Iranian n-stems, e.g. Ved. mvrdhán 'on the head' (Whitney 1924:158-9). The derivation of loc. sg. *dhghém? (= *[dhdhsém], cf. Hitt. tagāri) and similar forms was probably in origin similar to that of other oblique cases such as gen. *dhghm-és, except that here the ending was *-0, so that the accent of the ending was retracted one syllable leftward to the suffix, just as with Jer Deletion and Accent Retraction in Slavic (§4.2.3). * *
*
*dhegh om
(* (*)
0
> *dhghém.
To this endingless form, the "hic-et-nunc" particle *-i (i.e. "here-and-now" particle of near deixis) began to be added probably already in (late) PIE, so that the productive ending in those ancient IE languages which preserve reflexes of the locative is *-i, e.g. Skt., Av. -/, Gr. dat. -/, etc. (Brugmann 1911:174-85, esp. 179; Benveniste 1935 with refs.). The resulting forms, such as Ved. ksám-i, pitár-i, murdhán-i (to ksās, pitā 'father', murdhá 'head'), along with locatives to root nouns such as **h2ént-0 > PIE *h2ént-i (> Hitt. hanz(a), Skt. anti\ §2.2.1.1), **lcérd-0 > **£ēr-0 2 5 —> *kēr-i (> Horn. kε:ri 'at heart'. §3.5), retained the inherited stress and full-grade of the suffix and so continued to be 25
By Szemerenyi's Law (cf. §2.2.1.1). Neu (1980:31-3) discusses and rejects the proposed occurrences of Hitt. loc. SA-/r/kēr/ (parallel to E-i'r/per/ 'in the house'), which would continue this endingless preform.
35
distinguished from other oblique cases of the singular, and of the dual and plural. Initially, they must have been treated as endingless forms, with accented zero-ending and retraction, to which -/ was postlexically attached; this would have remained the most obvious analysis as as long as the earlier endingless forms remained as competing variants in the language. With the disappearance of *ksám, *pitár, míīrdhán, etc., however, the locatives in -/' could be reinterpreted as having accented suffix and ending -(i, i.e. *R-(S'-(i, e.g. *pit-(ar-(i, *rmlrdh-(an-(i. As the only oblique case form with a stressed full-grade suffix, however, they would have been prime candidates for remodeling: hence later Skt. murdh-n-i after the pattern of gen. mvrdh-n-ás, dat. murdh-n-é, instr. múrdh-n-ā(cf. §6.5, fn. 16). Such "regularized" locatives behave exactly like all other oblique forms built to the "weak" stem: the stress falls on the accented ending -/, just like Gr. gen. -as, Skt. gen. -ás, dat. -é, instr. -á. The existence of this class, with its alternation between root-stress (here interpreted as default initial stress by left head projection from line 1) and ending-stress, confirms the analysis offered in 1 and suggests that, at least originally, the endings of the direct and oblique cases in PIE were unaccented and accented, respectively. The associated ablaut alternations between the strong and weak forms must date from a period when unstressed morphemes automatically surfaced with 0- (or in certain cases, such as the "strong" suffixal variant of amphikinetic nouns, o-grade) and stressed morphemes with e-grade. Given the large number of exceptions to this generalization in the reconstructible PIE protolanguage — one need only recall simple thematic presents such as 2pl. *bhér-e-te. which dates at least to the last common ancestor of non-Anatolian PIE, or *sepuņ 'seven' — this originally phonological rule had long since become morphologized in alternating noun and verb paradigms; thus its operation in amphikinetic (and protero- and hysterokinetic) nouns, as in root nouns, indicates that the type is archaic and must already have existed in pre-PIE, when ablaut was phonologically associated with surface stress. We have seen that amphikinetic inflection may be confidently reconstructed for only a small number of (animate) nouns, neuter collectives, and the oldest athematic type of present and perfect active participle; in many of these cases, this reconstruction is based principally on reflexes of suffixal o-grade, e.g. in coll. *wéd-ōr 'the waters' or perf. act. ptcp. *wéyd-wōs, *-wos-iņ. Thus this type of accentual mobility must have been recessive already in PIE, perhaps even more so than the root-accented aerostatic class described in §2.2.1.2, and this raises some questions. If, as has long been assumed, amphikinesis in suffixed nouns is the exact counterpart of the root-ending alternation of root nouns, why do 36
only a handful of such nouns survive in even the archaic PIE languages? Certainly not because of some typological, much less universal, constraint against stress alternations over more than one syllable: as we shall see in chapter 4, Balto-Slavic offers numerous examples of unaccented stems which stress either the initial syllable or the ending, depending on whether the ending is unaccented or accented: cf. Russ. skovorod-á 'frying pan', pi. skóvorod-y, póvest' 'story', gen. pi. povest-éj, Lith. kātil-as 'kettle', loc. pi. katiluosé 'in the kettles', atmint-is 'memory', ace. ātmint-ļ. As with accented roots and aerostatic paradigms, it may not be possible to determine the reason for the disappearance of the root-ending alternation. Let us next consider proterokinetic and hysterokinetic inflection and see if their analysis offers any clues to the development of the PIE system of nominal accentuation. 2.2.1.4. Proterokinetic stems In this type, reflexes of which are widely attested in ancient IE languages, the stress alternates between the root in the strong cases and the suffix in the weak cases. In the strong cases, the root occurs in full-grade and the suffix in zero-grade; conversely, the weak cases exhibit zero-grade of the root and full-grade of the suffix. In other words, the stressed morpheme surfaces with full e-grade, and the unstressed morphemes (stem and ending in strong cases, root and ending in weak cases) with zero-grade, so that the alternation may be summarized in the formula *R(é)-S(0)-E(0) ~ *R(0)-S(é)-E(0) (cf. Schindler 1975b:263-4). This alternation may be reconstructed for the majority of i- and «-stems, such as deverbatives in *'-ti— *-téy- and *-'-tu— *-téw-; most suffixed neuter r/n-stems, including e.g. those formed with the suffix *'-wŗ ~ *-wén- (Schindler 1975a:9-10); almost all neuter ¿-stems (Schindler 1975b); and certain types of «-stems, in particular neuter nomina actionis formed with the suffix *'-mņ ~ *-mén-. Examples of each of these types are given below. nom. sg. ace. gen.
*h2éw-i-s 'bird' *h2éw-i-m *h2W-éy-s
nom./acc. sg. *péh2-wŗ 'fire' gen. *ph2-wén-s
*dhéhļ-t-i-s 'setting' *gwém-t-u-s 'coming' *dhéh|-t-i-m *gwém-t-u-m h *gwiņ-t-éw-s *d h]-t-éy-s *genhi-(o)s 'race, kind' *bhér-m-ņ 'burden, load' *gņhi-és-(V)s *bhŗ-m-én-s
37
In the case of /- and «-stems, particularly deverbatives in *-ti- and *-tu-, the evidence for proterokinetic inflection is especially robust: 1. The existence of accentual alternants in Vedic, e.g. usual mati- vs. rare máti'thought'. 2. Forms with and without Verner's Law in Germanic, e.g. Goth, ga-qump'assembly', lit. 'coming together') < PGmc. *kwumþiz < *gwiņ-ti- vs. missadēd- 'misdeed' < PGmc. *dzēðiz < *dhē-tí-. 3. Fluctuation between full- and zero-grade of the root, e.g. Goth, (missa-)dēd-, OE died, OHG tāt 'deed, act' < PGmc. *d<ē- < PIE *d h eh]- vs. Gr. thésis 'setting, situation' < PIE *d h hi- (regular zero-grade in Greek), Ved. vés 'bird'< *h2wéy-s (vis with secondary zero-grade) vs. Lat. avis < *h2éw-i-s. (Note the mismatch in ablaut and accent in Goth, (ga-)qump- vs. missa-dēd-\) Cf. also Hitt. tēpu- 'few' < *débu- vs. pānku- 'all' < *bņgu-, Hitt. Salli- < *selh2-i'great' vs. CLuv. Salhi(tti)- 'growth' < *sļh2-i- (Melchert 1994a:89). For suffixed r//i-stems such as 'fire', Schindler (1975a:9-10) notes the existence of reflexes of full-grade of the root in Hitt. pahhur, Luv. pāhur < PA *páHur < *péh2-wŗ vs. zero-grade in Gr. pu:r, gen. pur-ós, Umbr. pir [plr] < PItal. *pur, Goth./o/z, gen. fun-in-s < PGmc. *fun-, both remodeled from *ph2-un-(és) <— *ph2-wén-s; the full-grade of the suffix is attested in the Hitt. gen. pahhwenaS. The assumption of original proterokinetic inflection for neuter s-stems elegantly accounts for the contrast between o-grade of the suffix in nom./acc. *-os and e-grade in oblique *-es-; the former has replaced zero-grade *s, which survives in s-stems to laryngeal-final roots such as *kréwh2-s > Ved. kravis, Gr. kré(w)as (Schindler 1975b:265-6).26 In n-stems, finally, the alternation between zero- and full-grade of the suffix *-mn ~ *-men- is well attested in a number of languages in which the type enjoyed some productivity: cf. gen. GAv. -m-āņg, YAv. -m-qn (for -m-q*) < Plr. *-m-anh < *-men-s, e.g. GAv. ðašmaņg to Caiman- 'field of view', barasmqn to barasman- 'sacrificial straw' (Hoffmann and Forssman 1996:142-4); Olr. nom./acc. céim(m) 'step' < *-men < *-mņ, gen. céim(m)e < *-mēs < *-mens (also verbal nouns in *-(s)men-; Thurneysen 1946:212-3,453); OCS gen. vrē-men-e (to vrē-mq 'time') < PS1. *ver-men-e < *-men-es <— *-men-s (see §5.1.3). Once again, as with the amphikinetic class (§2.2.1.3, fn. 17), we find that the suffixal ablaut alternations have often been
26
The original root-ablaut alternation survives e.g. in Gr. krátos 'strength, might' (Horn, also kártos) < *kŗt- vs. Aiol. krétos < *kret-, continuing *krét-os, *kŗt-és- (cf. Horn, kratús 'strong', Skt. krátus, Av. xratuS '(mental) strength').
38
preserved, whereas the alternations of root ablaut have generally been leveled in one direction or another in the different languages.27 The analysis of proterokinetic stems appears at first to present an obstacle for the brackets-and-edges theory. As pointed out above in § 1.2.4, the selection of parameters posited for the PIE system easily accomodates stress alternations between initial and final syllables: the former receive their stress by default in unaccented forms, while the latter contain an underlyingly accented final morpheme, e.g. Russian ace. bórodu, ná borodu vs. nom. borodá. In order to account for the alternation between root and suffix, one must assume that the root is unaccented, and that the suffix in the weak cases, but not the strong cases, is accented: otherwise both strong and weak cases would surface with suffixal stress. But why should the strong allomorph of e.g. the /-stem nomina agentis suffix Midiffer in accent from the weak form *-tey-, as the two are obviously stress-conditioned ablaut variants of a single suffix? Such an ad hoc assumption, if correct, would require additional phonological and/or historical motivation. However, a paradigm with different suffixes in the strong and weak cases does exist, namely in r/n-stems of the type of 'fire'. For this stem class, one may reasonably assume that the strong suffix *-wŗ and the weak suffix *-wen- are underlyingly unaccented and accented, respectively. The metrical grids for nom/acc. *péli2-wŗ and gen. *ph2-wén-s are therefore *
*
(* *
(* *
*
*péh2 wŗ
/-*
*peh2 wén
*
es > *ph2-wén-s
Thus the strong case forms such as *péh2-wŗ are unaccented and surface with default initial stress, and the stress in the weak case forms falls on the accented weak morpheme *-wén-. I propose that this principle has been extended to the /-, u-, and /i-stems, resulting in an opposition between unaccented suffixes *-i-, *-u-, *-(m)n- in the nom.-acc. and accented *-ey, *-ew-, *-(m)en- in the oblique cases. In each of these stem classes, the suffix and ending have combined into a unanalyzable desinence, e.g. /-stem sg. anim. nom. *-is, ace. *-im vs. gen. *-eys, dat. *-eyey, etc. Although the exact timing of this development cannot be determined, it is clear that even in the most archaic IE languages such as Vedic Sanskrit, desinences such as /-stem -is, -im, -es, -aye were no lonser 27
0n athematic present active participles in *-ónt- ~ *-nt-, see above, §2.2.1.3.
39
immediately recognizable as consisting of discrete suffix plus ending, and were almost certainly learned and stored in the grammar as specifically /-stem inflectional morphemes; similarly for «-stem -us, -urn, -os, -ave.28 Most likely, then, proterokinetic stems such as *h2éw-i-s, gen. *h2w-éy-s, *gwém-t-i-s, gen. *gwiņ-t-éy-s were morphologically analyzed as *h2éw-is, *h2w-éys and *gwém-t-is, *gwiņ-t-éys, already in late PIE, with unaccented nom. *-is vs. accented gen. *-eys. The corresponding metrical grids would have been * (* *
* *\
*
*gwém t is
(* (*)
*gwem t éys > *gwiņtéys
In support of this hypothesis, note that proterokinetic stems are the only class of ablauting nouns (other than aerostatic stems) in which the weak-case ending is reconstructed as unstressed, with zero-grade ablaut. As we have seen above (cf. 2), aerostatic nouns, which constitute a marginal and disappearing class already in PIE, have almost entirely replaced their inherited unstressed zero-grade suffixes and endings with one of three possibilities: stressed full-grade suffix and zero-grade ending, as in proterokinetic stems; zero-grade suffix and stressed full-grade ending, as in amphikinetic and hysterokinetic stems; or generalization of the productive ending, e.g. gen. sg. *-es or *-os, without stress shift. Cf. Schindler's (1975a:6-7) examples of gen. *wéd-ņ-s, *dér-u-s, remade as *wed-én-s, *dr-éw-s (whence proterokinetic Skt. dóru, drós), or *wed-n-és, *der-w-és (*-ós; cf. Horn, dóru, do.rós < *dorwós), or *wéd-n-es, *dér-w-es (*-os; cf. Hitt. genuwaš). These morphological innovations betray a tendency to regularize the surface realization of individual morphemes, roots, stems, and especially endings, by generalizing one of their variants, usually the full-grade stressed variant. As a result, each IE language 28
In Hittite, «-stem proterokinetic endings such as gen. -awaS (< *-ew-os), dat. -awi (< *-ewey; Melchert 1994a: 138, cf. already Hrozny" 1917:9) have become specialized to adj. inflection, e.g. tēpuš 'few' < *dhébh-u-s, gen. tēpawaš <— *dhbh-éw-(o)s, whereas nouns have generalized -u(w)- from the nom. and ace, e.g. haSSuS (LUGAL-uS) 'king', gen. -(u)waS, dat./loc. -(u)e, -/', abl. -waz, genu 'knee', gen. gēnuwaš, abl. gēnuwaz. (/-stem adjectives have apparently undergone loss of intervocalic *-y- and deletion of the suffixal vowel, e.g. gen. sg. *-ey-os > *-eos > Hitt. -aš; cf. Melchert 1994a: 176). As the plene spelling of tepawaS indicates, Hitt. has eliminated the inherited stress alternation and generalized root stress to the oblique cases. Note that the BS1. oblique pi. /- and «-stem endings are still analyzable as weak stem vowel *-i-, *-u- and stressed suffix (PBS1. pi. gen. *-ō:n, dat. *-mas, instr. *-mins or *-mi:s, loc. *-su. du. dat./instr. *-mV), but these could have been reformed on the model of consonant-stem endings at time in the prehistory of PBS1. before the latter became assimilated to the /-stems and/or adopted /-stem endings. On this problem, and the reconstruction of the i- and w-stem pi. endings in general, cf. §5.3.
40
has generalized a single ending for each of the oblique cases to all non-thematic nouns except for proterokinetic stems. In the gen. sg., for instance, Greek, Celtic, and probably Anatolian and Tocharian have generalized *-os (> Gr. -os; Gaul, -os, Olr. -C; Hitt. -aš; TB -e), Balto-Slavic *-es (> OLith. -es, OCS -), Indo-Iranian either *-es or *-os (> PInlr. *as > Skt. -as, Av. -ō, OP -a); both *-es and *-os are continued in Italic, cf. Lat. -is but OLat. Vener-us, sēnātu-os.29 The proterokinetic stems, particularly the /- and «-stems, have "gone their own way" in Indo-Iranian, Greek, Italic, Celtic, Germanic, and Armenian, so that they must be treated as separate inflectional classes, with idiosyncratic "/-stem" and "w-stem" desinences, in the synchronic grammar of each of these branches. The preceding discussion holds consequences for the underlying accentuation of feminines with the "motion-suffix" *-ih2- ~ *-yeh2- to various athematic stems, including «-stem adjectives, present active participles in *-ónt- ~ *-nt-', and perfect active participles in *'-wos- ~ *-us-' (see §2.2.1.3). Alongside these derived feminines, one also finds a presumably older layer of ablauting proterokinetic stems, e.g. *s£éh2-ih2- ~ *skh2-yéh2'shadow' (cf. Skt. chāyā with generalized full-grade, Gr. skid: with generalized zero-grade of the root). The analysis of such nonderived ih2-stems is exactly the same as that of the proterokinetic nouns analyzed above: one must assume that the zero-grade variant *-ih2- in the nom./acc. is unaccented, whereas full-grade *-yeh2- in gen. *-yeh2S, dat. *-yeh2i, etc. is accented. * *
*slcéh2
* *\
*
ih2
f*\
*slceh2 yeh2 s > *slch2yéh2S
This assumption by itself, however, does not account fully for the reconstructible inflection of the previously mentioned classes of derived feminines. Judging from the evidence of IE languages, it appears that feminines to «-stem adjectives, for instance, had consistent zero-grade of the root, with stressed full-grade suffix *-éw- in the nom./acc. and zero-grade suffix *-u- and stressed "motion-suffix" *-yéh2- in the oblique. Thus, while the masc./neut. contrasted e.g. nom./acc. *mrégh-u- 'short' (cf. Lat. brevis, regularly remade as /-stem) with obi. *mŗgn-éw- (cf. brakhús, Skt. mŗhú- with zero-grade), the contrast of ^Similarly, the dat. sg. ending, in PIE underlyingly *-ey with the unstressed zero-grade allomorph *-i, has generalized *-ey in most IE languages. The discrepancy in function and semantics between the dat. and loc. renders implausible the older hypothesis that loc. *-i is in origin an ablaut variant of dat. *-ey: cf. §2.2.1.3.
41
Gr. fem. bar-eia, -eia:s 'heavy' < *gwŗH-éw-ya, *-éw-yās < *-éw-ih2, *-éw-yeh2 and Skt. fem. gur-ví, gur-vyās < *-w-fli2, *-w-yéh2-s points to an earlier alternation *-éw-ih2- ~ *u-yéh2-. Similarly, the feminine of present active and perfect active participles alternated between stressed full-grade suffix in the nom./acc. and stressed *-yéh2- in the oblique: cf. PIE nom. sg. *his-ént-ih2, gen. *hļS-ņt-yéh2-s (—> Skt. satí, satyās, Av. ace. haiti:m, Gr. Horn, éassa, eássε.s [on the stress see §3.6, fn. 33]; PIE nom. sg. *wid-wés-ih2, gen. *wid-us-yéh2-s (—> Skt. vidúsī, vidúsyās, Gr. Horn, iduia, iduia:s; Att. eiduia, eiduia:s with root ablaut from masc. eido.s, neut. eidós). These paradigms are recapitulated below for all three genders in the sg.:
nom. sg. ace. gen. dat.
nom. sg. acc. gen. dat. nom. sg. acc. gen. dat.
masc. neut. h *mréá -u-s *mréáh-u h *mrétí -u-m " *mŗsdh-éw-s *mŗJtíh-éw-i
fem. *mŗ;áh-éw-ili2 *mŗ>áh-éw-ih2-.rņ *mŗdh-u-yéh2-s *mŗ)gh-u-yéh2-i
masc. *hjs-on *his-ónt-iņ
fem. *hļS-ént-ih2 *his-ént-ih2-iņ *hļS-jjt-yéh2-s *his-ņt-yéh2-i
neut. *h]S-on(?) " *hļS-ņt-és *hjs-ņt-éy
masc. neut. *wéyd-wōs *wéyd-us (*-wos?) *wéyd-wos-rņ " *wid-us-és *wid-us-éy
fem. *wid-wés-ih2 *wid-wés-ih2-iņ *wid-us-yéh2-s *wid-us-yéh2-i
Thus the alternation in the fem., although traditionally described as "proterokinetic", does not involve the first two morphemes, i.e. root and suffix, but rather the two suffixes: the root appears in zero-grade throughout the fem. paradigm. I tentatively suggest that this pattern reflects a constraint, perhaps operative already in PIE, limiting ablaut/accent alternations to the final three syllables of the (phonological) word. Such forms as *mŗ;ghéw-ih2(-iņ) or *wid-wés-ih2(-rņ), then, are underlyingly unstressed, but the expected default stress does not fall on the leftmost syllable of the entire (phonological) word, but rather on the that of the stressable domain, hence on the suffix *-éw-, *-ént-, or *-wés-.30 30
Similarly, the fem. to amphikinetic PIE *péyh2-wōn, gen. *pih2-un-és 'fat' (Skt. plvan-, Gr. pi:(w)o:n; §2.2.1.3) was most likely *pih2-wér-ih2- - *pih2-ur-yéh2-; cf. Skt. pívarī, Gr. pi:e:ra.
42
Whether or not this explanation for the stress alternation of derived feminines is correct, the historical and synchronic analysis of proterokinesis in single-suffixed nominal stems seems clear. In sum, the generally accepted morphological reconstruction of root, stem, and ending in proterokinetic i-, u-, and n-stems, with stress alternation between root and stem, was almost certainly valid at some stage of pre-PIE. By the time of the breakup of PIE, proterokinetic nouns had been reanalyzed as consisting of a root and a stemspecific desinence, thus obviating the need for an ad hoc assumption of a suffix which was variously unaccented or accented (in strong and weak cases, respectively). The stress alternation in question therefore reduces to one between root and desinence, and may be analyzed exactly as that in unaccented root nouns in 1 above: strong-case desinences, e.g. /-stem anim. nom. *-is, ace. *-im, are unaccented, and weak-case desinences, e.g. gen. *eys, dat. *-eyey, are accented. 2.2.1.5. Hysterokinetic stems Alongside the numerous proterokinetic stems, one finds a smaller set of nouns which alternate in stress and full-grade between the suffix and ending, hence strong *R(0)S(e)-E(0) vs. weak *R(0)-S(0)-E(e). Unlike the proterokinetic class, which includes neuters (e.g. nomina actionis in *-'mņ ~ *-mén-), hysterokinetic stems appear to have been confined to animate gender, like amphikinetic stems (except for neuter collectives; cf. Schindler 1975b:263). The clearest examples of this type are the r-stem relationship nouns *ph2tēr 'father', *dhugh2tēr 'daughter', and a handful of n-stems such as *h2fsen 'male'. nom. sg. ace. gen.
*ph2t-er (< **-ér-s) *ph2t-ér-iņ *ph2t-r-és
*h2fs-ēn *h2fs-én-iņ *h2ŗs-n-és
In addition, several /'- and «-stem nouns with exceptional endings, as well as compound adjectives built to proterokinetic /'- and «-stems ("exocentric" derivatives), belonged to this class. These have been fairly preserved in Indo-Iranian; cf. Kuiper 1942, Beekes 1972 for discussion. Cf. the following uncompounded «-stem forms, taken from Hoffmann and Forssman (1996:129-32): nom. sg. *-āuš for *-auš < *-éws (long vowel from ace.) Av. hiβ-āuša 'companion', OP dahy-āuš 'land' (cf. YAv. daghuš, Ved. dásyus) ace. *-aum, *-āum, *-āwam <— *-ām < *-ēm < *-éwm (Stang's Law; cf. PIE *dyēm < **dyéwm, §2.2.1.1): 43
Av. daifh-aom, OP dahy-āvam, dahy-āum gen. *-was < *-w-és: Av. pas-uuō to neut. pasu- 'cattle' (Skt. pášu), xraβ-βō, Skt. krát-vas (alongside proterokinetic GAv. xrat-āuš, Skt. krát-os with *-aus < *-éw-s) to xratuš, krátus '(mental) strength' dat. *-wai < *-w-éy: Av. xraβ-βe, Skt. krát-ve instr. *-wā < *-w-éh]: GAv. xraβ-βā (alongside xrat-u, fn. 15), Skt. krát-vā nom. pi. *-āwas, *-was <— *-éw-es (?): Av. pasuuas-ca; dajjhāuuō, OP dahyāva ace. pi. *-āwas *-was <— *-éw-ijs (?): Av. pas-uuō, Ved. paš-vás; Av. dajjhāuuō, OP dahyāva gen. pi. *-wām < *-w-ōm: Av. pas-uuqm Hysterokinetic endings are also regular in compounds such as Av. daragō.bāzāuš 'longarmed', uγra-bāzāuš 'strong-armed' to proterokinetic bāzu- 'arm' (nom. bāzuša). Elsewhere in IE, Latin /-stem nouns with nom. sg. -ēs for usual -is, e.g. vātēs 'seer, prophet', vulpēs 'fox', are also probably hysterokinetic in origin, and nouns in -ēs of the type of māteriēs 'material, stuff (alongside materia) continue a pre-Latin paradigm with nom. sg. *-ēs, acc. *-ām < hysterokinetic PIE *-eh2-s (for *-éh2-s, with long-grade from sonorant-stems), *-eh2-m, with leveling after either nom. or ace; cf. Mayrhofer 1986:1334 with refs. Other alleged reflexes of hysterokinesis are less certain: cf. Schindler 1976 on the origin of the much-discussed type of Gr. hipp-eús 'horseman' (§3.4.2), and see §5.1.3 on Slavic n-stems in PS1. *-e., supposedly from hysterokinetic *-in. Finally, Schindler (1975a:9) is surely correct in interpreting OHitt. neut. utnē, gen. utniyaš as an original PIE collective *udn-ēy, *udn-i-' (ultimately related to *wed-(e)n-, obi. stem of 'water'? cf. Melchert 1994a: 161). The accentual analysis of these stems encounters the same initial obstacles which faced the proterokinetic stems in §2.2.1.4 above. If the stem is accented in the strong cases, whence the stem stress of e.g. acc. *ph2t-ér-iņ, why should it not also be accented in the weak cases as well? Yet instead of the expected gen. *ph2t-ér-s, the comparative evidence points to *ph2t-r-és, with zero-grade of the stem and stress on the ending: cf. Gr. patrós, Lat. patris, Olr. athar < *atros, Arm. hawr < *patros (Skt. pitúr < *pitŗs for *pitrás has acquired originally aerostatic *-ŗs from 'brother'; cf. §2.2.1.2). At least in cases such as 'father', the traditional analysis into root + stem + ending may not accurately reflect the actual synchronic morphology of late PIE. Consider the five 44
reconstructive r-stem relationship nouns reconstructible for PIE: *ph2tēr 'father', *méh2tēr (or *meh2tēr) 'mother', *bhréh2tēr 'brother', *dhugh2tēr 'daughter', *swésōr 'sister'. The last of these obviously stands apart, with its amphikinetic nom./acc. o-grade suffix and differing shape (cf. 3 above). The other four, however, all contain the sequence *-h2tēr ~ *-h2ter- ~ *-h2tŗ-, which suggests that it is the latter, and not merely *-ēr ~ *-er~ *-ŗ-, which should be taken as the suffix. In that case, we are left with a "root" for 'father' consisting of nothing more than *p hardly a viable root shape for any 31 recoverable stage of the protolanguage. It therefore seems reasonable to analyze 'father' as consisting of a unitary root *ph2ter- ~ *ph2tr- and number-case endings, with full-grade ablaut and stress characterizing the former in the strong cases and the latter in the weak cases; similarly for 'mother' and 'daughter'. The accentual structure of ace. *ph2tér-iņ and gen. *ph2tr-és is exactly parallel to that of an alternating root noun: * (* *
*ph2tér
* *\
*
in
*ph2ter
(* r*\
es > *ph2tr-és
Once such a pattern had become established, it is conceivable that it could have spread to mark other nominal stems, in particular the exocentric derivatives mentioned above of the Av. -bāzāuš 'X-armed' type, but the exact details of their (synchronic) metrical analysis remain unclear. With the exception of these hysterokinetic stems and of derived feminines in *'ih2- ~ *-yéh2- (§2.2.1.3), then, it appears that all of the stress patterns reconstructed for PIE find a ready explanation within the brackets-and-edges framework. The root ~ ending alternation of root nouns and amphikinetic stems is easily analyzed in terms of unaccented strong vs. accented weak endings; aerostatic roots are underlyingly accented and therefore surface with columnar root stress. In the proterokinetic and hysterokinetic classes, the alternations of root ~ suffix and suffix ~ ending reconstructed on comparative evidence had probably become simple root ~ ending alternations (or root ~ desinence in proterokinetic nouns) by late PIE.
3
'To be sure, the root of 'daughter' under this analysis would be *dhug-, which is a possible PIE root shape; however, no such nominal or verbal root in reconstructible for PIE.
45
In closing, it should be emphasized that the evidence adduced above for the various ablaut and accent classes varies greatly from one class to another, and also with respect to ablaut and to stress. As we have seen, some of the accentual types reconstructed for PIE are represented by a limited number of stems: this is true of the aerostatic and amphikinetic classes, and to an extent also for the hysterokinetic class. Moreover, most of the evidence for these reconstructions is indirect, being confined to alternations in ablaut grade in suffixes and endings, and very occasionally in roots; such alternations are especially numerous in Indo-Iranian and Greek, but also survive in Germanic, Italic, Celtic, Tocharian, Armenian, etc., particularly in r- and n-stems (cf. 4 and 5 above). Direct reflexes of accentual alternations are frustratingly sparse, at least outside Balto-Slavic (on which see chapter 4): excepting the well-attested alternation of root ~ ending in root nouns in Sanskrit and Greek (plus a few other nouns, e.g. Greek ace. gunaik-a, gen. gunaik-ós 'woman'), there exist only a handful of relics in Anatolian, particularly in Hittite: aerostatic nom./acc. wātar 'water', gen. widenaSa, amphikinetic coll. widār (itself from *wedōr, replacing *wédōr!), nomVacc. tēkan 'earth', dat. taknī, loc. t(a)gān. 2.2.2. Thematic steins The preceding considerations merely highlight what has long been recognized, that the "kinetic" system of alternating ablaut grades and stress on the root, suffix, or ending was already archaic and of decreasing productivity in PIE. With the partial exception of Anatolian, this system is continued for the most part by unproductive historical relics in the oldest IE languages; except for a handful of derivational suffixes, e.g. proterokinetic *-mņ ~ *-men-, the ablauting inflections are in retreat already at the earliest recorded stages of Sanskrit and Greek. Beginning in PIE, the nonablauting thematic stems in *-o- and *-eh2become steadily more numerous and productive: they are well attested in all IE languages, and in each branch, their frequency with respect to the older ablauting stems increases over time. By the 1st millennium BC, when Homeric Greek and Old Latin are first attested, oand eh2- (> ā-) stems have become far and away the dominant classes of nouns and adjectives, to which the inflectionally more archaic consonant-stems tend to become assimilated. In Sanskrit and ancient Greek, thematic stems are always marked by columnar stress: they are either stem-stressed (barytone) or ending-stressed (oxytone). With the exception of the vocative, which receives initial stress as a result of a postlexical rule of accent erasure, alternations of underlying accent are absent from the inflectional paradigms 46
of o- and eh2-stems in these languages. This columnarity is evident from the paradigms of two typical Vedic a-stems (< PIE o-stems), barytone vŗka- 'wolf and oxytone devá'god', as well as the d-stems sénā- 'army' and priyā- 'dear, friend (f.)':32 priy-á priy-ām priy-áyā priy-ā-yāi priy-ā-yās
dev-á-s dev-á-m dev-éna dev-āya dev-āt dev-á-sya dev-é dév-a
sén-ā sén-ām sén-ayā sén-ā-yāi sén-ā-yās
nom./acc. du. vŗk-ðu instr./dat./abl. vŗk-ābhyām gen./loc. vŗk-ayos
dev-āu dev-ābhyām dev-áyos
priy-é sén-e sén-ā-bhyām priy-ā-bhyām sén-ayos priy-áyos
vŗk-ās vŗk-ān vŗk-āis vŗk-ebhyas vŗk-ānām vŗk-esu vŗk-ās
dev-ās dev-ān dev-āis dev-ébhyas dev-ānām dev-ésu dév-ās
sén-ās sén-ās sén-ā-bhis sén-ā-bhyas sén-ā-nām sén-ā-su sén-ās
nom. sg. acc. instr. dat. abl. gen. loc. voc.
nom. pl. acc. instr. dat./abl. gen. ioc. voc.
vŗk-a-s vŗk-a-m vŗk-ena vŗk-āya vŗk-āt vŗk-a-sya vŗk-e vŗk-a
?!
sén-ā-yām sén-e
?í
priy-ā-yām príy-e
priy-ā-s priy-ā-s priy-ā-bhis priy-ā-bhyas priy-á-nām priy-ā-su príy-ā-s
In Greek, the original columnarity has been disturbed by the operation of the ThreeSyllable Rule in stems of more than one syllable; cf. §3.3.1 for details. As far as monosyllabic stems are concerned, however, the only deviation from columnar stress is found in the gen. sg. of barytone a.7ε:-stems (< PIE e/12-stems), where the PIE pronominal ending *-eh2ōn > PGr. *-a:ho:n > Horn. -á:o:n receives stress by the ThreeSyllable Rule, then undergoes contraction to produce the always circumflex-stressed ending -5:n in classical Attic. The columnarity of stress in Greek o- and a.7ε:-stems is illustrated below for barytone lóg-o-s 'word' and k''o:r-a: 'country', and oxytone mis&ó-s 'reward', bo:l-ε: 'will, counsel': nom. sg.
lóg-o-s
khó:r-a:
tnisAo-s
bo:l-ε:
32The endings here are Classical Sanskrit. In Vedic, one also finds older endings in instr. sg. -ā, nom./acc. du. -a (more frequent than -āu in the Vedas), as well as nom. pl. -Ssas (cf. Av. -áqhō, OP aniyaha bagaha 'the other gods', PWGmc. *-as < *-asaz?) and instr. pl. -ebhis (originally pronominal, like dat./abl. -ebhyas and loc. -esu). In Vedic, acc. pl. -an becomes -Snis in internal sandhi before a following vowel: this ending has perhaps been remodeled from *-ān < *-ōn < *-ons by Szemerenyi's Law (cf. R. Kim 200Ib: 129-30).
47
acc. gen. dat. voc.
lóg-o-n lóg-o: lóg-o:i lóg-e
khá:r-a:-n khá:r-a:s khó:r-a:i khá:r-a:
misth'ó-n mist^n: misth-5:i misth'é
bo:l-é:-n bo:l-ε:s bo:l-ε:i bo:l-é:
n./a./v. du. gen./dat.
lóg-o: lóg-oin
khá:r-a: khá:r-ain
misfró: misth-oín
bo:l-á: bo:l-aīn
nom./voc. pl. acc. gen. dat.
lóg-oi lóg-o:s lóg-o:n lóg-ois
khá:r-ai khá:r-a:s kho:r-5:n khó:r-ais
mist,uoí mistluó:s misth-5:n misth-oīs
bo:l-aí bo:l-á:s bo:l-ð:n bo:l-áis
How are these two accentual classes of thematic nouns to be analyzed? The synchronic interpretation of barytone and oxytone columnarity depends first on the underlying form of the thematic vowel *-%-. If the thematic vowel is accented, barytone stems must also be accented in order to receive the stress by the parameters of stress computation given in §1.2.4, whereas oxytone stems may be underlyingly unaccented. In that case, the metrical grids for PIE *é£w-o-s 'horse' and *yug-ó-m 'yoke' would be * (* (*.
(*)
*élcw
OS
* * *yuil
* (* (*)
óm
If, however, the thematic vowel is unaccented, barytone stems must also be unaccented, with default initial stress by left head marking on line 1 and Line 2 Projection. In that case, oxytone stems must be prespecified with a left bracket to the right of their final or only syllable head; i.e. oxytone stems are underlyingly postaccenting. Under this analysis, the metrical grids for *élcw-o-s and *yug-ó-m would be * (* *
*\
*é£w os
* (* *)
*(
*yug óm
The latter alternative is to be preferred, for a number of reasons. First of all, the ancient IE languages provide no compelling indication that the thematic vowel was accented (in nominal inflection; on verbal thematic inflection cf. §§2.3.2,5.5). In addition, this analysis
48
requires fewer morphemes to be marked as accented in the grammar and for that reason is more transparent to the (child) learner. The underlyingly unaccented nature of barytone thematic nouns, in particular ostems, is supported by two further pieces of evidence. As we shall see in the following section, masculine and neuter o-stems were capable of forming derived collective nouns with the suffix *-eh2. Data from a variety of IE languages suggest that the collectives corresponding to barytone o-stems carried stress on the suffix *-éh2, i.e. were oxytone. If barytone stems are underlyingly accented — as would be required if the thematic vowel *e /0- were accented — one would have to assume that collective *-eh2 is a dominant accented suffix, which erases any underlying stem accent. Under the second hypothesis, *eh2 is merely an ordinary accented suffix attached to an unaccented stem, yielding a form with oxytone stress, e.g. *orséh2 'rear end area' to *órsos 'ass', *kw(e)kwléh2 'rings, wheels' to *kwékwlos 'wheel'. Finally, unaccented o-stem nouns are actually attested in Slavic, particularly in Old Russian and, to a lesser extent, Middle Bulgarian manuscripts: they surface with default initial stress when occurring in isolation (i.e. when constituting a stress domain by themselves), but "lose" their stress to an enclitic or proclitic, or to a closely linked preceding word (for examples cf. §4.3.2). Obviously one ought to be wary of positing an analysis for PIE which has left direct traces only in Slavic, one of the latest recorded IE branches (and in medieval Slavic languages at that!). Yet i/this property is an archaism, it follows that the nominal stems in question must have been unaccented in PIE, and that they have generalized default initial stress in Sanskrit and Greek as well as Germanic and Lithuanian. If this interpretation is correct, it follows that thematic stems in (late) PIE fell into two accentual classes, unaccented (barytone) and postaccenting (oxytone). Significantly, there appear to have been no stem-accented nominal thematic stems of the form * (*. We shall see in §3.4.1 that this original distribution has been well preserved in ancient Greek, where accented o-stems with columnar penultimate stress of the type olígos, pedíon are rare, and a number of which are due to Greek-specific phonological developments or else demonstrably independent creations. 2.2.3. Collectives The above analysis of barytone thematic nouns as containing underlyingly unaccented stems is supported by the evidence of a derivational category reconstructible for 49
PIE, the collective. Most scholars now agree that alongside the familiar plural in *-ōs (< **-o-es; replaced by pronominal *-oy in many IE branches) to non-neuter o-stems in *-os, PIE possessed a derived collective in *-eh2 (cf. especially Eichner 1985) to o-stems, both non-neuter and neuter. Thus non-neuter (usually masculine) o-stem nouns may contrast a collective in *-eh2 with an ordinary "count" plural. This contrast is especially well preserved in Anatolian, e.g. Hitt. alpeS 'clouds, die Wolke' vs. coll. alpa 'cloud-bank, mass of clouds, das Gewolk' to sg. alpaS 'cloud' < PIE *albhó- 'white', but traces survive elsewhere as well, e.g. in Gr. Horn, kúkloi 'wheels' vs. kúkla 'rings on a shield' (see below). In most IE languages, this category has been restricted to neuter o-stems, to which it has become an ordinary nom./acc. pi. The non-plural origin of these plurals in *-eh2, however, survives in the synchronically peculiar agreement rule by which neuter plural subjects in Hittite, ancient Greek, and Gatha Avestan take 3sg. verbal endings, e.g. Gr. pánta ret 'all things flow'. It is no surprise, then, that these collectives could develop into singular nouns as well, and in fact they are believed to be the source of suffixed nouns in *eh2- or " e / i 2 - s t e m s " ' t n e source of the familiar d-stems of the classical IE languages. In all non-Anatolian IE branches, the suffix *-eh2- has become the exponent of feminine gender to thematic masc. *-os, neut. *-om: as a result, most easterns are feminine in agreement and take adjectives in *-eh2, whence the familiar pattern of Skt. návas, návā, návam, Gr. néos, néa:, néon, Lat. novus, nova, novum.13 One frequently finds an accentual contrast between thematic nouns in non-neuter *os or neuter *-om and their corresponding derived collectives. In particular, PIE stemstressed or barytone o-stems appear to have formed oxytone collectives with stressed suffix *-éh2: •
1. The accentual contrast between Gr. órros 'ass' < PIE *órsos (Engl, arse) and o:rá: 'tail, rear end' < *orséh2, of which the meaning of the latter fits well with an originally collective sense.34
•
2. Gr. kúklos 'circle, ring', TB kokále 'chariot' < PT *kwákwlé, OE hwēol 'wheel' < PGmc. *χwéχwIaz, all continuing PIE *kwékwlos, vs. Horn, kúkla 'rings on a shield' (with productive recessive stress in neuter nouns, cf. §3.6), Ved. pi. cakrā (whence
■^This pattern also occurs in Lycian, at least in an incipient fashion, but it is probably an independent development. On the evolution of ¿/i2-stems in Anatolian in general and in Lycian in particular, see Melchert 1994c and Hajnal 1994b, respectively. 34 Note also the differing treatment of Proto-Greek *-rs- in pretonic vs. posttonic position.
50
backformed sg. cakrám), OE hweowol, hweogol 'wheel' < PGmc. *χwéywlaz <— PIE *kw(e)kwléh2. •
3. The mobile accentual paradigm of Slavic, which comprises underlyingly unaccented noun stems (AP c; cf. §§4.2.3, 5.1), alternates between initial-stressed sg. and endingstressed nom./acc. pi., e.g. Russ. pole 'field', pi. poljá, SCpolje /polje/, pi. pdlja /polj'a/. Although the neuter has not been preserved in East Baltic, OPr. pi. tant. mensa vs. Russian mjáso, SC méso 'meat' < PS1. *m'|so (AP c) indicates that this alternation isofPBSl. date.
•
4. Gr. nomina actionis (action nouns) with o-grade of the root are generally barytone when ending in -os,and oxytone when ending in -a:/-ε:, e.g. phóros 'payment, tribute' vs. phorá: '(act of) carrying, freight, load' (to the root pher- of phéro: 'I bear'). 35 Given that feminine e/^-stem nouns are likely to go back to an original collective formation in *-eh2, this contrast between R(ó)-os and *R(o)-éh2 fits the pattern of 1-3 above.
•
5. The same pattern of root-accent in sg. and ending-accent in collective has been posited for the PIE suffixes *-mo- (forming action nouns, e.g. *h2óg-mos > Gr. ógmos 'furrow' to *\\2é£- 'drive'), *-tro- (forming "instrumental" nouns, e.g. *h2erh3tro-m 'plow' > Gr. áratron, Olr. arathar, Arm. arawr, —> Lat. arātrum, to *h2erh3'plow'), and *-no-: *R(e)-mó- [< *R(é)-mo-?] ~ *R(0)-méh2, adj. *R(0)-mó- (Hamp 1982); *R(é)-tro- ~ *R(0)-tréh2 (Olsen 1988); *R(é)-no- - *R(o)-neh2 (Barton 1993).
As noted in §2.2.2, this pattern may easily be explained if one assumes that the singulars are underlyingly unaccented, i.e. that both the root and the ending *-os or *-om are unaccented, whereas the collective ending *-(eh2 is accented: hence sg. *ors-os vs. coll. *ors-(eh2. The corresponding grids would then be * (* *
*\
*órs
os
* (* (*)
"ors
éh2
The reflexes of these roots and endings in accentually archaic IE languages such as Russian would then preserve these underlying specifications, e.g. for Russ. pole vs. poljá (see further §5.1.1, end):
j:>
As opposed to nomina agentis (agent nouns), which are oxytone masculine o-stems: cf. p orós 'bearing', tómos 'slice' vs. tomós 'cutting' to *tem- in témno:, Horn, támno: 'cut'. h
51
(* *
pól
*)
*
(* (*)
e
polj
á
Interestingly, several oxytone o-stem nouns appear to exhibit exactly the opposite accentual shift, forming barytone collectives in *-eh2. The evidence for this reverse pattern, although not quite as extensive as that presented above, contains several •
1'. Two relics in Greek: mε:rós 'upper thigh' vs. coll. mi.ra 'thigh-pieces' < *mēms-ro- (cf. Lat. membrum 'limb', Olr. mir 'portion'); astéir 'star' < *h2St-er (ti2S-ter?) vs. Horn. coll. ástra 'constellation' <— *h2ést-r-eri2 (*h2és-tr-eh2), with -a for expected *-a:, whence backformed sg. ástron.
•
2'. Lith. šimtas 'hundred' alternates between accentual classes 2 and 4, i.e. if one substracts the historical effects of de Saussure's Law (cf. §4.2.3), it follows either a barytone or oxytone paradigm. This could reflect a contrast between the securely reconstructed PIE oxytone sg. *£iņt-óm (Skt. šatám, Gr. he-katón, Goth, hund < PGmc. *hundā) and a collective or plural *kjņt-eh2 or *£émt-eh2 '(many) hundreds'.
•
3'. In his preliminary examination of the varied and controversial evidence for fullgrade *-ro- formations in PIE, Vine (1999a) notes that many of the best examples for *R(e)-ro- occur in collectives, primarily in Greek: ágra 'the hunt, prey' <— PIE *h2ég-reh2 (cf. agrós, Skt. ájra-, Lat. ager 'field'); ákra 'peak, summit, citadel' <— PIE *h2é£-reh2 (cf. adj. ákros 'extreme, outermost, at the tip', OCS ostru 'sharp'; Lat. ācer 'sharp', Skt. ášri'sharp edge, comer'); téfra 'ashes' <— *d h ég wh -reh 2 (cf. Lat. febris 'fever' < *d h eg wh -ri- or secondary for *febro-). The sg. of full-grade nouns in *-ro- can have been backformed to such collectives, e.g. Skt. ájra-, Gr. agrós, Lat. ager <— coll. 'fields, countryside, driven lands', Gr. do:ron 'gift' <—do:ra <— *déh3-reri2 'gifts, giftage', Lat. adj. minis 'wonderful' (secondary to neut. mirum 'wonder' < *sméy-rert2?). Based on this pattern, Vine tentatively concludes that full-grade PIE *R(e)-ro- was "either fundamentally collect. R(é)-reh2, or with a tendency to develop in that way" and that there is "relatively little evidence for synchronic R(e)-ro- adjs.; some such [were] perh. backformed to other types of formations."
How is this opposing accent shift to be explained? The ending-stress of the sg. indicates that the nominal stems in question are underlyingly postaccenting (cf. §2.2.2). But then the 52
collective in *-eh2 should also be oxytone: under the assumption that a bracket enclosing no syllable heads, i.e. the first of two adjacent left brackets in the metrical grid, is erased (Halle 1997:282), the coll. to e.g. *mēms-ró- should be "*mēms-réh2": * *(
*
(* *)
*mēms rós
*(
(* (*)
"*mems réh2"
We must therefore seek some other explanation for initial stress in collectives to postaccenting o-stems. I tentatively propose that stray erasure did not occur in this case in PEE, but rather that a "clash" of two adjacent left brackets triggered retraction in order to "resolve" the clash: Bracket Clash Resolution * ( ( — > ( * (
Thus when the accented coll suffix *-eh2 is attached to a postaccenting root, retraction occurs, and the form surfaces with initial stress. The metrical grid for *mems-reh2 then becomes *
*mēms reh2 —>
*
*mēms réri2
If this proposal is correct, it follows that collectives to postaccenting o-stems did not have default initial stress, i.e. were not "enclinomena", but rather were underlyingly accented on the final syllable of the root as a result of leftward bracket retraction; the initial stress of these collectives. This pattern immediately calls to mind the alternation of postaccenting o-stem neuter nouns in Slavic, i.e. nouns of AP b, between columnar ending-stress in the sg. and stem-final stress in the pi.: cf. Russ. sel-ó 'village', polotn-ó 'linen', pi. sel-a, polótn-a, SC sélo, vlákno 'fiber', pi. séla, vlákna (ðakavian seló, vlākno, pl. séla, vlákna). More recently, retracted pl. stress has spread to feminine and masculine nouns of AP b in East Slavic and Serbo-Croatian, e.g. Russ. ¿end 'woman', dolgotá 'length', pl. ¿eny, dolgóty, SC kbnac 'end', pl. konci (see §5.1). Halle (1997:282-3) captures this retraction for
53
Russian by proposing a lexically restricted rule of Bracket Retraction exactly analogous to that posited above for PIE. The derivation of Russ. pi. sela, polótna is thus * i*
*( (*) se
la —>
* (*
*
(* (*)
se la
*
*(
(*)
* (*
*
(*)
po lot n a36 —> po lot na
In contrast to feminine and masculine postaccenting nouns, where it is clearly an innovation, retraction appears to be old in o-stem neuter plurals: except for a relatively limited number of derived nouns, e.g. Russ. botestv-ó 'deity, divinity', pi. boiestv-á, plurals to oxytone neuters have stem-final stress in almost all East and South Slavic languages.37 Rather than assume that retraction has spread from the loc. pi. in *-éxú, where it was regular (Stang 1957:82-3; cf. §§4.2.1-2 on "Stang's Law"), I therefore tentatively conclude that the Slavic phenomenon of stress retraction in the plural, in its original domain of (monosyllabic) neuter o-stems, is inherited from PIE. 2.3. The verb 2.3.1. Athematic stems As in the noun, so also in the verbal system of PIE one may draw a sharp distinction between the archaic athematic conjugation, with numerous alternations of ablaut and stress, and the productive thematic classes, which lack stress or (intraparadigmatic) ablaut alternations. Furthermore, the athematic conjugations may be subdivided into a number of distinct types, parallel to the ablauting nominal classes. Let us consider each of these in turn.
36
Not underlyingly /poIotEn-a/, because then the gen. pi. would be polot(En-0 > "polotén". with retraction to the vocalized "jer" vowel, instead of polóten (the nomVacc. pi. would be polot(En(a > polótna with Jer Deletion and Accent Retraction, cf. §4.2.3). Historically, of course, the stem was indeed /polotEn-/ < PS1. *polt-In; cf. oknó 'window', underlyingly /okOn-/, nom./acc. pi. ok(On-(a > ókna, gen. ok(On-(0 > okón, which has preserved the jer-conditioned alternation (§5.1. fn. 2). 37 Cf. Garde 1976a:237-9. On the basis of its absence in Bulgarian, Garde (238) argues that plural retraction is a Slavic-specific innovation in neuter nouns as well. As Bulgarian has heavily restructured the inherited PS1. accentual alternations, however, it is entirely possible that pi. retraction in postaccenting o-stem neuters was lost there. See Alexander 1975:153-66 on this and other sg. - pi. alternations in neuter nouns in the Torlak (southeast Serbian) and neighboring northeast Macedonian and west Bulgarian dialects.
54
2.3.1.1. Root presents and aorists with alternating stress This type is represented by some of the most securely reconstructible preforms in all of IE grammar, including such fundamental present roots as *hies- 'be', *hļey- 'go', *gwhen_ 'strike'. Since Brugmann, these presents have been reconstructed with full-grade of the root and root stress in the sg., and zero-grade of the root and stress on the ending in the du. and pi. This pattern survives intact in Indo-Iranian and Greek, and traces of the root ablaut alternation survive in other branches, particularly for *hies- ~ *hļS- 'be'. lsg. 2 3 ldu. 2 3 lpl. 2 3
*hiés-mi *h|é-si *h|és-ti *h|S-wós *hls-t*hls-t*h|S-mós *his-té *hjs-énti
lsg. 2 3 lpl. 2 3
*hiéy-mi *hiéy-si *hiéy-ti *h|i-mós *h,i-té *hjy-énti
3sg. 3pl.
*gwhén-ti *gwhn-énti
Hitt. ēšmi ēšši ēšzi
(ēšteni) ašanzi
kuenzi kunanzi
Skt. ásmi ási ásti svás stás sthás smás sthá sánti
Gr. (Att.) e.mi ei esti
Lat. {sum)% es est
eston eston e.men este e:si
sumus (estis) sunt
émi ési éti imás ithá yánti
eimi ei(s) eisi imen ite ía:si
OCS jesmī esi jestīa (jesvé) (jesta) (jeste) ijesmv) (jeste) sgtī
OLith. esmi esi ēst(i)
(esmé) (esté)
eimi eisi elti eimé eité
hánti ghnánti
tOld Latin ESOM (Garigliano Bowl, cf. reference)
Other examples include *tlcey- ~ *tk*i- 'dwell', *tlcéy-ti > Skt. kséti (GAv. šaēití), *tlciyénti > Skt. ksiy-ánti, GAv. šiieiņtl, Gr. Myc. ki-ti-je-si /ktijensi/ (Horn, ktízo:); *kwer- ~ *kwr- 'cuts', *kwér-ti, *kwr-énti > Hitt. kuērzi, kuranzi; and *ses- ~ *ses- 'sleep', *sés-ti > Skt. sásti (cf. GAv. lsg. hahml), *ses-énti > Skt. sasánti. For root aorists, the distribution of ablaut and stress is somewhat more complex. In Sanskrit and in Greek, one generally finds full-grade stressed roots not only in the sg., but in the du. and 1 and 2pl. as well, i.e. in all forms outside the 3pl. Cf. the paradigm of *dhehi- ~ *dhh]- 'place' in Sanskrit and Greek:
55
1 2 3
Sanskrit sg. á-dhā-m á-dhā-s á-dhā-t
du. á-dhā-va á-dhā-tam á-dhā-tām
Greek sgé-^εi-n é-flε.-s é-flε:
pi. á-dhā-ma á-dhā-ta á-dh-ur
du. é-t"ε:-ton e-thé:-tε:n
pi. é-t^ε.-men é-fiε.-te é-t"e-san^
Thus the alternation between "strong" and "weak" forms disagrees with the neat division of sg. vs. non-sg. found in the root present. A similar pattern is found in Hittite, where the 1 and 2pl. usually occur with full-grade of the root in the preterite, and often also in the present indicative and imperative: pres. 2pl. ēš-teni 'ye are', pret. lpl. ēš-wen, 2 -ten, iptv. 2pl. ēš-ten; pret. lpl. kuewen (< *kuen-wen) 'we killed', 2 kuen-ten, iptv. 2pl. kuen-ten. Nevertheless, zero-grade roots and ending-stress are occasionally found in Vedic to roots of the shape TeRT, especially in the imperative (see below), e.g. 2du. kŗtám 'do!', gatám 'come!', šrutám 'hear!', 2pl. kŗta, gata, šruta; cf. also Gr. Horn. 3du. bátεin 'the two of them came', with zero-grade of the aor. root bε:- ~ ba-. As there is no reason why the ablaut of the root aorist should originally have differed from that of the athematic present, I assume that the du. and pi. of the root aorist, like that of the root present, had stress on the ending and zero-grade of the root in PEE, and that the Vedic imperatives just cited are relics of this original distribution: cf. Cardona 1960:10, Tedesco 1968:8. Contra Rix (1976:214) and Pinault (1994:195), the extention of full-grade root ablaut and root stress to the du. and 1,2pl. could be an independent parallel innovation of Indo-Iranian and Greek. The accentual analysis of root presents and aorists is exactly equivalent to that of alternating root nouns (§2.2.1.1): the root and sg. ("strong") endings are underlyingly unaccented, and the du. and pi. ("weak") endings are accented. Thus the unaccented sg. forms, being composed of two unaccented morphemes, surface with default initial stress, whereas the stress in the du. and pi. falls on the accented ending. Below I have provided grids for the lsg. and 3pl. of the root present *gwnén- ~ *gwnn- 'strike' and the root aorist *léykw- ~ *likw- 'leave (behind)': * *
* *\
* /*
wh
*g én m i
*gwhen
38
ent
*\
¡ > *gwhn_énti
The Skt. and Gr. 3pl. forms have replaced *á-dh-an, *e-the-n < PIE *é-d h h r ent, with the perfect ending -ur < *-ŗs and the productive Attic-Ionic 3pl. ending -san from the sigmatic aorist, respectively.
56
* (* *
(* *)
*
*leykw W
r*\ w
*leyk
ent > *likw-ént
Both root presents and root aorists have tended to disappear over time, as a consequence of thematization: see §2.3.2.1 for details. The same division into unaccented and accented endings applies to the imperative. In PIE, the imperative of athematic verbs is formed with a special set of endings: sg. 2 *dhi, 3 *-tu, 2 *-te, 3 *-entu. Reflexes of these endings have been preserved especially well in Indo-Iranian and Greek, and traces of 2sg. *-dhi are attested in Anatolian, Tocharian, and probably also Slavic. For pres. *hjes- and aor. *gwem-, the following forms are attested: 2sg.
*his-d h i
>
3 2du. 3 2pl. 3
*hiés-tu
>
*hts-té *hļS-éntu
> >
*gwiņ-dhí *gwém-tu
> >
*gwiņ-té *gwm-éntu
> >
2sg. 3 2du. 3 2pl. 3
GAv. zdī, Skt. edhí(<*sa-dhí, with analogical full-grade), Gr. isfli (with unexplained initial i for expected *ésthi) Hitt. ēšdu, Skt. ástu (cf. Gr. ésto:) Skt. stám, Gr. éston Skt. stem, Gr. éstε.n Skt. síá(na), Gr. éste (cf. Hitt. ēšten with full-grade) Hitt. ašantu, Skt. sántu (cf. Gr. éstoirí) Skt. gadhi, gahi, GAv. gaidī Skt. gántu (for *jántu), GAv. JantQ gatám (gantám) gantām Skt. gata (gánta), Skt. gámantu
Here 2sg. *-dhí, as well as both of the pi. endings, are underlyingly accented and trigger zero-grade of the root; the 3sg., on the other hand, falls into line with the present sg., with unaccented ending and stressed full-grade root.39 Let us turn to the mediopassive. The discovery and proper evaluation of Anatolian and Tocharian have led to a refined reconstruction of the PIE mediopassive endings, which closely resemble those of the perfect. In contrast to the active, all of these are accented, sg. as well as du./pl.: the stress falls on the ending, and the root appears in zero-grade, as in the aorist mediopassives of Skt. kar- 'do, make' (pres. 3sg. kŗ-ņó-ti) and Gr. thε:- 'put, place' (pres. 1 sg. tí-^lε:-mi): 39
If the type of Lat. 2sg. I 'go!', ad-es 'be there! be present!' with stressed full-grade root, the ending *-0 would pattern with iptv. 3sg. *-tu and the other athematic sg. endings.
57
lsg. 2 3
lpl. 2 3
Skt. *k ŗ-h2é —> á-kr-i™ *kwŗ-th2é —> á-kŗ-thās *kwr-ó —> *-tó- > á-kŗ-ta á-kŗ-mahi *k w ŗ- me ( s )d hh 2 w á-kŗ-dhvam *k ŗ-(s)dh2we *kwŗ-ró —> *-ņtó > á-kŗ-ata41
Gr. *d hi-h2é —> e-thé-mε:n *dhhi-th2é —> *-só > é-the-o *d h hi-ó—>*-tó > é-the-to *d h hi-me(s)d h h 2 e-thé-metha *dhhi-(s)dh2we é-the-sthe h *d hi-ró—> *ņtó > é-fie-nto h
w
Thus the traditional division of "strong" vs. "weak" forms comes down to act. sg. vs. act. pi., mediopassive, and iptv. 2sg. *-dhi; the latter are accented, whereas act. sg. *-m(i), *s(i), *-t(i), iptv. *-tu are unaccented. 2.3.1.2. Narten presents The alternations described above apply to the vast majority of classically reconstructed root presents and aorists. Recent research, however, has brought to light a small, residual class of presents marked by stress on the root in all person-number forms. In a classic article of 1968, Narten posited the existence of "proterokinetic" presents on the basis of paradigms such as pres. act. lsg. Ved. stáu-mi, GAv. stāu-mī 'praise, sing of, 3sg. stáu-ti (RV inj. (a-)stau-t), ptcp. GAv. stauu-at-, mp. 3sg. RV stáv-e, Av. stao-ta, Gr. steu-tai (inj. steu-to), lpl. stao-maide, ptcp. TV stáv-āna-, Av. stauu-ana-, which reflect PIE act. sg. *stēw- ~ du./pl. *stéw-, ptcp. *stéw-ņt-, mp. *stéw-, ptcp. *stéwiņhjno-. Like the aerostatic root nouns of the second type discussed in §2.2.1.2, these roots always carry surface stress, and occur in lengthened-grade in the sg. act. and in fullgrade in the rest of the act. and the mp. In addition to 'praise', aerostatic presents may be securely reconstructed for the following roots:42 •
*h!ēd- ~ *hiéd- 'eat': pres. 3sg. Lat. est, OCS jastījéstī < *hļēd-ti (*[h]ēts ti]), OLith. pres. lsg. em/, 3sg. est(i) (mod. edu, éda); Skt. pres. 3pl. ad-ánti (with analogical stress on the ending) <— *ád-ati < *h|éd-ņti (cf. Narten 1968:15fn.44); cf. 40
On the origin of the ending -/, see Cowgill 1968. The older endings 3sg. *-o, 3pl. *-ro tend to survive in stative usage, e.g. Ved. save, šere (Av. sōire), Av. áijhāire 'they are sitting/seated', Skt. a-šay-a-t, a-še-ra-n 's/he, they were lying', ásthi-ra-n 'they were standing' (with old endings -a, -ra < *-o, *-ro plus secondary -/, -n); cf. Narten 1968:9-10, Jasanoff 1994:150-2. On the prehistory of the middle and perfect in general, see Jasanoff 1994, 1997b. The aberrant stress and ablaut of 'be lying' and 'sit' is discussed in §2.3.1.2 below. 42 I omit Ved. tāsti 'makes, builds', 3pl. táksati, ptcp. táksat-, Av. tSSti 'forms, shapes' (inj. tSSt), as it is likely that the 3pl. and ptcp. continue a reduplicated pres. *tekts-ņti, *telcts-ņt- (or sim.) < PIE *te-t¿-ņti, *te-uc-ņt- to root aor. *ték- - *tek-' (cf. Gr. tikto: 'beget' < *ti-tk-e/0, thematized aor. étek-on); this was reinterpreted as a Narten present with stressed full-grade root *tekts-, and the sg. with lengthened-grade was rebuilt on the model of the other aerostatic presents. 41
58
•
• • • •
Hitt. lpl. edwāni (OHitt. in Neo-Hitt. script), 2pl. ēzzatteni, etc. for expected ēdwani*. ēzzattani* < *hiéd-weni, *hiéd-teni (with *e > *a in posttonic open syllable in -wani, -tani; Melchert 1994a: 137-8) *h)ēh 2 g wh - ~ *hiéh 2 g wh - 'drink': Hitt. pres. lpl. akuwani < ♦hjér^g^-weni (? cf. MidHitt. ekuwani, influenced by edwani), TB yok-, TA yok- < PT *yokw- < *yekw- ~ *ok w - < *ēg wh - ~ *āg wh -; cf. Gr. Dor. impf. 3sg. naphe to nzpho: 'am drunk' <— pres. ptcp. népho:n <—«-stem *ņ-hļéh2gwh-ōn, gen. *ņ-hih2gwh-n-és or sim. (Weiss 1994:93ff.; for reconstruction and discussion, see R. Kim 2000b) *kēHs- ~ *k"éHs- 'instruct': Skt. act. 3sg. šās-ti* (subj. RV šās-a-ti), 3pl. šās-ati, ptcp. šās-at- (RV), mp. 3pl. šās-ate (RV), ptcp. šās-āna-, GAv. sās-íī, iptv. 3sg. sāstu *dēHJc- - *déH¿- 'honor': RV act. 3sg. dās-ti, ptcp. dāš-at*wēlc- ~ *wélc- > PA *wēg- ~ *wég- > Hitt. wek- ~ wekk- 'demand, ask for', pret. lsg. wekun, pres. 3pl. wekkanzi (Eichner 1972:78, cf. Melchert 1994a:68) *wēh 2 - ~ *wéh2- > PA *wéh- ~ *wáH- > Hitt. weh- - wahh- 'turn', pres. 3sg. wehzi, lpl. wahhweni
As the list indicates, aerostatic forms with lengthened-grade of the root in the act. sg. or full-grade in the remainder of the paradigm are especially well preserved in Indo-Iranian and Hittite, but relics survive in other languages as well, particularly for 'eat'. In addition, a number of mediopassives reconstructed for PIE exhibit constant stress on the root, which appears in zero-grade. The best known example of this type is the medium tantum verb for 'be lying (down)', PIE pres. 3sg. *k"ey-o-r, 3pl. *key-r-o-r, reflected in CLuv. zīyari, ziyar, Luv. sijēni (Melchert 1987a: 195-6, 1989:26-7, 1992:1925) and, with substitution of active primary *-i (*[-y]) for *-r, *lcey-o-y, *lcey-r-oy > Ved. šáye, šére (GAv. so:ire/sae:re);the byforms *lεey-t-o-r, *£ey-ņt-or, with active 3sg. *-t- and 3pl. *-nt- for archaic mp. *-0-, *-r-, are continued by OHitt. kitta,43 Pal. kltar, Lye. sitēni and (once again with active *-y for *-r) Skt. šéte, Av. saēte, Horn, keītai, kéatai. Other examples include PIE *wés-tor 'have on, wear' > Hitt. wašīari, Ved. váste, GAv. vastē 's/he clothes', cf. ptcp. Ved. vásāna-, Av. vaņhāna-; PIE *éh]S-or, *ehjS-ror 'sit, be seated' —> Ved. āste, āsate, Av. āste, áijhāire (preserved in stative meaning), áijhaņīe, Gr. ε:stai, ε:atai; cf. Hitt. 3sg. ēša 'sits (down)'; PIE *ewgh-tor, *-ņtor 'speak' > Ved. 3pl. óhate 'praise, exalt', Av. aoxte, aojaite (inj. 3sg. GAv. aoγadā, YAv. aoxta).
4j
On the loss of word-final *-r already in Proto-Anatolian and the origin of the ending -ri in Hittite and Luvian mediopassives, cf. Melchert 1988, Yoshida 1990:103ff.
59
Although the only Narten present which occurs in both act. and mp. is *stēw- ~ *stéw(Ved. stáu-mi, stáv-e, etc.), such aberrant full-grade mediopassives are probably to be placed in the same class as the root-stressed active forms above. The reconstruction of columnar root-stressed verbal paradigms for PIE is now generally accepted (cf. Beekes 1995:244, Hoffmann and Forssman 1996:199-205); such paradigms are now regularly referred to as "Narten presents" or "Narten aorists". However, debate remains as to whether the aerostatic class is characteristic of particular stems or constitutes a morphological category by itself, equivalent to e.g. root presents or sigmatic aorists. Kiimmel (1999) argues that a number of stems exhibit a Narten present alongside an "ordinary" root aorist in PIE, i.e. one and the same verbal root may form both a columnar root-stressed present and an ablaut- and stress-alternating aorist, which would argue for the latter hypothesis. From the perspective of metrical bracket theory, the analysis of columnar rootstressed Narten paradigms immediately presents itself: the root is underlyingly stressed, and all the forms made to that root surface with initial root stress (as opposed to default initial stress in unaccented forms, cf. above). The metrical grids for pres. 3sg. *hiēd-ti (= *[hiétsti]), 3pl. *h]éd-ņti and mp. 3sg. *lcéy-o-r, *wés-tor are therefore * /*
* r*
*hiéd t i > *hiēd-ti
*hjed ent i > *hjéd-ņti
*
*
*
*£éy o r
*wés to r
Just as aerostatic nominal roots appear to have been a relic class already in PIE (cf. §2.1.1, 2), accented verbal roots exhibit a tendency to disappear or be replaced by unaccented roots in stress-alternating paradigms. To take the example of 'praise', whose aerostatic forms were given above, Indo-Iranian has already introduced zero-grade into the non-sg. forms of the pres. act. paradigm, e.g. Vedic 3pl. stuv-ánti (vs. ptcp. stáv-āna-, mp. stáv-e). Avestan has taken the unsurprising further step of altering the lengthened-grade of GAv. stāu-mī to YAv. staomi, staoiti, inj. staot, thus producing a regular ablauting paradigm in the active. It is thus not unthinkable that columnar root-stressed presents could have come to be perceived as a separate inflectional type in post-PIE, at least in isolated cases, whence 60
the coexistence of Ved. pres. 3sg. krāmati 'strides, steps' (thematized from *krámiti < *k(w>Rē'mH-; mp. krámate) and root aor. á-kramīat < *k<w)RémH-t,44 or pres. dās-ti 'honors, offers, awaits', ptcp. dāš-at- < *dēlc- (generalized long-grade; or *dēHk- ~ *déHk*-? see the discussion in Narten 1968:14fn.33, 15fn.43) and Gr. root aor. 3sg. dék-to 'received' < *delc- (cf. §6.6, fn. 38). In view of the uncertainty of these and Kummel's other examples, however,451 maintain the hypothesis that aerostatic Narten inflection in the earliest reconstructible PIE verbal system was a property of particular roots, namely those which were lexically marked with underlying accent. 2.3.1.3. Hysterokinetic alternations: the optative and nasal-infixed presents Alongside the indicative and imperative, root presents and aorists, like other presents and aorists, also form a subjunctive and optative in Indo-Iranian and Greek. The antiquity of the subjunctive has been challenged in recent years: note that Anatolian possesses not a trace of the thematic vowel *-e/0-, the subjunctive suffix underlying Skt., Av. -a- and Greek - e / 0 - (cf. also Lat. fut. era 'will be' < subj. *hļes-e/0- and OscoUmbrian s-futures from s-aorist subjunctives; on the latter in Old Irish, see McCone 1991:55-83), and that Tocharian attests only two likely examples of subjunctives to PIE athematic roots. 46 Whatever the prehistory of this suffix, and of the subjunctive as a category, it is likely that the classical thematic sufix did not exist in PIE, and that its appearance in the "Brugmannian" IE languages is in some way connected to the origin of the simple thematic present (cf. Schmidt 1992:108-11, Jasanoff 1997b, Ringe 2000:13 Iff.). On the other hand, the optative in *-yeh]- - *-ihj- is well attested in Tocharian as well as the classical languages. The original paradigm has best been preserved with *h les'be':
44 Indo-Iranian only, cf. Sogd. impf. γr'm [yram] 'come!', Persian xirdm < PIr. *xram(LIV:328-9). 45 Particularly *hied- "eat' and *tetk- 'build, construct', for which the evidence for a root aorist is unconvincing (on *tetk- see fn. 42 above), and *welhi- 'want', which probably was originally an aoristic root. 46 Namely TA 3sg. šmās, TB šam-n (verse, prevocalic) 'will come' < PT *samya(-s3) < PIE w *g ém-e-ti (GAv. jamaitī, Jimaitl, fimat, 3pl. Jiman, Ved. gámat, 3pl. gámanti, gáman [Hoffmann 1955]; pres. Goth, qimiþ, OHG quimit < PGmc. *kwimiþi) to aor. *gwém - *gwm- (§2.3.1.1) and TB kantār 'will happen' < PT *ka natsr <— PIE aor. subj. *génh,-e-tor to aor. *g'enh r - *g'ņh r , pres. *gņh]-yé/ó(Hackstein 1995:239-42; §2.3.2.2a). For other possible examples of Toch. cl. II subjs. < PIE aor. subjs cf. Ringe 2000:132-6.
61
lsg. 2 3 lpl. 2 3
Skt. syām syās syāt syāma syāta syúr
GAv. xiiām xiia
xiiāta xiiāma xiiātā YAv. hiiāra
Gr. eíε:n eíε:s eíε: éimen éite éīen
Lat. sim sīs sit (OLat. sied) sīmus sītis sint
Based on these forms, one may confidently reconstruct an alternation between sg. *hjsyéhi- and (du./)pl. *his-ihi-'; that Tocharian likewise inherited these variants is shown by impf. 3sg. TB sey ~ sai (TA se-s) < PT *se- + -(s)y- < *his-yehi(-t) + productive zerograde optative suffix *-ihj-, cf. Adams 1988:98, Pinault 1989:128-9). Although this mood, like the subjunctive, finds no cognates in Hittite or the other Anatolian languages, the alternation of ablaut grade and stress reflected in Indo-Iranian and Greek — and indirectly for 'be' in Tocharian — makes it probable that this suffix and category are of PIE date. The analysis of this alternation, however, presents some difficulties. If the root is underlyingly accented, i.e. forms a "Narten" present or aorist (§2.3.1.2), the opt. exhibits columnar root-stress with zero-grade of the unstressed suffix *-ihj-, e.g. *hjéd-ihi-t > Lat. edit, *wél-ih]-t > Lat. velit, Goth. wili. For unaccented roots such as *hies-, one must assume either that the root was for some reason unstressable or that recessive stress in the unaccented sg. forms could not retract to the left of the optative suffix: hence 3sg. *h|Syeh]-t instead of "*hiés-ihj-t".47 I can offer no satisfactory explanation for this peculiarity at the present time. Alongside the athematic root optative, another verbal category appears to be characterized by the same stress alternation between suffix in the sg. and ending in the du./pl., namely nasal-infixed presents of the type of Ved. 3sg. yu-ná-k-ti, 3pl. yu-h-j-ánti < PIE *yu-né-g-ti, *yu-n-g-énti. Here, however, the stress alternation may be non-peripheral on the surface only: in light of recent studies of infixation within the framework of Prosodic Morphology (cf. McCarthy and Prince 1993, 1995), one may analyze the nasal infix as a morpheme attached to the left edge of the word — in other words, underlyingly prefixed — which undergoes prosodic alignment at the right of the nucleus of the root 47
The most archaic sigmatic aorist optatives in Greek, e.g. Cretan wérksien 'may they carry out', are of no help for historical reconstruction: since (classical) PIE formed root optatives to sigmatic aorists — a situation preserved in Indo-Iranian, cf. GAv. opt. lpl. taenia 'we might win' to inj. zāh-* < * g h ē h r s - (Ved. has-; §6.6, fn. 33) — these forms must have been remodeled with the -sformant of the indicative and subjunctive, like later Av. opt. 3sg. fra-za-h-lc.
62
syllable, i.e. is metathesized with the CV to its right. In that case, the underlying metrical grids for PEE *yu-né-g-ti, *yu-n-g-énti would be * (* *
ne
*
*)
yewg ti
>
né-yug-ti
>
PIE *yu-né-g-ti
>
n-yug-énti
>
PEE *yu-n-g-énti
* *
ne
*
(* *)
yewg enti
The same derivations will produce the Vedic forms, except that the ablaut alternation in the nasal infix, which was phonologically conditioned by the placement of stress in (pre-)PIE, had already become morphologized by (late) PEE (cf. §2.2.1.3). 2.3.2. Thematic stems Even in the oldest attested IE languages, Hittite (and Palaic and Luvian), Vedic Sanskrit, and Homeric Greek, the inherited stratum of athematic verbs coexists alongside a very large number of thematic verbs, marked by the presence of a suffix containing the thematic vowel *-e- or *-o-. Just as nonablauting thematic o- and e/22-stems gradually outnumber the archaic ablauting types and come to predominate in nominal formation and borrowings (cf. §2.2.2.), so thematic inflection eventually prevails in the verbal systems of virtually all surviving IE languages. This stage has already been reached in Classical Latin, where only traces of athematic inflection survive (e.g. sum, velim); similarly for the old Germanic languages or Old Church Slavic, which preserve only a handful of athematic verbs (cf. lsg. Goth, im, OE eom, OCSjestnJ 'I am', OE dam, OCS daml 'I give').48 To be sure, this is unquestionably a long-term process, and in IE languages with relatively archaic verbal systems such as Vedic or Tocharian — to say nothing of Hittite, renowned for its complete lack of simple thematic presents (cf. §6.1) — athematic formations are numerous and play a large role in the verbal morphology. Furthermore, a number of athematic formations retain some productivity down into the immediate prehistories and early histories of Indo-Iranian and Greek: cf. the "Aiolic" (and Arcadian, Cypriot) presents in -ε:m(m)i for Att. -ea:, e.g. kálε:mmi (Sappho) for Att. kaléo: 'call', Arc. 48
0n reflexes of athematic inflection in Germanic, see most recently Ltihr 1990. The accentuation of the Slavic athematic presents is examined in §5.5.
63
poiensi for Ion. poiéo.si, Att. poio.si (Buck 1955:123). Nevertheless, a long-term trend toward the eventual extinction of athematic conjugation is shared by most IE languages: witness the development of Indo-Aryan and Greek from their earliest attested stages to the present day, or the near-complete elimination of mi-verb forms, some twenty of which are found in Old Lithuanian and/or the modern dialects, in the modern (literary) language (cf. Senn 1966:286-97). Thematic verbal stems, like thematic nouns, are generally reconstructed with columnar stress for PIE, e.g. suffixed pres. *gwiņ-yé/ó- 'come', *pŗ(lc)-slcé/ó- 'ask', simple pres. *bhér-e/0- 'carry'. In the case of suffixed thematic presents, columnar stress on the thematic vowel is attested in Sanskrit and indirectly confirmed by Germanic and, of course, zero-grade ablaut of the root (cf. 2 below); similarly, originally athematic presents and aorists which have undergone thematization in post-PIE always exhibit columnar stress, often on the thematic vowel but sometimes on the preceding thematized root. These patterns are confirmed by the Balto-Slavic data, where the verbs in question were uniformly stressed on the thematic vowel or the immediately preceding vowel (cf. §6.6). For simple thematic presents, however, the evidence is conflicting, with Balto-Slavic (particularly Slavic) exhibiting an intraparadigmatic alternation quite unlike any found in Sanskrit. I therefore restrict the following discussion to thematized PIE athematic presents and aorists and suffixed thematic presents; for the accentual reconstruction and analysis of simple thematic presents, cf. §§6.1-2, 6.5. 2.3.2.1. Thematized presents and aorists Although this class is strictly speaking not of PIE date (cf. §2.3.1.1), almost every IE language possesses thematic presents and aorists which are reconstructed as athematic for the protolanguage, either because athematic forms are actually attested for earlier stages of that language or in other languages, or because the presents of different IE languages disagree in their root ablaut or are formed with different suffixes (e.g. *-e/0- vs. *-ye/0-), suggesting that the verb in question has been thematized independently, after the breakup of PIE. In general, the mechanism of thematization is clear: first the 3pl. ending *-énti. *-ént is replaced by *-ónti, *-ónt under the influence of ptcp. *-ónt- (~ *-ņt-, cf. §2.2.1.3). The latter is reanalyzed as *-ó-nt(i), with thematic vowel *-o-, and the rest of the paradigm is then rebuilt with thematic vowel *-e/0-.49 In the classical IE languages, the resulting 4
9Thi s sort of paradigmatic remodeling, by which the vowel of a vowel-initial ending (usually the 3pl.) is reanalyzed as a suffix and spreads throughout the paradigm, has occurred independently on
64
thematized presents join the ranks of simple thematic presents with root or thematic stress, e.g. later Skt. ksiy-á-ti 'dwells', táks-a-ti 'makes, creates' for Ved. ksé-ti, tās-ti (cf. 3pl. ksiy-ánti, táks-anti\ §2.3.1.1-2). In Slavic, as we shall see, athematic presents which were thematized at a sufficiently early date were assimilated to the accentual paradigm of the inherited post-PIE simple thematic presents; cf. §6.6 for details. Likewise, thematic aorists are a post-PIE creation of the individual languages. As Cardona (1960) has demonstrated, only two thematic aorists are attested in more than two separate IE branches, namely TB 2, 3sg. lac /bcá/ 's/he went out', Olr. luid 's/he went' < *hiludh-és, *-ét, Horn. Gr. é:luthe 's/he went' < *é-h]ludh-et (Cowgill, cited in Winter 1962:22) and Gr. eide < *éwide 's/he saw', Ved. ávidat, Arm. egit 's/he found' < *é-widet (cf. Schwyzer 1939:747, Godel 1975:114), and both of these in turn are likely to be parallel thematizations of root aorists, preserved in Lat. pf. vldi 'I saw, have seen' < aor. *wéyd- (~ *wid-'), Skt. ró(d)hati 'grow, ascend', Goth, liudan 'grow, rise, spring up' < aor. subj. *hļléwdh-e/0- to *hiléwdh- ~ *hiludh-' (Bammesberger 1985:71-4; cf. R. Kim 2001b:120fn.4 with refs.) In most IE languages, root aorists have been thematized in the manner described above, with 3pl. *-ónt <— *-ént being reanalyzed as *-ó-nt; thus their reflexes in the daughter languages should be stressed on the thematic vowel. Such is the case in Greek: although virtually all finite verb forms have generalized enclisis (reflected as recessive stress, cf. §3.3.1), the active and middle participles and infinitives, as well as the imperative middle 2sg., preserve the expected thematic stress, e.g. ptcp. act. lip-0:n, mid. lip-ó-menos, inf. act. lip-e:n, mid. lip-e-sfiai, iptv. mid. Horn, lip-éo (> Att. lip-d:) to the root lip- 'leave (behind)'; cf. also the iptv. act. 2sg. forms elth-é 'come!', heur-é 'find!', eip-é 'speak! tell!', id-é 'see!', lab-é 'take!', which exceptionally preserve the thematic stress (vs. regular lip-e 'leave!'). As for Sanskrit, it is generally agreed that at least some of the Class VI presents, marked by zero-grade of the root and stressed thematic vowel (the "rwi/art'-type", named after one of its members), are in origin thematized aorists which have shifted aspect: in addition to tud-á-ti 'pushes' <— PIE aor. *(s)téwd- ~ *(s)tud-' (cf. Ved. nitundate 'pushes him/herself on', Lat. ¡undo 'push, hit' < nasal pres. *(s)tu-né-d- ~
65
*(s)tu-n-d-'), one may cite diš-á-ti 'shows' <— PIE aor. *déyk- ~ *di£-' (Gr. deik-nu-mi; sim. Lat. dlcōl cf. LIV:92-3 and see §6.6, fn. 32).50 2.3.2.2. Suffixed thematic presents: *-yé/<s-, *-sJcé/ó-, *-éy e / 0 All three of these suffixes are well documented, or securely reconstructible for the immediate prehistory of, Indo-Iranian, Greek, and Italic. Moreover, the continuants of *yé/ó- and *-slcé/6- are productive suffixes in Hittite and Luvian, and Hittite attests at least two reflexes of presents in *-eyé/0- (cf. below). In Tocharian, *-sj£é/ó- has become exceptionally productive, serving to form "causatives" (i.e. marked transitives) and, at least at one point, denominatives; probable examples of presents in *-yé/ó- and *-eye/0- may be identified as well. One may therefore reconstruct all three suffixes for "true" PIE, where at least *-yé/ó- and *-slcé/(5- appear to have been productive present formants. 2.3.2.2a. Zero-grade presents in *-yé/óThis, the most widespread of the suffixed thematic presents, finds clear reflexes in virtually every IE branch, including Anatolian, Tocharian, Italic, Celtic, Greek, Germanic, Indo-Iranian, and Balto-Slavic. The following are among the most widely represented (cf. the corresponding LIV entries for details): *his-i(y)é-ti 'throws' (with "Sievers's Law" treatment of *y after a consonant cluster, cf. Mayrhofer 1986:164-7): Hitt. si(y)ēzzi, cf. Ved. ásyati, Av. aijhiieiti with full-grade *hjes- from root aor. *hjés- ~ *his-' (GAv. inj. 2sg. as 'sent', subj. ayhat, Ved. viasan 'throw from one another'); *gņhi-yé-tor 'be born, come into existence' > Ved.jayate, Av. zaiieiti, Olr. rel. 3sg. gainethar 'who is born' (alongside redupl. *ge-génhi- ~ *ge-gņhi-' > Ved. aor. [<— impf.] ájījanat, thematized pres. Av. 3pl. zīzanaņti, Gr. gignomai, Lat. gignō; Lat. nāscor probably backformed to past ptcp. (g)nātus with inchoative -sc-, TB knastār, 3pl. knaskentār < PT *kanassa- ~ *kanaskéwith present-forming *-sk-, cf. §2.3.2.2b), to root aor. *génh]- ~ *gņhr' (Ved. mp. Isg. ajani, Gr. egéneto, them. 3pl. egénonto), subj. *génhi-e/0- (—> pres. Ved.janati, OLat. 3pl. genuntl), *dénhļ-e-tor > TB subj. kantár (cf. Hackstein 1995:239-42) numerous occasions in IE languages; such a morphological change is favored by the greater ease with which the new paradigms, lacking consonant clusters at the boundary of root and ending, may be acquired. Cf. the Greek "alphathematic" sigmatic aorist in -sa-, with -a- generalized from lsg. -s-a < *-s-rņ, 3pl. -s-an < *-s-ņt. 50The progress of thematization in the aorist may be observed from the Rg-Veda through the Brahmanas to the classical language (Cardona 1960:10-51), but since lexical stress was lost by classical Skt., our best evidence for the accentuation of thematized aorists comes from the Class VI presents.
66
*kh2p-i(y)é-ti 'grabs, schnappen' > Gr. kápto: 'get hold of, swallo', Lat. capit, 3pl. capiunt 'take, grab', Goth, hafjan 'raise, lift' (secondarily root-stressed, hence no Verner's Law voicing); *mņ-yé-tor 'thinks (of)': Ved. mányate 'thinks, take for', GAv. 3pl. mainiieņtē 'take for', Gr. mainomai 'rave, rage, am mad', Olr. domainethar 'thinks' (rounded to moin-, -muin-; Thurneysen 1946:129, 354), to root aorist *men- ~ *mn- (GAv. mp. mantā 'thinks of, Skt. ámata 'thought of); *mŗ-yé-tor 'dies' > Lat. morior, Ved. mriyáte, Av. miriiete, Lat. morior, OCS umīrjetú 'will die', to root aor. *mér- ~ *mŗ-' (Ved. ámŗta, OCS u-mrē-tú < PS1. *mert or sim.; Gr. Hesych. émorten • apéfianen 'died' for *émorto?); *nigw-yé-ti 'washes' > Olr. -nig (thematic *nig-e/0- <— *nigye/0- < *nigw-ye/0-, Thurneysen 1946:115, 378), Gr. nizo:, mp. nizomai 'wash, bathe' (cf. Horn. khér-nib-on 'hand-wash basin'), to root aor. *néygw- ~ *nig w -' (Ved. mp. ptcp. -nijāná- 'washed', lsg. anijam); *sh2-i(y)e-ti 'ties, binds': Hitt. 3pl. išhiyanzi (whence Ai-conj. 3sg. išhāi), CLuv. hišhiyanti, Ved. syáti, Av. subj. 3pl. hiiqn 'will bind', to root aor. *séh2- ~ *sh2-' (Ved. inj. sāt); *swid-yé-ti 'sweats': TB /ssya-/ in verbal noun syālhe 'sweating', Skt. svidyati, OHG swizzen; *wŗg-yé-ti 'works': Av. varaziieiti 'does, performs', Gr. (w)érdo: 'fiihre aus, opfere' (< *wérzde/0- < *wergye/0- <— *wŗg-ye/0-, cf. Myc. wo-ze /worzei/ 'fiihrt aus'), Goth, waúrkjan, OE wyrcan, to root aor. *wérg- ~ *wŗg-' (GAv. opt. lpl. varazimā-ðā [Hoffmann 1968:3-5]; sigm. aor. inj. vāraš*, subj. varasaitK *wérd-se-ti for root aor. *vareš, *varazaití? cf. §6.6, fn. 33). In most of these verbs, *-y e / 0 - performs the function of a present-forming suffix to an "aoristic" root, e.g. pres. 3sg. pres. mp. 3sg. *dņhj-yé-tor, root aor. *génhi-tor. This continues an inherited pattern, probably already in place in PIE, by which "presentic" (atelic or imperfective) roots formed derived, e.g. sigmatic aorists, whereas conversely "aoristic" (telic or perfective) roots formed root aorists and derived presents; cf. §6.6 for the continuation of this system in Slavic. Melchert (1997b) has recently argued that the suffix *ye/0- may also have marked imperfective aspect in early Anatolian, on the basis of Hitt. and CLuv. verbs which appear to take *-y e / 0 - in the present indicative only, in particular Hitt. karp(i)ye-, kar(a)p- 'lift'.51 In addition, *-ye/0- served to form denominative verbs to nouns of various stem classes in PIE. Tocharian class XII presents, for instance, contain the suffix PT *-nn3/e- < *-n-ye/0-, (with regular development of sonorant + *y to palatalized geminate; cf. Adams -"It has been generally assumed that Anatolian did not inherit the classical PIE distinction of imperfective vs. perfective vs. stative aspect (i.e. present vs. aorist vs. perfect), and that the suffixes *ye/0- and *ské/ó- carried only imperfective, i.e. iterative, durative, etc. Aktionsart at the time of the separation of Anatolian from the rest of PIE: cf. Strunk 1994.
67
1988:35,44fn.4; Ringe 1996:116-7), abstracted and generalized from denominatives to nstems: cf. TB subj. /wen3/e-/ 'say', lsg. wenau <— FT *wenn3/e- < *wek-nn3/e- to *weka (TB wek) < *wokw-iņ 'voice' (Winter 1977:133-4), /tankwa-nna/e-/ 'love', pres. act. ptcp. tanwannenca to tanw, tānkw- 'love'. Denominative presents in *-e-ye/0- to thematic nouns are continued in Indo-Iranian, e.g. Skt. ŗta-yá- (ŗtā-yá-), Av. ašaiia- 'act truthfully, righteously' to Skt. ŗtá-m, Av. aša- (< *árta-52) 'truth, righteousness', and as the large and productive class of Greek verbs in -eo:, e.g. philéo: 'love' < *phiIe-je/0- <—philos 'dear'; furthermore, Hittite offers likely examples in duwarni/a- 'break' < *dhwerne-yé/ó- <— *dhwerno- (cf. Luv. l(a)warr(i)-; Melchert 1994a:270) and iter. ušnešk?/a- 'offer for sale' (pret. 3sg. ušneškattá) to ušne/a-* < *usné-ye/0- <— *usnó- (Melchert 1984:36-7; cf. independently created Ved. vasnayáti to vasnám 'price', Gr. o.néomai < *wosne-ye/0- to o.nos 'purchase price' < *wosno-; Lat. vēnum dare 'offer for sale' < *wesno-). Although a large number of -ya- presents in Sanskrit are stressed on the stem, the consistent zero-grade of the root in the verbs listed above indicates that the suffix *-ye/0was accented in PIE. Where the root was unaccented, the suffix received the surface stress, which has been preserved in e.g. Hitt. siēzzi and Skt. mriyáte. The accentual derivation of *shļ-i(y)é-t-i and *mŗ-yé-t-or is hence *
*seh] ye t i
*
>
*shii(y)éti
*mer ye t or >
*mŗyétor
In denominatives to thematic nouns, the expected stress *-e-yé/¿- has largely been altered to *-é-ye/0-, probably under the influence of iterative-transitive o-grade presents in *-éye/0(see §2.3.2.2c below): cf., already in Vedic, mantrá-ya-te 'speaks, takes counsel' <— mántra- 'counsel, advice', arthá-ya-ti, -te 'seeks (as goal, desire)' <— ártha- 'goal, object', Brugmann and Delbriick 1916:248-9, Macdonell 1910:398-9, Whitney 1924:387). Similarly, CLuv. tarmi-ltarmai- (pret. 3sg. tarmita, iptv. 3pl. tannaindu) 'nail down' <— tarma- 'nail, peg' and Luv. kumezidi, 3pl. kumezeiti 'sacrifice, worship' <— kumaza'priest' agree in having lenited 3sg. endings, i.e. CLuv. pret. -ta, Lye. pres. -di. This lenition between two unstressed vowels, which dates back to Proto-Anatolian ("Eichner's second lenition rule"; cf. Eichner 1972:100fn.86, Morpurgo Davies 1982/3:260ff.),
52
On the historical phonology of the Av. noun, see Hoffmann 1986:166-9 (cf. OP (a)rta-
þrtá-]).
68
demonstrates that the immediate preform must have been *-é-ye/0- for PIE *-e-yé/ó(Melchert 1994a:318, 1997:134-6). 2.3.2.2b. Zero-grade presents in *-sl£é/óThis suffix is profusely attested in both Anatolian and Tocharian as well as the classical BE languages, and so may without difficulty be reconstructed for PIE. Among the wide range of functions attested in IE, *-síé/ó- forms presents to subjunctive, i.e. punctual or perfective (aoristic) verbal roots, like *-yé/ó-. Cf. the examples below: *cjņh3-slcé-ti 'recognize' > OP subj. xšnāsātiy (with full-grade), Gr. gi-gno:sko:, Lat. (g)nōscit, Arm. ðanað'em (for *c-), Alb. Isg. njoh 'know' (alongside *gņné-h3-ti —> Skt. jānāti, Av. 3pl. -zānaņti [with full-grade], TA 2sg. knānat, Olr. ad-gnin 'knows', Goth, kunnan 'know', Lith. Zinoti 'know' [Bammesberger 1993]); sigm. aor. *cjnēh3-s- > Hitt. ganēšš-, TA pret. Isg. knasu, 2 knasast (Hackstein 1993, contra Jasanoff 1988a); *gwiņ-slcé/ó- > Skt. gácchati, Av.jasaiti (for *gasaiti) 'go, come', Gr. básko: 'come', alongside Gr. baino:, Lat. veniō< *gwiņ-yé/ó- to *gwem- 'come' (Rix 1976:213-4; for root aor. forms see §2.3.1.1). *h\ļ-séé/ó- > Skt. ŗccháti 'reaches, arrives at, comes to/upon', OP subj. ni-rsātiy 'may come down', Gr. érkhomai 'go, come' (Rix 1969:98-100), alongside Ved. ŗņóti 'brings, takes (to a place)' < *h^-néw- (Ved. root aor. ārat 'came to, arrived at' < *hļér-);53 *h2is-slcé/ó- > Skt. iccháti, Av. isaiti, OLith. Isg. ieszku (modern ieškóti, íeško), PS1. *jisk-a-ti, *jisk-je-ti (OCS iskati, ištetī; cf. Ukr. s'káty) 'seeks, desires' (cf. GAv. sigm. aor. inj. āiš 'wishes' < *h2ēys-s-t); **milc-slcé-ti 'mixes' > *mislcéti > Ved. pres. mp. ptcp. miccha-māna- 'uprooted', Gr. misgo: (< *mig-sko: with *-g-), Lat. misceō, -ēre (rebuilt with *-eye/0-), Olr. mesca (rebuilt as cf-pres.), OE miscian (if not borrowed from Lat.; rebuilt as ō-pres.) to root aor. *mey£- ~ *milc-' (Gr. Horn, émikto), cf. Ved. miš-rá'mixed (up)'; **pŗlc-slté/ó- > *pŗslcé/ó- > Skt. pŗccháti, Av. Isg. parasā 'ask', Arm. aor. 3sg. eharc' 'asked', Lat. poscō 'demand' (< *porsk-; cf. prec-ēs 'prayers' < *prelc-), Olr. Isg. arco 'ask for' (cf. OHG denom. forscōn 'ask, inquire' <—forsca 'question, inquiry' < post-PDB deverbal *p^slc-éh2) alongside Lith. praSyti (3sg. prāšo) 'ask, request', OCS prositi, 3sg. prosit! 'ask, demand' <— *prolc-
-^And perhaps Hitt. arlkizzi 'arrive repeatedly, invade' vs. arnuzi 'move (tr.), transport' (LIV:211-2, LIV2:238 s.v. *h¡er, on arnuzi see §6.4), although an iterative in -Sk- to ārhi 'arrive' could of course have been formed at any time in Hittite. Kimball (1987) and Melchert (1987b) have demonstrated that initial * h r survives in Hittite as h-, so these verbs must continue a root *h,er/*h,or-/*h,ŗ-(ibid., 22fn.l0).
69
éye/o_ (j-aor. Ved. áprāt< *prēk>s-t, GAv. subj. mp. lsg. frašī 'consulted'; cf. TB preksa, TA prakās).5A In addition, *-slcé/ó- is found in the following functions: •
• • • •
•
Hittite "marked imperfectives" in -ike ška-, which may carry descriptive/progressive, iterative, durative, distributive, or inchoative meaning (cf. Melchert 1998:414-6 for a brief survey with examples and references), and the cognate iterative suffixes Luv. -z-, Lye. -s-; Tocharian derived present stems and marked transitives (i.e. transitive-causatives) in Proto-Tocharian *-sss- ~ *-ske-:55 the East Ionic "iterative" imperfect, e.g. Horn. ékhe-skon 'I used to/would have', dóskon 'I used to/would give', strépsa-skon 'they used to/would turn over' (limited to thematic stems and 3rd person in Herodotos; Schwyzer 1939:710-2, Rix 1976:229);56 Latin fientives to -ē- statives of the type alb-ē-scō 'become white' (to albus 'white', alb-eō, alb-ē-re 'be white'; Watkins 1971:66-72); Avestan inchoatives in -sa-, e.g. Svsa- 'put oneself in motion' < *£yu-slcé/ó-, tafsa'become warm' < *tep-s£é/6- (vs. causative tapaiieiti 'make warm, warm (tr.)', cf. §6.1; Hoffmann and Forssman 1996:188), cf. also xvafsa- <— *sup-slcé/ó- vs. Russian CS1. u-svple- 'fall asleep' < *sup-yé/ó- (Tedesco 1948:375-6); Lithuanian intransitives to heavy stems with -sta- < *-sk*é/ó-, e.g. dyg-ti, pres. 3p. dygsta (pret. dyg-o) 'germinate', tifp-ti, tifp-sta 'become torpid' (vs. «-infixed intransitives to stems of the shape CeC, e.g. tik-ti, pres. 3p. ti-n-k-a [pret. tik-o] 'fit^prasti /prat-ti/, pra-h-t-a 'become used to' — in turn replacing earlier *-ye/0-? Cf. Tedesco 1948:349, fn. 10, 384-5; on Slavic intransitives in *-ne/0- from earlier *-ye/0-, see §6.6.)57
54
Note that *h|βs- 'be' also apparently had a *-s¿é/6- present already in PIE, to judge from Palaic iptv. 2sg. ¡Ska 'be!* < *sk"ó< *h,s-séó (cf. Watkins 1993:477-8), TB 3sg. ste, star-, 3pl. skente, skentár- < PT *skate, *-ta'r, *skēnte, *-ár < *ske-tó(r), *-ntó(r) <— *hļS-s¿éto, *-ónto, OLat. escit, Horn. éske. I hope to discuss the semantics and usage of *hļS-siEé/6- vs. *h]es- on another occasion; for opposing views, cf. Hackstein 1995:272ff.274-5, 278-2, Melchert 2000:146. 55p¡ŗ£ *-ské/ó- presents to monosyllabic roots have been reanalyzed as simple thematic presents in Tocharian, e.g. TB /pask-/, TA pask- 'protect' (mp. 3sg. pāstār) < PT *pask- < *ph2-ske/(5(cf. Hitt. pah(has)s- 'protect', OCS pasetl 'watches over, tends', Lat. pastor 'shepherd' < *pehi-s-). TB /nask-/, TA nāsk- 'bathe' < PT *nask- <— *nh2-sk'é/ó- (cf. Ved. sndti 'bathes', Lat. nāre 'swim' < *(s)neh 2 -); cf. Ringe 2000:121-2. That *-slcé/ó- must also have served as a denominative suffix at some stage of pre-PT is revealed by TB /waynask-/, TA wināsk- 'honor', cf. TB wlna 'liking, pleasure'. 56 Continued in certain modern Greek dialects as -iskon, -iska: cf. Schwyzer 1939:712 with refs. On Greek -ske/„- presents in general, see now Zerdin 2000. 57 0 n the distribution and formation of the Lith. «-infixed and íía-intransitives, cf. Stang (1966:338-49), who however does not recognize the derivation of the latter from *-sk'é/ó-. The suffix -c'- of the classical Arm. pres. and aor. subj. in -ic'-, e.g. sg. 1 sir-ic'-em, 2 -es, 3 -é to pres. sirem 'I love', ber-ic', ber-c'-es, -é (< *ber-ic'-es, *-é) to root aor. e-ber 's/he carried', sirec'-ic', sires-c'es, -é (< *sireac'-ic'-em, *-es, *-e) to weak aor. sireac' 's/he loved' has been connected in various ways with PIE *-ské/ó-; see Schmitt 1977:142-3, 145 for references.
70
It is likely that we are dealing here with a suffix which was originally associated with imperfectivity, whence Hitt. inchoatives derived from statives (e.g. šešk*/a- 'fall asleep' to šeš- 'sleep'), Latin fientives in -ēscō to ē-statives, and East Ionic ste^-imperfects. Such an imperfectivizing suffix, reinterpreted as a true aspectual marker of the familiar type, could have begun to be employed already in classical PIE to form derived present stems to aoristic roots, as in Skt. gácchati, Gr. básko: and the other examples given above.58 Whatever the initial function of *-sEé/ó-, it is clear from the derived presents to 'come', 'reach', 'seek', and 'ask', as well as the Avestan inchoatives švsa-, tafsa-, that the root appears in zero-grade. We may therefore reconstruct underlying accentuation for this suffix, with an unstressed root receiving zero-grade.
*
* (* /*
*hjer sice
*)
ti > *hiŗ-slcé-ti
The evidence of Skt. ŗccháti, iccháti, pŗccháti confirms this hypothesis; gácchati for expected *gaccháti has probably acquired its root stress from simple thematic presents with root vowel a. 2.3.2.2c. o-grade presents in *-éye/0As is well known, PIE o-grade presents containing the suffix *-éye/0- are continued in Indo-Iranian by Sanskrit class X presents and Avestan presents in -aiieiti < *-ayati, e.g. raoCaiieiti 'kindles', tapaiieiti 'warms (tr.)'; in Germanic by a number of weak class I presents with old o-grade, e.g. Goth, wasjan 'clothe' < *wo-seye/0-, lagjan 'lay' < *loghéye/0- (cf. OE licgan 'lie' < *\eg-ye/0-),fra-wardjan, OHG far-werten 'corrupt, spoil' < *wort-éye/0- (with Verner's Law treatment of *t, cf. Goth, waírþan, OHG werdan 'become' < *wért-e/0-); and in Balto-Slavic by o-grade verbs with infinitives in Lith. -yti, OCS -if/, whose presents have been remodeled: Lith. vartyti (3p. vafto) 'turn, turn over, move (repeatedly)', OCS vratiti (3sg. vratitT) 'turn, turn around (tr.)' <— *wort-éye/0-; 58
The only usage of *-s¿é/ó- which cannot be so easily reconciled with "imperfectivity" is the type of Lith. dyg-sta, tifp-sta. Senn, however, considers the original meaning of Lith. verbs in -sta- to have been "intransitiv-inchoativ": "Doch ist der inchoative Sinn in der ipf. Grundform oft verlorengegangen und nur in den pf. Formen erhalten; z.B.: ipf. pykti 'ziirnen, zornig sein' und pf. papykti, supykti 'zornig werden'" (1966:266). The stativity of pykti (pres. 3p. pyk-sta), etc. may hence be a secondary development resulting from their imperfective aspect within Lithuanian; for a parallel from Slavic, cf. Russ. impf. snēt 'is dying (off)' (of fish) < *sup-ne-t <— PS1. *súp-ye-fí vs. Russ. CS1. pf.u-svple- 'fall asleep'.
71
Lith. prašyti (3p. prāšo) 'ask, request', OCS prositi (3sg. prositī) 'ask, demand' <— *prok-éy e / 0 - (cf. §6.2). 59 Based on the evidence of these branches, as well as clear survivals elsewhere such as Lat. moneō 'remind, advise, warn' (< 'make aware/mindful', to *men-; cf. meminī 'remember, am mindful of < pf. *me-mon-), Gr. sobéo: 'drive away (birds)' (< *'make retreat'; cf. sébomai 'feel fear/awe' > 'worship', Ved. tyájate 'submit' < *tyeg w -), scholars have traditionally reconstructed *Co(R)C-éye/0- as the productive iterative-causative formation for PIE: cf. Brugmann and Delbriick 1916:244ff. and most recently LIV:22-3.60 In addition to the well-known reflexes in the classical languages, Tocharian appears to preserve at least one or two o-grade *-éye/0- presents; cf. R. Kim, forthcoming, fn. 70 for discussion. As for Anatolian, it is now generally agreed that Hittite exhibits cognates of at least two well-established o-grade causatives, namely lukkezzi 'kindles' < *lowkéyeti (Ved. rocáyati, Av. raoCaiieiti, Lat. lúcēre; Watkins 1973:68-9, contra Hoffmann 1968:215-6, Oettinger 1979:273 [< *léwketi; note that Ved. attests only a mp. pres. rócate 'shines']); cf. pret. lukkatta 'it has become bright, it has dawned' (> adv. 'in the morning'; —> NH pres. lukzi 'it dawns') <— OH lukta (lx) < *luk-tó, reanalyzed aor. inj. (Melchert 1994a: 149 following Oettinger 1979:273-6, contra Watkins 1971:69 [< *léwketo]); and waššezzi 'clothes' < *woséyeti (Ved. vāsáyati, Goth, wasjan; Eichner 1969:31-2, Melchert 1984:31-5), with geminate šš from athematic weš-, e.g. pret. 3sg. wēšta [wēs.sta] < *wés-to (Melchert 1984:3 l-2fn.64, 1994a: 152). If Melchert (1984:37-8, 1994a: 176-7, contra Oettinger 1979:63-7, 1992:219-20) is correct, lāizzi, lānzi 'release, let go' continues *laēti, *laōnti < *lohi-éye-ti, *loh]-éyo-nti, with loss of both intervocalic *hj and *y and resulting contractions (for the root cf. Lat. lē-tum 'death', lē-nis 'soft', Lith. lé-nas 'quiet', OCS lē-tī 'permission', lē-nú 'slow', and with root extension Goth, lētan 'let, allow' < PGmc. *laet- < *lehj-d-, Lat. lassus 'tired' <— *lh,d-to-; LIV:358-9, LIV2:399-400).61 This type is also attested in all the other Anatolian languages except for Palaic, where its absence is almost certainly due to chance: 59
The Slavic verbs of this type may be distinguished from denominative /-verbs in their accentuation: cf. §6.6, fn. 41. 60 In addition to Lith. vartļfti, examples of iterative or "frequentative" value include Gr. cpopéco 'carry repeatedly/habitually; carry around, here and there' to phérv: 'carry', óχεco 'hold fast, keep doing' to εχco 'have' (vs. 'let ride' to fεχco* 'bring, carry', Pamphy. impv. 3sg. f εχεTco); cf. Delbriick 1897:109-15. 61 Another possible example of this type in Hittite is uwate- 'carry', if from preverb *au- + h *wod -éye/0- 'lead'; see Lehrman 1985:258, Melchert 1994a:66, 134
72
CLuv. 3sg. dúpiti (HLuv. tu-pi-ri), 3pl. dupainti, Lycian 3sg. tubidi, 3pl. tuheiti 'strike' < *(s)towbhéye/0-, with 3sg. *-īdi < *-éydi < *-éyedi < *-éyeti (Melchert 1994a:277; cf. Morpurgo Davies 1982:267-8); for the root cf. Gr. stuphelizo: 'strike hard, treat roughly'; CLuv. wiši-lwišai- (pret. 3sg. wišita, 3pl. wiSainta), Milyan wisi-lwisei- (pres. 3sg. wisidi) '(op)press' < *woyséye/0-; Lyd. fa-karsed 'cuts (out)' < *kors-éye-di, cf. Hitt. karS- 'cut'; Lyd. karared (kat-šared) 'stands watch' < *sor-éye-di to the root *ser- (Av. (ni)har- 'watch over, guard'), cf. sarēta 'protector, patron' (Melchert 1992:46-7, 1994a:380). Notice that CLuv., Lye, and Lyd. agree in showing lenited 3sg. endings in the pres. and pret.: CLuv. 3sg. -til-di, -tal-da (whence HLuv. rhotacized -ri), Lye. -di, -da, Lyd. -d. Thus the accentuation of these denominatives in PA was *-éye/0-, and they may with confidence be connected with causatives elsewhere in IE of this type. Although it is not clear whether o-grade presents in *-eye/0- should be reconstructed for PIE, much less as a productive transitive-causative,62 the evidence of Anatolian, Germanic (cf. fra-wardjan, OHG far-werten 'corrupt, spoil' < *wort-éye/0above), and possibly Slavic (§6.6, fn. 41) unanimously points to stress on the first vowel of the suffix *-éye/0-.63 Whatever the origin of the o-grade root vowel, then (perhaps related to o-grade deverbal nomina agentis of the type of Greek phóros 'bearer' to phero: 'bear, carry'?), the columnar stress of this suffixed present results simply from the accent 62
The existence of Hitt. lukkezzi, waSSezzi suggests that at least some verbs of this shape had acquired causative meaning already in the protolanguage, although it is always possible that *woséy e / 0 - and *Iowkéye/0- became semantically specialized in this function independently in Anatolian and the rest of IE. Note, however, that none of the other Anatolian *-éye/0- presents listed above is unambiguously causative; rather, Luv. dupiti 'strikes', Lyd. fa-karsed 'cuts (out)', and the like appear to be functioning as simple transitives. Likewise, the original function of the Indo-Aryan presents in -áya- was neither iterative/intensive nor causative, but merely transitive (first proposed by Thicmc [1929:17-30] and argued at length by Jamison [1983a: 183-9 and passim]; according to Macdonnell [1910:393], "of about 150 causative stems appearing in the RV. at least one-third lack the causative meaning"); the causativity so familiar from classical Sanskrit grammar has arisen within the (pre)history of the language. The picture which emerges from this discussion is one of an incipient category — perhaps comprising only a few verbs at the PIE stage, of which *lowkeye/0- and *woscye/0- are two likely candidates — which then gained in numbers in late PIE and eventually became productive and developed marked transitive (i.e. causative) function in a number of "classical" IE daughter languages. The well-known -áya- causatives of Indo-Iranian, along with related forms in Germanic, Italic, etc., would then be largely the product of post-PIE, and probably at least partially independent, developments. 63 Tedesco (1923:302; cf. Mayrhofer 1989:13fn.27) notes that the Western Middle Iranian (e.g. Middle Persian) present must derive from the PIr. transitive-causative *bar-áyā-mi, *bar-áya-si, *baráya-ti, etc., with stress on the first suffixal vowel just as in Skt. -áya-.
73
of the suffix. The contrast between transitive-causatives such as pres. 3sg. *lowk-éye-ti 'kindles, sets alight/afire' and denominatives to o-stem nouns such as *usne-yé-ti (*wosne-yé-ti) 'offers for sale', would then be represented as follows: *
*
(*
(*
*lowk é
ye
ti
*usn
e
yé
ti
The accentuation of the PIE verb thus closely parallels that of the noun in contrasting an archaic stratum of athematic formations with complex ablaut and stress alternations, and a newer, productive thematic type marked by columnar stress on the thematic vowel (or, in the case of *-éye/0-, the first vowel) of the suffix. Let us next turn to the accentual systems of some of the IE languages which preserve lexical stress and examine how much information they contribute to the reconstructions advanced above, and how much the reconstructed PIE system has been preserved and/or transformed.
74
Chapter 3 Continuity Amid Innovation: The Development of Proto-Indo-European Accent in Greek 3.1. Introduction1 In recent years, certain scholars have questioned whether the accentual system found in the Attic-Ionic dialects of ancient Greek, in particular the mostly archaic Ionic of the Homeric epics or the Attic of classical Athens, continues that reconstructed for ProtoIndo-European (PIE).2 In his study of Indo-European nominal accent, Lubotsky (1988:121) states: "The original accentual distribution, however, was most disturbed not by the phonetic shifts but by analogical patterning. Already in prehistoric times Greek had generalized a uniform accentuation for many categories and suffixes. A wellknown example is the recessive accentuation of the finite verb. Moreover, all neuters (including those in -on) became barytone, with only few exceptions; also barytone are feminines in -a, i-stems and substantives in -us, while adjectives in -us, -los, -nos, and -ros show pervasive oxytonesis. There are several indications that this process of generalizing a single accentuation pattern for every category went on in historical times. A good example is the suffix of nomina actionis -mo-, which shows both types of accentuation in Homer but is almost exclusively oxytone in later texts." Immediately afterwards, in his discussion of Greek /-stems, he concludes that "in this light, the identical accentuation found in Gk. pósis 'husband' and Skt. páti- 'id.' or in Gr. ois 'sheep' and Skt. ávi- 'id.', which is mentioned time and again as proof of the original identity of the Sanskrit and Greek accentual systems (cf., e.g. Kuryiowicz 1968:20), is not significant. The accentuation of these Greek words is ambiguous; it might be old, but it might also be analogical" (121). Thus, according to Lubotsky, even the earliest (alphabetic) Greek of Homer had undergone major analogical shifts of accent and generalized a single accentuation for most nominal or adjectival formations, preserving little of the original PIE pattern. This view is adopted by Halle (1997:301, 304-5), who goes so far as to claim that the accentual system of ancient Greek had no historical connection with that of PIE: "...Greek reintroduced lexi'Note that the "spurious diphthongs" εi and ou are here spelled I, 6, acute i, <5, circumflex ε. ō (vs. "genuine" εi, ou), reflecting their likely pronunciations [e:], [o:] in Attic Greek of the 5th century BC. As usual, a colon denotes vowel length. ^Data from other dialects will be mentioned below where relevant.
75
cally accented morphemes. The lexical accentuation of Greek, however, is an original development that is unrelated to the accentuation of the protolanguage" (301). According to the latter view, the PIE system of underlyingly accented or unaccented roots, stems, and endings, supposedly reflected in Vedic Sanskrit and BaltoSlavic, was entirely lost in the ancestor of Greek; in other words, ancient Greek holds little or no relevance for the reconstruction of PIE accent. Halle (1997:303-5) explicitly proposes that prehistoric Greek first passed through a stage like Archaic Latin or contemporary Czech, in which all underlying lexical accentuation was lost and initial accent generalized according to the parameters given for PIE in §1.2.3. Subsequent undefined developments then led to the contrast observed in Homeric and Attic Greek between e.g. barytone (stem-stressed) and oxytone (ending-stressed) o- or d-stems. 3.2. Was PIE accent preserved in ancient Greek? Lubotsky (op. cit.) is certainly correct when he notes that most Greek words sharing a particular suffix also exhibit the same accentuation. For instance, most (all?) ustem adjectives are oxytone, whereas most w-stem substantives are barytone, e.g. barús 'heavy', brakhús 'short', eurús 'broad, wide', platús 'flat, broad' vs. pε:khus 'forearm', ástu 'city'. Similarly, most adjectives in -ros (< PIE *-ro-) are accented on the "thematic" vowel o, e.g. eruthrós 'red', elaphrós 'light, nimble', as opposed to the mostly root-ac cented nouns in -mos (< PIE *-mo-), e.g. ógmos 'furrow'. Some of these patterns are undoubtedly due to the generalization of particular surface patterns and their corresponding underlying accentual representations within the prehistory of Greek; for some likely examples, see §3.6. To conclude on the basis of these distributions that Greek has almost entirely given up any trace of PIE accent, however, is demonstrably wrong. Many of the patterns known to earlier generations of scholars of Greek and emphasized by Lubotstky are at least partially inherited from the parent language. For example, there is strong evidence that the nominal suffix *-ro- was underly ingly accented in PIE; the reconstructed PIE adjective in *-ro- was marked by "zerograde" vocalism of the root and oxytone accent.3 If we find that Greek adjectives in -ros are as a rule oxytone, the obvious inference is that Greek has inherited and preserved the underlying accent of the PIE suffix *-ró-.
■^Excepting a handful of archaic root-accented forms with "full-grade" e-vocalism (Vine 1999a), of which at least some are backformed to substantivized collectives.
76
More importantly, Lubotsky omits any mention of intraparadigmatic alternations of accent and their significance for the reconstruction of (pre-)Greek accent. To cite only one classic example, the contrast between sg. nom. pó:s, acc. póda and gen. podós, dat. (<— loc.) podi exactly matches that between Skt. sg. nom. pat, acc. pādam and gen. padás, dat. padé, instr. padā, loc. padí. Ablaut alternations, such as we find in athematic verbs (i.e. verbs lacking the thematic vowel -o
e-) or archaic r-, n-, r/n-, i-, and «-stem nouns, are
especially valuable for recovering what might have been the PIE or pre-Greek state of affairs prior to the various accent shifts that produced the attested Greek forms, e.g. columnarization of accent in nominal paradigms. In the r-stem kinship nouns, for in stance, the ablaut contrast between pi. nom. patér-es, acc. patér-as and dat. patrá-si suggests that the last of these forms continues *patŗ-sí <— PIE *ph2tŗ-sú (cf. Skt. pitfsu, which has likewise undergone columnarization); cf. §2.2.1 for details and discussion. Finally, the sheer number of matches in lexical accent and paradigmatic patterning between Greek and Vedic — not to mention Anatolian and the indirect evidence of Verner's Law in Germanic4 — absolutely excludes the possibility that Greek could have lost and then reintroduced underlying accent of morphemes.
In addition to the
paradigmatic alternations given just above, note the following small sample of accentual word equations:5 Gr. hippos (<— *ippos, cf. PN Alk-ippos), Skt. ášvas (ON jór, OE eoh < PGmc. *eχwaz) < PIE *ék\vos 'horse'; Gr. néos (< *néwos), Skt. návas (Hitt. nēwaš, Luv. nāwaš < PAnat. *néwos) < PIE *néwos 'new'; Gr. lúkos, Skt. vfkas (Goth. nom. pi. wulfðs, OHG wolf < PGmc. *wulfaz) <— PIE *w|k w os 'wolf; cf. Skt. vŗkí, ON ylgr (< PGmc. *wuly w iz) <— PIE *wļkwíh2 'she-wolf; Gr. alphós 'white leprosy', Hitt. alpāš 'cloud' (cf. Lat. albus 'white') < PIE *aībhós 'white'; Gr. klutós 'famous', Skt. šrutá- 'heard' < PIE *lclu-tós 'audible' to the verbal root *lclew- 'hear'; likewise other verbal nouns in *-tó- with zero-grade of the root; Gr. he-katón, Skt. šatám (Goth, hund, OHG hunt < PGmc. *hundā by Verner's Law) < PIE *krņtóm 'hundred'; Gr. zugón, Skt. yugám (Hitt. iukan, never "ivkan") < PIE *yugóm 'yoke'; Gr. ókris, ákris, Skt. ášris < PIE *h20Jc-r-i-s, obi. *h2élc-r-i- 'point' (aerostatic, cf. §2.2.1,2);
4
For discussion of Verner's Law and its significance for the reconstruction of PIE accent, see
§1.3. 5
Other examples may be found in Schwyzer 1939:380-1 and throughout Euler 1979.
77
Gr. pósis, Skt. pátis < PIE *pótis 'master, husband' (originally aerostatic? cf. §2.2.1,2); Gr. dóru 'spear' (Horn. gen. do.rós < *dorwós), Skt. dāru, gen. dróh 'tree' < *dóru, obl. *dréw- 'tree' (cf. above); Gr. patεsr, Skt. pitá (Goth. voc. fadar, OE feeder, OHG fater < PGmc. *faðer- by Verner's Law) < PIE *ph2tē(r) 'father' (§2.2.1, 5); Gr. phrá:tε:r 'member of a clan/phratry', Skt. bhrātā (Goth, brōþar, OE brōðor, OHG bruoder < PGmc. *brōþer-) < PIE *bhréh2tē(r) 'brother' (§2.2.1, 2). To be sure, there are apparent Vedic-Greek cognates in which the attested stresses do not correspond. Many such pairs, however, raise other phonological difficulties or are suspect of being independent post-PIE creations; moreover, systematic discrepancies in accent may often prove useful in reconstructing the original accentual pattern of the ancestral language (cf. §2.2.1, esp. 4). The weight of impeccable equations such as those just listed clearly demonstrates that Greek inherited lexical accent from the parent language in at least some forms, and strongly disfavors the hypothesis of Lubotsky and Halle. Given the principle that phonological mergers are irreversible by internal linguistic means (cf. Garde 1961:389, Labov 1994:31 Iff.), it would have been impossible for Greek to eliminate the distinctive lexical accent of PIE, then somehow recreate or recover the correct distribution of, say, barytone vs. oxytone o-stem nouns in so many cases — much less the alternation of stemstress in the nom. and ace. vs. ending-stress in the gen. and dat. of monosyllabic consonant-stem nouns. Of course, the accentual system of Homeric and Attic Greek, as described by the grammarians of Hellenistic Alexandria, has undergone several well-known innovations, e.g. limitation of surface stress to the last three syllables of the word (the Three-Syllable Rule, henceforth abbreviated TSR) and near-total restriction of nominal stress alternations to monosyllabic stems. As a result of these extensive changes, the PIE accentual system survives into ancient Greek only to a limited degree. Nevertheless, Greek does preserve the underlying accentual representations of a number of PIE nominal and verbal morphemes and nominal stems, as I demonstrate below. In the following sections, I will apply the brackets-and-edges analysis to ancient 6
Greek. Despite several significant innovations, it will be shown that Greek does indeed 6
EarIier analyses of ancient Greek accent may be found in Kiparsky 1967, 1973:796-805 and Steriade 1988. Cf. also Noyer 1997 on the need for intermediate representations in the derivation of Greek surface stress (see fn. 9). In the following discussion, "word" is understood in the sense of "accentual word", i.e. a phonological unit, typically composing a single nominal or verbal form and one or more proclitics or enclitics, upon which stress is assigned. As the following study is devoted to
78
preserve the lexical acccent of PIE stems and morphemes, both in a number of nonfinite verbal forms (§3.3) and in the noun (§3.4). Based on an examination of relic forms and differences in the accentuation of PIE, Proto-Greek, Homeric, and Attic, many of the innovations which have altered the accentual system of Greek may be placed within a relative chronology (§3.5). Finally, a number of previously unexplained idiosyncrasies of nominal accentuation may be understood as reinterpretations of PIE alternating stems, in particular of the proterokinetic inflection (§3.6; cf. §2.2.1,4). 3.3. Relics of PIE accent in the Greek verb 3.3.1 Exceptions to recessive verbal accent One of the accentual peculiarities of Vedic Sanskrit is that the underlying accentuation of finite verb forms surfaces only in subordinate clauses and at the beginning of a clause or pāda (verse). Otherwise, a postlexical rule of destressing results in forms such as pres. act. 3sg. yunakti, 3pl. yunjanti 's/he yokes, they yoke' for yunákti, yunjánti < PIE nasal-infixed *yu-né-g-ti, *yun-g-énti (cf. Whitney 1924:223-6). If we assume that the same rule was operative already in PIE — or at least in the dialect of post-PIE that gave rise to Greek — it follows that pre-Greek, like Vedic, must have contrasted unstressed finite forms in main clauses with stressed forms elsewhere. A generalization of the unstressed variants to subordinate clauses under these circumstances would not have been surprising. As a result, all finite verb forms at that stage of pre-Greek would have been unstressed, i.e. enclitic, a situation preserved in Attic and (to the extent that its evidence is independent) Homeric in most forms of the presents of e:mi 'be' and phε:mi 'say' (Wackernagel 1877:457-8; cf. Chantraine 1942:381, Risch 1975:475-6, Rix 1976:43, 199-200). The introduction of the Three-Syllable Rule (TSR) then assigned surface stress to the mora farthest to the left permitted by the final syllable, resulting in the familiar "recessive" accent of Attic-Ionic:7 1. Forms ending in a syllable containing a short vowel (or certain short diphthongs, e.g. nom. pi. -oi, -ai or mediopassive and infinitival endings in -ai\ cf. Lejeune 1972:296, Risch 1975:473, Rix 1976:47-8) receive antepenultimate stress — in demonstrating the historical continuity of PIE accent in Greek, I will not discuss the accentuation of clitics here: cf. Schwyzer 1939:386-9, Lejeune 1972:299, Rix 1976:43-4 for details. 7 Cf. Schwyzer 1939:378-9, Lejeune 1972:295-6, Risch 1975:471-2, Rix 1976:42. In the East Aiolic dialect of Lesbos (Lesbian), all words carry recessive stress, so that stress is no longer distinctive (Schwyzer 1939:383, Lejeune 1972:298, Risch 1975:475, Rix 1976:43): cf. Sdeus,M:mos 'soul, spirit', pótamos 'river' vs. Att.-Ion. Zeus, thu:mós, potamós. On exceptions to the TSR, e.g. gen. sg. póleo:s, pl. póleo.n to pólis 'city', see §3.4.2, fn. 22.
79
disyllabic words, initial stress, with circumflex intonation on a long vowel:8 Horn. aor. 3sg. kéleuse 's/he ordered', lábe 's/he took', heure 's/he found'; pres. 3sg. ágetai 's/he takes for her/himself, inf. é.nai 'to be'. 2. Forms ending in a syllable containing a long vowel or diphthong (excluding the diphthongs in 1.) are stressed on the penultimate, with acute intonation on a long vowel or diphthong: pres. lsg. kli.no: 'I lean (tr.)', khairo: 'I enjoy, am happy', 3sg. klíinei, fd'aírei. 3. Long monosyllables receive circumflex intonation: Horn. aor. 3sg. thε: 's/he put', impf. 3sg. ε:n 's/he was'. Thus, the previous underlyingly unaccented nature of finite verb forms is masked by the subsequent operation of the TSR. As Hoenigswald (1998:271) succinctly puts it, "recessivity is the guise taken perforce by their former enclisis, which can be reconstructed by comparative evidence".9 Should one conclude that PIE lexical accent in verbal inflection was lost without a trace in Greek? Almost completely: excepting a few fossilized thematic aorist active imperatives in -é (and the regular aorist mediopassive imperative ending -oú < *-éo; cf.
8
The so-called "sa.-fε.ra-rule" (after ace. sg. so:tε:ra 'savior', an example of it), which assigned circumflex intonation to long penults followed by a final short vowel, had the effect of eliminating intonational distinctions in penultimate position: the intonation of a long penult is entirely conditioned by the quantity of the vowel of the final syllable, so that ' and A contrast only word-finally. Not all Greek dialects appear to have undergone this innovation: cf. Doric ná.sos 'island', gunaikes 'women' (Schwyzer 1939:377, 384, Lejeune 1972:297, Rix 1976:44-5; otherwise Risch 1975:474-5). 9 Cf. Wackernagel 1877:459, Schwyzer 1939:389-90. Subsequent to the assignment of recessive accent, loss of intervocalic *y, *h (< *s), and later *w led to the contraction of vowels across the resulting hiatuses (cf. Chantraine 1942:382-3, 384). Examples include so-called "contract verbs" (i.e. verbs with vowel-final stems, e.g. pres. lsg. ti:mð:, 3sg. ti.má.i 'honor' < ti:má-o:, ti:má-ei) and the subj. and opt. of athematic verbs and the stative/passive aor., e.g. pres. subj. lsg. titho.\ tithε:i < *tithé :-o:, *tithé :-ei to títhε:-mi 'I put', aor. subj. khar5.\ kharε:i < *kharé :-o:, *kharε :-ei to e-k''úrε:-n 'I was happy'. The exceptional accentuation of athematic pres. opt. mp. 3sg. didoito, titheito, histaíto (to dído:-mi, títhε:-mi, hístε:-mi 'I give, put, stand (tr.)') probably results from contraction following the loss of laryngeals in intervocalic position: (post-)PIE *di-dh r ih r to, *d h i-d h h,-ih,-tó, *s(t)i-sth 2 -ih r tó > pre-Gr. *didóīto, *tithéi'to, *histáí'to —> didoito, tifoito, histaito (sim. act. lpl. didoimen, tifieimen, histaimen\ Risch 1975:476-8). These were further supported by the corresponding aor. opt. forms doito, theito, staito, as opposed to príaito, dúnaito (to aor. e-priá-mε:n 'I buy', diina-mai 'I am able'), which have adopted regular recessive surface stress. For other possible survivals of laryngeal contraction, see fn. 26. On the implications of these contractions for the surface opacity of recessivity, cf. Noyer 1997. I leave aside the accentuation of verbal compounds such as iptv. 2sg. apó-dos 'give away!'. sum-pró-es 'send forth together!', sun-é-skhon 'held together', in which the surface stress cannot retract beyond the final syllable of the last preverb (hence not "ápodos", "súmproes") or the augment (hence not "siinesk''on"; Chantraine 1942:384, Risch 1975:476); on the cyclic computation of stress in such forms, cf. Noyer 1997:522-4 with refs.
80
Chantraine 1942:382, Risch 1975:475-6, Rix 1976:42-3, 215 and see below, §3.3.2), finite verbal forms are synchronically unaccented and receive recessive surface stress. Non-finite verbal forms such as participles and infinitives, however, would not have been subject to postlexical destressing. To be sure, one finds recessivity in a number of participles, including several in masc. -menos, fern, -menε:, neut. -menon (stem -meno), illustrated here with forms of pres. graph-e/0-, fut. graps-e/0-, aor. graps-a- 'write': pres. mp. graphómenos, graphoménε:, graphómenon 'writing for oneself, being written'; fut. mid. grapsómenos, -ménε:, -menon 'going to write for oneself, who/which will write for oneself; fut. pass. graphthe:sómenos, -ménε:, -menon 'going to be written, which will be written'; aor. mid. grapsámenos, -ménε:, -menon 'who/which writes for oneself, wrote for oneself. These correspond to the recessively stressed infinitives gráphe-sthai,
grápse-sthai,
graphthé:se-sthai, and grápsa-s^ai, respectively. In the absence of probative evidence for underlyingly accented (graph-, I adopt the default hypothesis that verbal stems, and hence the above participles, are unaccented, with recessive stress determined by the TSR. Three others, the pres. and fut. act. ptcp. in masc. -o:n, fern. -o:sa, neut. -on (stem -ont-) and the "first" (i.e. productive) aor. act. ptcp. in masc. -a:s, fern. -a:sa, neut. -an (stem -ant-), appear at first glance to be recessive as well: pres. act. phéro:n, phéro:sa, phéron 'bearing, carrying', baíno:n, baíno:sa, baínon 'going, stepping'; fut. act. oíso:n, oíso.sa, oison 'going to bear, that will bear'; aor. act. deíksa:s, deíksa.sa, detksan 'showing, that showed'. As with grap1'-, there is no reason to assume accented (pher-, (ois-, (deiks- here (but cf. fn. 10). However, an examination of longer forms, e.g. those of paideúo: 'instruct, teach', reveals an irregularity in the neut. nomVacc. sg. in -on, -an: paideúo:n,paideúo.sa,paideúon 'instructing'; paideúso:n,paideúso:sa,paideúson 'going to instruct, that will instruct'; paideúsa.s, paideúsa.sa,paideúsan 'instructing, that instructed' Why do we not find recessive "paídeuon", "paídeuson", "paídeusan"! The answer lies in interaction of the TSR with the paradigms of these participles: the neut. nom./acc. sg. was the only form in which stress could retract to the penultimate syllable of the verbal 81
stem, in this case pa/. 10 This sparsity of phonetic (surface) evidence for recessivity caused a phonological reanalysis: instead of pai[deúo:n, pai[deúso:n, pai[deúsa:s, with stress determined by the TSR, speakers of Greek at some point abstracted underlyingly accented forms pai(deúo:n, pai(deúso:n, pai(deúsa:s. (For a parallel case of reanalysis in the noun, cf. §3.4.1.) Although the stress of pres. and fut. act. ptcps. in -o:n, -o:sa, -on may be explained starting from a recessive origin, there remain four other participles whose non-recessive stress cannot be account for in this way (cf. Rix 1976:199-200,215): 1. active participles of the thematic or "second" aorist, ending in -ó:n, -o:sa, -ón, e.g. lipo:n, lipd:sa, Upon 'leaving (behind)', elfióm, elfio.sa, elthón 'coming'; 2. aorist passive participles (still largely stative in Homer) in -é:s, -é:sa, -én (stem -ént-), e.g. Horn. kharé:s, kharé:sa, kharén '(being) joyous, happy', Att. lufié.s, luflé.sa, luflén 'released'; 3. perfect active participles in -o:s, -uta, -as (stem -ót-), e.g. Horn. gegono:s, gegonuia, gegonós 'being (X years) old', Att. pepaideuko.s, pepaideukuia, pepaideukós 'having instructed'; 4. perfect mediopassive participles in -ménos, -ménε:, -ménon, e.g. gegramménos, -ε:, -on 'having written for oneself, having been written'.
10
Cf the following paradigm for pai(deúo:n 'instructing' (excepting the dual), with TSR limitations marked; paideúso.n and paideúsa.s are entirely parallel: nom. sgace. gen. dat.
masc. neut. pai[(deúo:n [pai(deúon pai[(deúonta pai[(deúontos pai[(deúonti
fem. pai[(deúo:sa pai[(deúo:san pai(deu[ó:sε:s pai(deu[ó:sε:i
nom. pi. ace. gen. dat.
pai[(deúontes pai[(deúonta pai[(deúontas pai(deu[ónto :n pai[(deúo:si
pai[(deúo:sai pai(deu[ó:sa:s pai(deu[ó:so:n pai(deu[ó:sais
The corresponding infinitives match their participles in stress: pres. act. pai(deú-e:n, fut. act. paifdeúse:n, aor. act. pai(deús-ai (-ai here counts as short; cf. opt. 3sg. paideiisai with "long" -ai, like 2sg. paideiisais). This innovative accentuation appears to be limited to nonfinite forms only: cf. iptv. mp. 2sg. paideiisai with recessive accent, thus excluding a stem pai(deusa-. — On the so-called "'Aiolic infinitive" ending -men and East Aiolic (Lesbian) -menai found with athematic presents in Homer, e.g. ém-men(ai) 'to be' < *eh-m- for Att. é.nai < *eh-enai (to pres. lsg. e:mi< *eh-mi), f-men(ai) 'to go* for Att. i-énai (to eí-mi), éd-menai 'to eat' for Att. éd-e:n (to thematized pres. éd-o:), cf. Rix 1976:238. I leave unresolved for now whether the same reanalysis occurred in monosyllabic pres., fut.. and ("first") s-aor. act. stems, e.g. whether pher-, ois-, deiks- remain underlyingly unaccented, which leads to no problem in their participial stress, or have been reinterpreted as accented (pher-, (ois-. (deiks- after polysyllabic stems.
82
As has long been known (cf. Wackernagel 1877:459-60, Schwyzer 1939:389-90, Chantraine 1942:378ff.), these forms preserve an older accentual pattern which was lost in the corresponding finite forms when the latter generalized their destressed variants (whence recessive stress in Greek).11 In the next two sections, we shall discover that the nonrecessive stress of these four participles provides evidence that the inherited underlying accentuation of at least certain PEE verbal morphemes has survived into Greek. 3.3.2. Survivals of lexical accent: the thematic aorist As is well known, the Homeric and Attic "second", i.e. thematic aorist continues PIE root aorists which have been thematized on the basis of 3pl. *-ó-nt <— *-é-nt, e.g. Horn. 3sg. lipe 's/he left (behind)' <— *likw-é-t <— PIE *léykw-t after 3pl. Upon < postPIE *likw-ó-nt < PIE *likw-é-nt (cf. §2.3.1.1, §2.3.2.1). The diachronic origin of both the "zero-grade" vocalism and the oxytonesis of Greek aor. act. ptcps. such as lipó:n, lipó:sa, Upon 'leaving (behind)', eltho:n, elthō:sa, elthón 'coming' is thus clear: both were generalized from the 3pl. which, along with other plural and dual forms, contrasted with the full e-grade and stem-stress of the singular. One may correspondingly reconstruct *likw-ónt-, *hiludh-ónt- as the thematized aorist participial stem in post-PIE. But how should the accent of these participles be analyzed synchronically? If the zero-grade aorist roots were underlyingly postaccenting, e.g. eWl(-, lip(-, then the contrast in accentuation between these participles and pres. act. phéro:n would immediately follow: since pres. pher- is unaccented, phér-o:n receives recessive accent according to the TSR (but cf. fn. 10), whereas in lip-6:n it is the thematic vowel which re ceives the accent according to the above rules of prosodic computation. Similarly in the infinitive, pres. phér-e:n contrasts with aor. lip-e:n. Within the brackets-and-edges framework, the innovative TSR may be implemented by limiting Line 0 Projection to the final two or three syllable heads. I tentatively propose the following two steps: 0a) Insert a left boundary to the left of the rightmost (i.e. final) syllable head if the latter is a long vowel or diphthong (excluding nom. pi. -oi, -ai, mp. and inf. verbal endings in -ai; cf. above), i.e. if it "counts as long" for the purposes of accent. Ob) Construct a single binary foot to the left.
11
1 wish to thank an anonymous reviewer for the PWPL for emphasizing the need to distinguish between finite and nonfinite verbal forms.
83
The resulting boundary, which will be indicated in the examples below with the symbol [, thus demarcates the "stressable" portion of a Greek word: only those syllables which lie to its right may project an element onto line 0. 12 Otherwise the rules for stress computation in ancient Greek are the same as those given above for Russian. I repeat these for convenience: LineO Project syllable heads Project L boundary of lexically specified syllable heads Edge-Marking: RRR Head: L Line 1 Edge-Marking: LLL Head: L The metrical grids for phéro:n and lipám are thus * (* *
*) h
\p ér
o:n
*(
* (* *)
[lip
o:n.
12 Cf. the situation in Latin, where words take penultimate stress if their final syllable contains a long vowel and antepenultimate stress if their final syllable contains a short vowel; because Latin lacks lexical accent (and hence Syllable Boundary Projection), one may simply project the left boundary of a final long syllable on line 0, then construct a single foot to the left by ICC. For a different analysis, cf. Halle and Idsardi 1995:424-5. Note that the analysis proposed here is also required for modern Greek, in which historically long endings such as substantival o-stem gen. sg. -u, ace. pi. -us, gen. pi. -on (< ancient -o.\ -o:s, -a:n; cf. §3.4) still count as underlyingly "long" in step 0a. The contrast between e.g. nom. sg. ánβwp-os and gen. sg. anβróp-u may thus be represented by the grids
* (* * [án
* Qro
*\ pos
* (* * an
[βró
*) pu
which correspond perfectly to ancient Greek * (* * [an
* tf'ro:
*\ pos
* (* * an
84
[rVa:
*) po:
This explanation, however, is ruled out by unaugmented finite aorist forms, i.e. "injunctives", which survive in Mycenaean and are still quite frequent in Homer. Based on the evidence of lipe, etc., the aorist root cannot have been postaccenting — unless the thematic vowel -e/0- is a dominant cyclic affix in Greek, for which there is no evidence. These unaugmented forms suggest that the root is instead underlyingly unaccented, and that the initial accent of unagumented aorist forms such as lipe is due to the LLL and HeadL parameter settings for line 0. But if that is the case, the oxytonesis of lipðm, etc. presupposes that the participial endings are still underlyingly accented in Greekl It is therefore not true that ancient Greek has completely lost lexical accent in the verbal system: although the majority of both finite and infinite categories exhibit recessive accent, the decidedly non-recessive oxytonesis of the participle of the thematic aorist can only be analyzed in synchronic terms by positing an underlyingly accented ending -(o:n, -(o.sa, -(on. Further support for this analysis comes from the imperative of the thematic aorist, which furnishes the only nonrecessive finite verb forms in all of Greek. The fossilized active imperatives elthé 'come!', heuré 'find!', eipé 'speak! tell!', idé 'see!', labé 'take!' suggest that iptv. act. 2sg. -(e, like the participial ending, was also originally accented (cf. Schwyzer 1939:799); similarly pi. elftéte, heuréte, eipéte, idéte, labéte, with accented 2pl. -(ete. These relics apparently reflect an earlier stage of the language in which all imperatives to thematic aorists were oxytone, before analogy to the recessive accentuation of the rest of the paradigm resulted in the majority pattern of Horn, lipe 'leave (behind)!', pie 'drink!', Att. bale 'throw!',páthe 'suffer!'.13 Likewise, the regular aorist middle iptv. 2sg. ending Horn, -éo < *-éso (Att. contracted -o:) has preserved the accentuation of the theme vowel, e.g. genéo 'become!', idéo 'see for yourself!', labéo 'take for yourself!' (ibid.). To recapitulate, we may trace the accentual development of the thematized aorist from PIE to Homeric Greek as follows. In PIE, the past-tense ("temporal") augment *(ewas underlyingly accented: cf. Skt. aor. 3sg. vid-át, á-vid-at 's/he found' (Horn, ide, eide < *wíde, *éwide), where the stress of the unaugmented variant disappears after the uniform á-. As seen above, the thematic aorist which arose in late or post-PIE had an 13
Cf. Schwyzer 1939:389-90, 799. Synchronically, these five must almost certainly have been treated as lexical exceptions: the unusual oxytone aor. iptv. was entered as part of the underlying specification of the verb. An anonymous reviewer for the PWPL has argued for treating these roots as exceptionally postaccenting rather than assuming that iptv. -e, -ete are accented only when suffixed to them, but cf. above for the evidence against such an analysis.
85
accented thematic vowel *-(e/0-- This vowel has lost its accent in all finite forms by Homeric times, so that post-PIE unaugmented aor. ("injunctive") lsg. *likw-ó-m, 3sg. *likw-é-d develops to Horn. Upon, lipe. *
*
(* * w
post-PIE
(* *
r*\
*Iik
óm
—>
*)
Homeric [lip on
In Homeric, the augment is also no longer accented: the placement of stress in lsg. [élipon, lpl. e[lipomen, or for that matter in "normal" s-aorists such as lsg. e[paideusa, lpl. epai[deúsamen 'instructed' or imperfects such as lsg. [élegon, lpl. e[légomen 'was saying', is completely determined by the TSR. The correspondence between lsg. PIE *élikw-o-m and Horn. ¿Upon 'I left (behind)' is hence merely due to coincidence. *
* *
(* (*
*
lik
*é
w
(*)
om
—>
(* *
*
*)
[é
lip
on
The thematic vowel retains its original lexical accent only in the non-recessive active participle (and infinitive: lip-e:n, id-e:n), as well as in the mid. iptv. 2sg. in -éo (Att. -o:) and five residual act. iptvs. in 2sg. -é, 2pl. -éte. Here alone we find complete continuity between PIE (or rather post-PIE, see §2.3.2.1) and Homeric, not only in surface realization but also in underlying specification of accent: * *
*likw
(* (*
ónt
*
*) *
>
*
(* (*
*)
[lip
ónt
a
* *
*wid
(* (*
é
*
*)
te
>
*
(* (*
[id
é
* *
*likw
*)
te
*
(* (*
*)
*
(* (*
*)
é
s(wOo >
[lip
é
o.
86
Note that the underlying accentuation of the participial endings -(o:n, -(o:sa, -(on does not extend to the present, where we find ordinary recessive \phéro:n or pai[deúo:n with stress governed by the TSR. 14 Greek therefore exhibits a synchronic contrast between unaccented pres. act. -o:n, -o:sa, -on and accented them. aor. act. -(o:n, -(o:sa, -(on. 3.3.3. Survivals of lexical accent: the aorist passive and the perfect In addition to the thematic aorist, the perfect active and passive participles and the aorist stative/passive participle also indirectly continue the underlying accent of their PIE ancestors. Let us begin with the latter. It is universally agreed that the classical Greek aorist passive in -ε:-, -thε:- (in Homer still predominantly with stative meaning), e.g. Horn. ekhárε: 's/he was joyous, happy', Att. elúthε: 's/he, it was released', continues a stative formation inherited from PEE. The latter may be reconstructed with zero-grade of the root and stress on the suffix *-ehi-, e.g. PIE *hirudh-éhi- 'to be red' > Latin rubēre. The underlying accentuation of the old stative suffix *-(ē- has been lost in finite forms such as ekhárε:, elúthε:, but survives in the ptcp. and inf.: khar-é:s, khar-é:sa, kharén '(being) joyous', lutf'-é.-s, lufl-é-.sa, lufl-én '(being) released', with accented -(ent< *-ēnt-;15 inf. k^ar-ε-.-nai, lufi-t-nai.
Below is the metrical grid for masc. ace. sg. luthénta:
* (*
[luth
ént
a.
The perfect active participle is regularly stressed on the suffix, e.g. Horn, eido'.s, iduia, eidós 'knowing' (to oida 'I know'), gegonóis, -uia, -ós 'being (X years) old' (to gignomai 'I am born'), Att. pepaideukó.s, -uia, -ós 'having instructed'. Since the stress of these forms is not recessive, we must assume that the suffix is underlying accented -(o:s, -(uia, -(os.
14
As I hope to show elsewhere, simple, i.e. unsuffixed thematic presents — which are (almost) entirely a post-PIE innovation, only one or two unclear examples being attested in the archaic Anatolian languages — had unaccented theme vowel, whereas present-forming suffixes such as *-(ye/0-. *-(ske/0- were underlyingly accented. Greek has lost the accent of such suffixes, e.g. in khairo: "go, step' < *khar-ye/0- < *ņ\-yél6-,básko: 'go, step' < *gwrņ-sk'é/ó-, so that all thematic presents in Greek share recessive stress. I5 Here *-€- has been shortened before tautosyllabic nasal within the historv of Greek by "Osthoff s Law": *V:RC > *VRC (cf. §2.2.1.2, fn. 11).
87
How may we relate this to the accentuation of the perf. act. ptcp. in PIE? Consider the PIE paradigm of the archaic unreduplicated ptcp. of *woyd- ~ *weyd- ~ *wid- 'know', which we have considered in §§2.2.1.3-4: nom. sg. ace. gen. dat.
masc. neut. *wéyd-wōs *wéyd-us (*-wos?) *wéyd-wos-rņ " *wid-us-és *wid-us-éy
fern. *wid-wés-ih2 *wid-wés-ih2-iņ *wid-us-yéh2-s *wid-us-yéh2-i
We have seen that the masc. and neut. oblique case endings (gen. *-es, dat. *-ey) are accented, whereas the ablauting root *weyd- ~ *wid- and the suffix *-wos- ~ *-us- are unaccented, as in other amphikinetic paradigms. The analysis of the feminine, with its proterokinetic contrast between stressed suffix *-wés- in strong cases vs. stressed feminine "motion-suffix" *-yéh2- in weak cases, is more complicated, and not all the details are fully clear.16 Nevertheless, it appears that at some point in the prehistory of Greek, the suffix *-wos- ~ *-us- was reanalyzed as accented, resulting in the following paradigm reconstructible for the Mycenaean stage, i.e. the 15th to 13th cc. BC: nom. sg. ace. gen. dat./loc.
masc. neut. *weydwōs *weyd-wós *weyd-wóh-a " *weyd-wóh-os *weyd-wóh-i
fern. *wid-uh-ya *wid-uh-yan *wid-úh-yā-s *wid-úh-yā-y.
Subsequent changes in suffix (substitution of *-wot- for *-woh- once *h had ceased to be a conditioned allophone of *s; cf. Szemerenyi 1967a) and ablaut (spread of e-grade to the fern., cf. Horn, iduia vs. Att. eiduia) produced the paradigm of classical Greek. As with *-(wos- ~ *-(us-, the accented ending -(enai of the pf. act. infinitive always carries the stress, e.g. in eid-énai 'to know', gegon-énai 'to be (X years) old', pepaideukénai 'to have instructed'.17 The pf. mp. ptcp. and inf. are likewise characterized by penultimate stress: Att. pepaideuménos 'having been instructed', gegramménos (< /gegraph-ménos/) 'having been written', pepaideústhai 'to have been instructed', gegráphthai (< /ge-graph-sthai/) 'to have been written' topaideúo: 'I teach, instruct (children)', grápho:
16
Cf. §2.2.1.4 on the possibility of a PIE constraint limiting stress alternations to the final three syllables of the (phonological) word. Might this restriction, evident in such ferns, as *mŗgh-éwih 2 (-iņ), *wid-wés-ih2(-iņ), have contributed to the eventual generalization of the TSR in Greek? 17 In Homer we find inf. id-men, id-menai 'to know', with the Aiolic endings (cf. fn. 10).
88
'I write'. Since PIE apparently did not have a pf. act. inf., nor any pf. mp. at all, these forms must be independent innovations — paralleled in other IE branches such as IndoIranian, but specific to Greek — and the underlying accentuation of their stems is almost certainly due to the model of the pf. act. ptcp. 3.4. PIE nominal accentuation in Greek Although it has undergone several innovations, the Greek nominal system has preserved relatively more of the original PIE system of accent. Below are the case/number endings of the noun in Homeric and Attic Greek, excluding the dual and Horn, instrumental -phi:is singular nom. ace. gen. dat. plural nom. ace. gen.
o-stems
a:/ε:-stems
consonant-stems
-OS
-a:, -ε: -a:n, -ε:n -a:s, -ε:s
-s,-0
-a:i, -ε:i
-a
-ai -a:s Horn. -á:o:n >
-es -as -(o:n
-on Horn. -oio, HomVAtt. -o: -o:i -oi -o:s -o:n
-a, -n -(os
Att. -5.71
dat.
Horn. -ois(i), Att. -ois
Horn. -ε:is(i) HomVAtt. -ais.19
-(si
Let us investigate the accentual properties of each of these classes in turn. As is well known, o- and a;/ε:-stems, along with the majority of consonant-stems, regularly I8
Originally plural and confined to non-o-stems, as regularly in Mycenaean; the o-stem ¡nstr. pi., spelled -o in Myc, probably stood for *-ōis or, with shortening by Osthoffs Law, *-ois (fn. 15; cf. Skt. -āih,Av. -āiš, Lith. -aīs). See further Morpurgo-Davies 1970, with refs. — On a:/t.--stems with nom. sg. -a, ace. -an, which continue PIE stems in *-ih2- - *-yeh2-, see §3.6. I also leave aside the vocative, which originally had initial stress no matter what the underlying accentuation of the stem, to judge from the pattern of Vedic and Balto-Slavic (cf. SC voc. iéno iTtnol to iéna /Zena/ 'woman'; cf. Schwyzer 1939:391, 547). This pattern remains fairly productive in consonant-stems: cf. páter, Sophókle:s (< *Sophó-klewes), eúdaimon with recessive stress, to nom. patc'.r 'father', Sophoklε:s (< *Sopho-kléwε:s) 'whose fame is wisdom', eu-daimo:n 'well-spirited, fortunate'. Virtually all o-stem voc. sgs. in -e have acquired "columnar accent" (see below), leaving only a few isolated relics such as ádelphe to adelphós 'brother'. On the voc. of masc. a:/ε:-stem nouns with nom. -a:s/-ε:s, cf. fn. 21. 19 On the distribution and origin of the competing dat. pi. endings in Homer, Attic, and the other dialects, cf. Schwyzer 1939:556-9, Chantraine 1942:194-6, 201-2.
89
exhibit "columnar accent": the stress of the nom. sg. is maintained throughout all other forms of the paradigm as far as allowed by the TSR, with a circumflex becoming an acute as necessary. Where columnar stress would violate the TSR, it shifts one syllable to the right: cf. [pε:khus 'arm', gen. sg. [pé:kheos (Ionic); [hé:misus 'half, gen. sg. hε:[míseos (Rix 1976:43). 3.4.1. o- and a:/ε:-steins For o-stem nouns and adjectives containing at least three syllables, there are three possible patterns of surface stress: 1. X ' X os 2. X X ' os 3. X X ós
e.g.pátríos 'paternal', thánatos 'death' e.g. oligos 'few' e.g. potamós 'river', ouranós 'sky'
Type 1 may be analyzed as accented on the antepenult, i.e. (X X os, or as underlyingly unaccented, with surface stress provided by the TSR: [X X os. Type 3 includes all oxytone, or ending-stressed, o-stem nouns; it may be analyzed as postaccenting X X( os or, if one posits an underlying accented thematic vowel -o-, as X X (os. In the absence of positive evidence that the theme vowel is accented in Greek nominal inflection, I adopt the former analysis. Type 2, however, can only be analyzed as X (X os, with penultimate accent. Among other examples of this relatively limited class, one may cite poikilos 'pointed', pedion 'plain, flat area', and the oblique/feminine stem megálo- to the adjective mégas, megálε:, méga 'big, great'. At least for this class, therefore, one must assume an underlyingly accented stem, contrasting with both unaccented (type 1) and postaccenting (type 3) stems, much as in Slavic languages such as Russian or Serbo-Croatian.20 One might likewise expect the same three patterns in a:/ε:-stem nouns — i.e., unaccented (or accented on the antepenult), accented on the penult, and postaccenting — and indeed oxytone (Type 3) a:/ε:-stems can be analyzed as postaccenting: X X( a:/ε:. Interestingly, however, Homeric and Attic lack íz:/ε:-stem counterparts to the recessive type 1 of o-stem pátrios, thánatos.
20
The distribution of this class has undergone certain alterations within the prehistory of = Greek, e.g. Wheeler's Law (" ' > , e.g. agkúlos 'crooked' < *agkulós, cf. Skt. ankurá-) or Vendryes's Law C'~ >"'< e.g. hétoimos 'ready' < hetoimos; Att. only). Here " and " denote short vowels and long vowels/diphthongs, respectively. Cf. Schwyzer 1939:379, 382-3, Lejeune 1972:297-8.
90
l.*X' 1. - A X A a: a: 2. X X' a: e.g. oikia: 'house', agápε: 'love' 3. X X á: e.g. agorá: 'marketplace', arε.té: 'virtue' The absence of unaccented a./ε.-stems is likely to be due to the weakness of phonetic evidence which would distinguish their paradigm from that of type 2. Note that with the exception of nom. pi. -ai, all the a:/ε;-stem case/number endings contain a long vowel, so that according to the TSR stress cannot recede beyond the penult. Thus, the paradigm of unstressed */oi ki a:-/ would be nom. ace. gen. dat.
sg. /oi ki a:/ > oi[ki-a: /oi ki a:n/ > oi[ki-a:n /oi ki a:s/ > oi[ki-a:s
pi/oi ki ai/ >"[oúci-ai" /oi ki a:s/ > oi[ki-a:s /oi ki a: o:n/ > oiki-[5:n (Horn. oiki-[á:o:n) /oi ki a:i/ > oi[ki-a:i /oi ki ais/ > oi[ki-ais
Just as the neut. nom./acc. sg. of participles in -o:n, -o:sa, -on did not furnish enough evidence for recessive stress (§3.3.1), so here the nom. pi. alone did not constitute a sufficient basis from which learners of Greek could deduce an underlyingly unaccented stem. Instead, they reanalyzed the nom. sg. and other forms with a long vowel in the final syllable as underlyingly accented, i.e. /oi (ki a:-/. As a result, the nom. pi. became oikiai, thus falling into line with the rest of the paradigm and bringing about columnar accent on the surface.21 *
oi
*
(* (*
*)
[Id
a:
—>
*
(* (*
*)
[oi
Id
ai
Type 2 must therefore be analyzed as X (X a:/ε:, with penultimate accent. Unlike type 2 o-stems, which are relatively limited in ancient Greek, this class includes all non-oxytone, i.e. barytone or stem-stressed a:/ε:-stem nouns. 21
This was already seen by Kurytowicz (1958:118-9); he further adduces the accentuation of the voc. sg. of a./c.-stem masculines in -a:s/-ε:s, e.g. poli.ta 'citizen', hippóta 'horseman', which must likewise be analogical to the nom. sg.: po(li:t-a, hip(pot-a after po(li:t-ε:s, hip(pot-ε:s <—po[li:t-ε:s. hip\pot-ε:s. Cf. fn. 18 on non-recessive stress in the o-stem voc. sg. The lack of recessivity in vocatives such as Lukó-phron 'wolf-minded' (for "Lúko-phron"', to nom. Lukó-phro:n) is a related phenomenon, although complicated by the cyclic computation of stress in compounds (cf. fn. 9; Noyer 1997:512-5. 522-4 and passim).
91
3.4.2. Consonant- and semivowel-stems As noted above, polysyllabic consonant-stem nouns regularly show columnar accent. The same applies to /- and «-stem nouns, which are generally recessive: pólis 'city', gen. sg. póleo:s < *pólε:(y)os (by quantitative metathesis),22 nom. pi. póle.s < *póleyes; pε:khus 'arm', gen. sg. pé:kho:s (Ion. pε:kheos) < *pε:k h ewos, nom. pi. pé:khe:s < *pε:khewes. This, however, does not imply that polysyllabic consonant-stems have necessarily given up lexical accent in all cases. For example, nouns containing the femininizing suffix -is, gen. -id-os must carry an underlying accent on the final syllable of the stem, for otherwise we should expect, e.g., "aléktoris", gen. "alektórid-os",
"Atlantis", gen.
"Atlantid-os" instead of actually occurring alektoris 'hen', gen. alektorid-os, Atlantis 'daughter of Atlas', gen. Atlantid-os (formed to aléktoir 'cock' and Atlais, gen. Atlant-os, respectively). Similarly, the numerous masculine nouns in -eús, e.g. basileús 'chief (Mycenaean, Homeric), king (Attic, Hellenistic), emperor (Byzantine)', hippeús 'knight, horseman', or Homeric names in -eus such asAkHlleus 'Achilles', must be analyzed with accented suffix /-(eu-/ ~ /-(ε:w-/.23 A convenient list of nominal derivational suffixes with their surface accentual properties is given in Kurytowicz 1958:130-7. In contrast to polysyllables, monosyllabic consonant-stems still exhibit intraparadigmatic stress alternations, e.g. nom. sg. pass 'foot', ace. pód-a vs. gen. pod-ós, dat. pod-i. This pattern is easily accounted for if we assume that the nom. and ace. endings are underlyingly unaccented and the gen. and dat. endings accented, as in the table above (§3.4). The underlying forms for /pod-/ 'foot' are thus nom. ace. gen.
sg. /pod-s/ > pó.s24 /pod-a/ > póda /pod-(os/ > podós
pi. /pod-es/ > pódes /pod-as/ > pódas /pod-(o: n/ > podo:n
22
Whence gen. sg. póleo:n< *polε:o:n by analogy; cf. Lejeune 1972:296, Rix 1976:42. The gen. pi. forms of pólis and other /-stems make up the only synchronic exceptions to the TSR, since the gen. sg. may be analyzed as underlyingly /polε :(y)os/, with quantitative metathesis still a synchronic rule. Cf. the paradigm of basileús 'king' and other ¿«-stems: gen. sg. basiléo:s, ace. sg. basiléa:, gen. pi. basiléo:n, ace. pi. basiléa.s for underlying /basilε :(w)-os, -a, -o :n, -as/. 23 0 n the prehistory of this class cf. Schindler 1976. Cf. fn. 22 on quantitative metathesis in. and the underlying forms of, the ei/-stem paradigm. 24 The lengthening in this form is irregular and must be due to (imprecise) analogy with longvowel nom. sg. forms in -o:n, -o:s, -e:s, etc. Cf. Lat. pes, whose long vowel is likewise analogical to oblique forms with short c, e.g. ace. pedem, gen. pedis: the original PIE alternation of nom./acc. *o obi. *e (cf. §2.2.1.1) has been replaced by *€ ~ *e. In Greek, leveling of *o ~ *e to *o probably took place first, followed by introduction of the innovative long vowel into the nom. sg.
92
dat.
/pod-(i/ > podí
/pod-(sí/ > post
Interestingly, not only have Homeric and Attic generalized a requirement restricting mobile stress to monosyllables, but all monosyllabic nominal stems exhibit mobile stress, i.e. are unaccented. That this pattern remains productive is proved by disyllabic stems in Homer (representative of early Attic-Ionic) which have undergone vowel contraction in Attic: PGr. neut. *ows, *owh-at > Horn.
'ear' (= [óus] or [ð:s], with o: from 'eye'; Szemerenyi 1967b, esp. pp. 65ff.), nomVacc. pi. (= [ów.wa.ta], with Aiolic treatment of intervocalic *-wh-, or [ó:ata], with o: from 'eye' for expected o:); *<3:wat-a > Att. o:t-a, whence gen. sg. *ó:wat-os (Horn. ) —> o:t-ós (cf. Rix 1976:148);25 PGr. i-stem *ówis (cf. Argolic ówis) > Horn, ois 'sheep' > Att. ois, whence gen. oi-os [ó.i.yos] —>oi-ós [oy.yós]; PGr. *páwis, *páwid- > Horn, pais 'child' > Att. pais, paid-ós; PGr. neut. s-stem *pháhos > Horn. pháos 'light' > Att. pho:s, reanalyzed as /ph5:ts/ with a new stem pho:t-, whence gen. pháo:s (/pháeh-os/) —>pho:t-ós. In each of the above cases, an underlyingly accented polysyllabic stem has become monosyllabic by contraction across an intervocalic hiatus: the new stem exhibits stress alternation between direct and oblique cases, although the original polysyllable did not. This suggests that Attic-Ionic had a synchronic constraint against accented monosyllabic stems which remained in effect at least until the above contraction had taken place.26 2
^Cf. the contrast within Homer between older dat. pi. oúasi (for *o:wasi; Iliad 12.442) and contracted o.sin (with prevocalic word-final "movable ri"; Odyssey 12.200). Hellenistic Gr. nom./acc. sg. 5:s has adopted the stem vowel of a :t-. 26 In the root aor. ptcps. dó.s, thé:s, stá.s, root-accented dónt-os, -i, -o:n, dð:si (sim. thént-os, stánt-os, etc.) are either analogical to the pres. act. ptcps. didó:s, íithé:s, hisíá.s or indirectly reflect the columnar accent of disyllabic *doat-, *theat-, *staat- < *dh3-ņt-, *d h h r ņt-, *sth2-nt- (leveled after nom.-acc. *dont-, *thent-, *stant-, e.g. dó:s < *dónts, dónta, pl. dónt-es, dónt-as < *dlvrént-; Rix 1976:234). Cf. fn. 9 on contraction across laryngeals in the opt. of these verbs. Similarly Attic d:n, gen. ónt-os is contracted from Horn., Ionic eó.n, eónt-os <— PGr. *eont- ~ *eat- < *ehont- - *ehat- < PIE *h!S-ónt- - *h,s-ņt-' (Rix, ibid., cf. §2.2.1.3; Herakleian ént-es < *ε:nt- < *eant- <— *eont- - *eat, or secondarily influenced by 3pl. *ehenti > Myc. e-e-si /ehensi/, Att.-Ion. e:si. Dor. enti'l On Horn, fern, éassa < *ehat-ya see fn. 33 below.) As far as I am aware, these participles constitute the only exceptions to this constraint; it is possible, although not necessary, that the existence of a morpheme boundary between root and suffix, e.g. do-nt-, has played a role in their accentual development (D. Ringe, p.c). The root accent of gen. sg. i:r-os, dat. -i to nomVacc. ear 'spring' (Lindeman 1972:219) poses no problem if one sets up an underlying stem /ear-/, with the sequence /ea/ subject to contraction in forms of more than two syllables (cf. en-t''o:-siá(z)o: 'be inspired by a god, inspire' vs. theós 'god' with uncontracted /eo/; Rix 1976:54). The synchronically irregular root accentuation of gen. pl. á:t-o:n,paíd-o:n vs. o:t-ós, -í, o:sí. paid-ós, -i, paist (cf. also Horn. dó:r-o:n, gó:n-o:n < *dórw-o :n, *gónw-o :n vs. do:r-ós, -/', go:n-ós, -í to dóru 'wood', gónu 'knee') is an archaism, dating from the period prior to the contraction and resulting
93
3.5. Implications for the phonological prehistory of Greek Is it possible to draw any inferences about the relative or absolute chronology of the changes involved in the development of the Greek accentual system? Fortunately, a handful of isolated relic forms survive from older stages of the language and allow us to trace, if only approximately, the evolution of the productive accentual rules and restrictions of Greek of the 1st millennium BC. Hoenigswald (1997) has argued that Horn. kε:ri 'at heart', an old locative to the fossilized neuter noun kε.r 'heart', represents the sole survival of an earlier stage of Greek in which the dat. sg. ending -i — originally the PIE locative sg. ending *-i — was ««accented. This concurs with our current understanding of PIE nominal morphology, according to which the most archaic type of PIE locative was "endingless", and *-i was subsequently added (see §2.2.1.3). At some point in the prehistory of Greek, loc. (—> dat.) sg. -/ acquired underlying accent, becoming -(i. This accent must have been analogical, not only to gen. sg. -(os, but almost certainly also to the older PIE dat. sg. *-(ey which survived into Mycenaean (e.g. dat. e-me /hem-éy/ 'one' vs. Horn., Att. hen-i with stem-final -n- from nom. *hens —> he:s): as *-i and *-ey have undergone functional syncretism already in Mycenaean, with -e /-ey/ preponderant in ¿-stems and -i in other consonant-stem classes, one would expect these two endings to have influenced each other. Unless -/ remained unaccented until after the complete loss of *-(ey — which is highly unlikely — it is probable that the combination of gen. -(os and dat. *-(ey led speakers of Greek to associate underlying accent with oblique case and so bring originally unaccented loc. *-i into line with this pattern.27 "monosyllabization" of the stem. As observed by Lindeman (1972:218-9), forms such as o-stem gen. pi. lúko:n,oíko:n (to lúkos 'wolf, oikos 'house') would have provided more than ample support for 5:to:n,dó:ro:n, so that the latter could survive as such, reinterpreted as underlying /o:t-D:n/, /do:r-o:n/ with unaccented stem and unaccented (properly o-stem!) gen. pi. /-o:n/. Note also pa:s 'all (m.)\ which stresses the ending in gen. and dat. sg. but the root in both gen. and áat. pi.: pantos, panti \s. panto :n, pa: si (the circumflex of pð:s < *pánts is unexplained, as is that of hē:s < *héns 'one'; Lejeune 1972:296). I tentatively propose that this adjective is a fossilized pres. ptcp. to PIE *peh2- 'guard, take care of (hence ""watched, guarded, kept' > ""whole, entire' > 'all'; cf. Hitt. pah(has)s- 'protect', OCS pasg 'watch over, tend' < *peh2-s-), which has otherwise been lost in Greek: nom./acc. *ph2-ént- (or perhaps rather *péh2-ņt-?), gen./dat. *ph2-ņt-' > *pant- - *paat—> pánt-, as in dónt-, etc. above. (Cf. TB, TA pont- 'all' < PT *pont- < *pant- [Ringe 1996:23, 75], which Penney [cited in Adams 1999:402] suggests may be from *péh2-rjt-.) Could the complete isolation of pant- have led to its partial assimilation to the regular accentual pattern, in the sg.? 27 I wish to thank an anonymous reviewer for the PWPL for clarifying this argument. The accentuation of *-i in Greek is therefore independent of the parallel development in Sanskrit: cf. Ved. murdhán-i vs. mUrdhn-t, after gen./abl. mvrdhn-ás, dat. -é, instr. -ð (cf. §2.2.1.3).
94
The other probable accentual archaism, also discovered by Hoenigswald (1987), is Horn, aiei, Att. a:ei 'forever'. If, as Hoenigswald argues, these forms continue *ayyehi < (post-)PIE locative *h2eywes-i to an .y-stem *h2éywos (cf. Dor. aiés 'id.'), their final accent marks them as a unique relic, a "moderately ancient locative dating from the time after oxytonesis had been generally extended to oblique cases including the locative, but before it was limited to monosyllables" (51). These considerations lead to the following relative chronology of developments in the nominal accentuation of Greek:
Note that even in contracted monosyllabic stems with stem-stressed gen. pi. (a':t-o:n,paid-D:n) or gen. and dat. pi. {pánt-o:n, pá.si; cf. fn. 26), the gen. and dat. sg. always have stressed -ós, -í. Thus the gen. sg. is the only oblique case form which is always ending-stressed in Greek, without exceptions or relics.
95
loc. pi. *-(su —> *-(si after loc. sg. *-i Horn. Gr. kε:ri
i
sg. loc.28 -i —> -(i after dat. -(ey, gen. -(os
mobile accent extended to all paradigms v /
*ayw-es-(i > *ayy-eh-(i > Horn, aiei, Att. a:ei
I
mobile accent restricted to monosyllables and r-stems (productive pattern in Attic-Ionic)
1
end of composition of Homeric epics
1
Horn, oús, oúat-a, -os 'ear' —> Att. oiis, 3:t-a, o:t-ós Horn, ois, oi-os 'sheep' —> Att. ois, oi-ós Horn. pháos, pháo:s 'light' —> Att. ph5:s, pho:t-ós 3.6. More innovations in Greek nominal accent: continuity in disguise? As demonstrated in §3.4, Greek nominals have retained significantly more of the inherited PIE accentual system than the verb. Nevertheless, certain nominal classes have undergone significant innovations which have not yet received a satisfactory explanation in previous scholarship. Two of these are discussed below. Like their animate counterparts, monosyllabic neuter nouns are regularly unaccented and exhibit alternating stress: cf. nom./acc. pur 'fire', gen. pur-ós, dat. pur-i, like aiks 'goat', ace. aig-a, gen. aig-ós, dat. aig-i. As has long been noted, however, athematic neuter nouns, i.e. neuters other than o-stems ending in -on, of more than one syllable regular exhibit recessive stress in Greek: /•/«-stems, e.g. húdo:r 'water' < *udōr <— PIE coll. nom./acc. *wédōr, gen. húdatos < pre-Gr. *udņ-t-os <— PIE coll. obi. *ud-n- (cf. Hitt. coll. widār < *wedōr <— *wédōr; sg. wātar < *wód-ŗ, gen. wedenaš <— PIE *wéd-n-s; Schindler 1975a:3-7); ¿¿-stems, e.g. dóru 'wood', gónu 'knee' < PIE *dóru, *gónu (but cf. fn. 26);
28
0r dat., if functional syncretism of *-i and *-ey had occurred by this time.
96
s-stems, e.g. génos < PIE *génh r os (< pre-PIE *dénhļ-s), gen. géno.s < *genesos 'birth, race, race, kind' < PIE *dénhi-es- (< pre-PIE dņhi-és-), kréas 'flesh, meat' < *krewas < PIE *kréwh2-s (Schindler 1975b); n-stems, e.g. ónoma, gen. onómatos 'name' <— ónuma (cf. Horn. nd:num(n)os 'nameless') <— enuma* (preserved in Laconian Enuma-kratída:s) < *enomņ29 < *hiņh3m-ņ- (synchronically in the following class; cf. §2.2.1.2); /-stem (< n-stem) nomina actionis ("action nouns") in -ma, e.g. so:ma, gen. sóimatos 'body'. Whereas the recessive stress of (nearly all) finite verb forms reflects the generalization of the postlexical destressing rule found in main clauses in Vedic Sanskrit (§3.3.1), and that of vocatives continues the initial stress of vocatives in PIE (cf. fn. 18), this pattern seems to have no historical basis: as Hoenigswald (1998:272) observes, no related phenomena are found elsewhere among the ancient IE languages.30 However, an examination of the archaic accent-ablaut paradigms reconstructed for PIE reveals that the oldest stratum of neuter nouns always carries stress on the root in the nominative-accusative singular. As reconstructed by Schindler (1975a and especially 1975b:262-4), PIE athematic neuter nouns may belong only to either the aerostatic or proterokinetic inflection, i.e. with fixed stress on the root or stress alternating between root and suffix, respectively (cf. §§2.2.1.2, 2.2.1.4). aerostatic, fixed stress on root nom./acc. *wód-ŗa 'water' gen. *wéd-ņs
*hinēh3m-ņ 'name' *hjnéh3m-ņ-s
proterokinetic, stress alternating between root and suffix nom./acc. *genhi-(o)s 'kind, race' *X'-mņ (> Gr. -ma in nomina actionis) gen. *gehi-és-(o)s *X-mén-s Now if we assume that the initial stress of the "strong", i.e. nom.-acc. variant was generalized at some early stage of pre-Greek, all neuter nouns (other than o-stems) would
29
Cf. Vine 1999b:557-8 and passim on the precise conditioning of the rounding of *o > u in this and other Greek forms ("Cowgill's Law"). 30 Athematic neuter adjectives also appear to have generalized recessivity in prehistoric Greek, although the effects of columnar accent have largely leveled this out. Cf. álε:thes 'really?", khárien 'nice!', fossilized neut. nom./acc. sg. of alε:^é:s 'true', kharie:s 'beautiful, elegant' (vs. regularized alε:thés,kharíen\ Hoenigswald 1998:272), although the second of these, like eléε.mon 'full of pity', béltion, kállion 'better' (to eleé:na:n, beltío:n, kallh:n), may merely reflect an underlying unaccented stem of the type which has been eliminated in pres. and fut. act. ptcps. (cf. §3.3.1, fn. 10). On Att. ei~i 'well' < *éu vs. Horn, eus, eti, cf. Hoenigswald 1998:272-4.
97
have ended up with fixed stress on their initial syllable —just like vocatives.31 Subsequent application of the TSR then resulted in the observed recessive stress. Another class of nouns that regularly exhibits recessive stress in Greek is feminines with nom. sg. in short -a. In contrast, the cognate Vedic class of feminines in -I with obi. forms in -yd- (< PIE *-ili2- ~ *-yeh2-), although preponderantly barytone, also includes oxytone stems, e.g. devl 'goddess', after which the inflection is named in traditional grammars.32 Note, however, that Greek feminine adjectives in -a are often accented on a preceding suffix: bareia, f. to w-stem barús 'heavy'; iduta, f. to pf. act. ptcp. eido:s 'knowing'; lipo.sa, f. to aor. act. ptcp. lipa:n 'leaving (behind)'. These patterns may be explained, I believe, if we suppose that the same generalization of nom.-acc. stress to oblique case forms just proposed for athematic neuters also occurred in iT^-stems. Since this class, like e/12-stems (> Gr. a:/ε.-stems), was proterokinetic in PIE, the "strong" cases would have carried stress on the root, while the "weak" cases would have had stressed full-grade of the ending, i.e. nom. *X'-ih2-0, ace. *X'-ih2-iņ vs. gen. *X-yéh2-s, dat. *X-yéh2-i. If the nom. and ace. were reanalyzed in pre-Greek as having underlying accent on the root, generalization to the oblique cases would have yielded a columnar paradigm. The development of the ¿z-stem pa:sa 'all, every' < *pánsa (cf. Arcadian pánsá) < *pántya < *pānt-ili2 (cf. fn. 26) would thus have been nom. *pānt-ih2 ace. *pānt-ih2-iņ gen. *plnt-yéh2-s
—> *pānt-yeh2-s
31
> *pánt-ya > *pánt-yan > *pánt-yās
> pá. sa >pð.san > pá:sε:s
In the case of pur 'fire' <— PIE r/n-stem *péh2-wŗs, *ph2-wén-s (Hitt. pahhur, gen. pahhwenaS; cf. Schindler 1975a:7, 9-10), which has become monosyllabic within the prehistory of Greek, this initial accent has been eliminated by the constraint that monosyllabic noun stems must be underlyingly unaccented (§3.4). 32 In Vedic, a large class of about 70-80 oxytone derivatives in -J are inflected like root nouns in -7 (e.g. dhlh 'thought'), i.e. with non-ablauting stems in -I- - -iy-: hence vŗkíh 'she-wolf, gen. vrkfyah, kalyāņíh 'fair woman', purusíh 'woman', Yamlh (to vfka-, kalyāņa-, púrusa-, Yamá-, respectively), laksmí 'mark', m. rathí 'charioteer', m. ahí 'serpent' (cf. older /-stem áhi-, Gr. óphis 'snake'); cf. Macdonell 1910:268-70, Whitney 1924:128 (§355b). Interestingly, Whitney (1924:132 (§362, 2d)) notes that Vedic ferns, in -I which show accent shift from the corresponding masc, e.g. távisl 'might', párusņl 'reedy (name of river)', páliknī 'gray', róhiņī 'ruddy cow', follow ¿evl'-inflection only when not oxytone, in contrast to vŗkf, etc. Oxytone stems of the devl-lype are either analogical to the corresponding masc, e.g. devl itself to m. devá- 'god', or result from remodeling of the inherited stem stress, e.g. it-stem ur-v-í 'wide', pres. ptcp. ad-at-í 'eating' (m. urú-, adánt-; cf. Macdonell 1910:273 and see §2.2.1.4 and fns. 16, 33 above). By classical Skt., the vŗkí class was almost completely assimilated to the devl type; the resulting paradigm took nom./acc/voc. du. -íx-ð(tt), nom. pi. -ix-ah from vŗkí inflection (Whitney 1924:128-9, 132-3).
98
dat. *pānt-yéh2-i
—> *pānt-yeh2-i
> *pánt-yay
> pá:sε:i 33
Evidence for this hypothesis comes from the one a-stem that preserves mobile accent, i.e. mia 'one (f.)': nom. mia ace. mían gen. mia:s dat. mia:i
< *smiya < *smíyan < *smiyá:s < *smiyá:y
< *sm-ih2 <— *sém-ih2 < *sm-fli2-iņ <— *sém-ih2-m < *sm(i)-yéh2-s < *sm(i)-yéh2-i.
The accentual alternation of this form, which is entirely isolated within Homeric and Attic Greek, can only be explained by assuming that — at least at an earlier stage — the oblique endings of a-stems were underlyingly accented, just as for the consonant-stems: gen. sg. -(a:s, dat. sg. -(a:i. Why the stem mi- remained unaccented, in contrast to all other ci sterns, is somewhat less clear. Note that unlike the development proposed above, by which the accent of the nom./acc. was generalized to the oblique cases, 'one (f.)' has apparently generalized both the stem accent and zero-grade root ablaut of the oblique cases to PIE *sém-ih2(-iņ); the motivation for this divergent treatment remains unclear to me.34 33
Note that according to this view, the PIE inflection of fem. «-stems and perf. act. ptcps. proposed in §2.2.1.4 explains the non-recessive accentuation of these forms in Greek, e.g. nom. *mŗJáh-éw-ih2 ace. *m$h-éw-ih2-iņ gen. *mŗsgh-u-yéh2-s dat. *mŗ)gh-u-yéh2-i nom. *wid-wés-ih2 acc. *wid-wés-ih2-iņ gen. *wid-us-yéh2-s dat. *wid-us-yéh2-i
—> *m$h-éw-yeh2-s —> *mŗgh-éw-yeh2-i
> > > >
*brakh-éw-ya *brakh-éw-yam *brakh-éw-yā s *brakh-éw-yāy
> brakheīa > brakheian >brak''eía:s > brakheía:i
—> —> —> —>
> > > >
*wid-úh-ya *wid-úh-yan *wid-úh-yās *wid-úh-yāy
> (w)iduia > (w)iduian >(w)iduía:s > (w)iduía.i.
*wid-ús-ih2 *wid-ús-ih2-iņ *wid-ús-yeh2-s *wid-ús-yeh2-i
The same should have occurred in the archaic pres. ptcp. fem. of *h!es- 'be': *hļS-ént-ih2 , gen. *h,sņt-yéh2-s > *ehént- ~ *ehat-' —> PGr. *ehátya > "eás(s)a", eássε.s (cf. Skt. sail <— *sánt- - *sat-'; §2.2.1.4). Could Horn, éassa (gen. eássε.s by the TSR) owe its recessive accent to other ferns, in -ssa, e.g. thálassa 'sea', mélissa 'bee'? On the other hand, f. pi.e.ra 'fat' <— PIE *pih2-wér-ih2 (Skt. pívarl) for expected "pi.é.ra" can have its recessive stress from m. pi:o:n, itself from amphikinetic PIE *péyh2-wōn, gcn. *pih2-un-es (Skt. pfvan-; §§2.2.1.3-4). 34 Monosyllabicity alone cannot explain the unaccented status of mi-, as other monosyllabic a-stems exhibit no accentual alternation, e.g. nom. sg. f. pá.sa 'all', gen. pá:sε:s above or moira 'fate' (< *'that which is measured') < *smórya < *smór-ih2 (to the verbal root *smer- 'measure', cf. Horn. pf. émmore "shares, has as share' < *e-hmor-e with Aiolic treatment, plupf. mp. [hje.marto 'was ordained' < *e-hmar- < *e-smŗ-), gen. sg. moira:s. Has mi- secondarily become unaccented by analogy to masc./neut. hen- 'one"? Or has this root merely preserved the inherited proterokinetic pattern despite remodeling of nom. and ace, thanks to the development of *ih2 > *iya after the initial cluster *sm (PIE *sém-ih2(-m) —> *sm-íh2(-iņ) > *smíya(n) > mia, mían)l
99
Although the hypotheses proposed in this section remain tentative, note that they depend crucially on the assumption that pre-Greek did inherit the accentual patterns and distributions of PEE, e.g. the apparent restriction of ablauting neuter nouns to aerostatic and proterokinetic inflection, or the proterokinetic inflection of ifi2-stems. Despite recent claims to the contrary, then, full consideration of the IE background of Greek is necessary for understanding the many synchronic idiosyncrasies of Greek accent, and the extent to which the language has preserved and modified the system inherited from its PIE ancestor: in other words, the ancient Greek accentual system directly continues that of PIE, albeit heavily obscured and altered by a series of far-reaching innovations. Finally, the discussion in this chapter demonstrates that stress in Homeric and Attic Greek may be computed by means of the same parameter settings as for Balto-Slavic languages or Sanskrit (cf. § 1.2.3), thus making it all the more likely that these settings were inherited from, and operative already in, the accentual system of PIE.
100
Chapter 4. From Proto-Indo-European to Balto-Slavic: Dybo's and Hirt's Laws reconsidered 4.1. Introduction: some methodological considerations The past generation has witnessed a remarkable explosion of scholarship on the perennial problem of the historical reconstruction of Baltic and Slavic accentuation. Among recent major works of the last 25 years, one may mention Garde 1976a, Dybo 1981, Dybo, Zamjatina, and Nikolaev 1990, as well as Zaliznjak's 1985 history of Russian accentuation and the articles of Kortlandt (e.g. 1975, 1978, 1988) and Stankiewicz 1979a, 1986c (cf. also the historical remarks in Stankiewicz 1993:1-38, 329-51). As a result of these studies, we now have a far more precise understanding of the reconstructed accentual systems of Proto-Balto-Slavic and Proto-Slavic and their development into the accentual systems of the medieval and modern Baltic and Slavic languages. Yet much research on BS1. accent during the 20th century has largely ignored or downplayed the IE background of these languages, limiting its scope to what may be determined from the Baltic and Slavic evidence alone. In the decades following the appearance of van Wijk (1923), few studies made a serious attempt to incorporate the data of Vedic Sanskrit, ancient Greek, or other IE branches with distinctive lexical accent into their accounts. Illie-SvityC's 1963 monograph,1 which attempted to connect the nominal accentual classes of Baltic and Slavic to those of other IE branches, is a notable exception to this pattern, and occasional references to PIE phonology appear in the works of Dybo and his students. More often, however, investigation of the prehistory of BS1. accent has proceeded in a vacuum separated from the larger picture of IE, with negative results for both Balto-Slavicists and general Indo-Europeanists: scholars of BS1. accent, particularly but not exclusively in eastern Europe, have at times betrayed a lack of familiarity with many of the recent advances in — or even fundamentals of! — the reconstruction of the IE noun and verb, whereas Indo-Europeanists not specializing in BS1., understandably intimidated by the complexity of the entire topic, have by and large remained in the dark about the work of scholars such as Garde and Dybo. BS1. accentology, and to some extent BS1. in general, has unfortunately become relegated to an exclusive, "specialists-only" subfield, its forbidding reputation dissuading outsiders from examining the facts and Hereafter cited from the 1979 English translation; the original was not available to me.
101
scholarly literature for themselves and drawing connections with the rest of IE historical linguistics. In this and the following chapter, I investigate the PIE origin of some features of BS1. accent, proceeding from three fundamental principles. First, Balto-Slavic must not be treated as a linguistic isolate for the purposes of accentology: rather, proper reconstruction of Proto-Balto-Slavic must systematically take into account all available evidence from other ancient IE languages with lexical accent. In addition to the standard methodological practice of reconstructing backwards from the modern languages and their older records (e.g. Old Lithuanian, Old Russian, or Middle Bulgarian, as well as Old Prussian), we ought also to reconstruct forward from the PIE system, postulated mainly on the basis of the generally clear evidence of Sanskrit, Greek, and Hitite and outlined in chapter 2. To be sure, only about 12 centuries separate contemporary Slavic from its relatively uniform Proto-Slavic (PS1.) ancestor, whereas the latter had already been diverging from PIE for at least 4000 years, given the currently accepted view that PIE began to break up around the mid- or late 4th millennium BC (cf. Cowgill 1986:13). For Baltic, of course, our earliest information on accent, from the Old Prussian Third Catechism and Dauksa's Old Lithuanian Postilla, does not predate the early modern period. Nevertheless, one may argue that even a relatively recent achievement of IE historical reconstruction such as the archaic accent-ablaut types of nominal and verbal stems (cf. §§2.2.1, 2.3.1) is in many ways better understood, and enjoys wider scholarly acceptance, than the prehistory of BS1. accent. Second, one must adopt the starting hypothesis that the BS1. system of underlying accented and unaccented morphemes (on which see below) historically continues that of PIE. The only alternative is to assume that the two types of contrasting accentuation on BS1. stem syllables, suffixes, endings, etc. are entirely a creation of the post-PIE period, in which case, nothing can be gained from a comparison with other IE languages and the reconstructed protolanguage.
Such a negative view suffers from the scientifically
objectionable drawback of being in principle unprovable: once one either assumes or concludes that BS1. accentual specifications are independent of those of Sanskrit, Greek, and Germanic, any correspondences between BS1. and extra-BSl. forms may be dismissed as coincidence. As the greatly reduced number of PIE word equations involving BS1. makes a statistical analysis impossible, any study which aims to elucidate the accentual prehistory of BS1. must take into account the most securely reconstructed aspects of the prosodic phonology and morphology of its PIE ancestor. 102
Finally, any historical investigation of Balto-Slavic accent must concentrate in the first instance on reconstructing the prehistory of nominal or verbal classes, e.g. o- and istems or simple and derived thematic presents — and the underlying accentuation of stems and inflectional endings — from the PIE preforms reconstructive principally on the basis of Sanskrit, Greek, and Hittite. Such an approach is necessary because the BS1. languages have undergone massive lexical replacement of PIE nominal forms with derivatives or new items; consequently, as will shortly become clear, there exist relatively few reliable word equations for a comparison of nominal or verbal stems with direct cognates in other IE languages. I follow the by now generally accepted hypothesis that Baltic and Slavic share enough distinctive innovations to warrant the reconstruction of a common post-PIE ancestor, namely Proto-Balto-Slavic (cf. Szemerenyi 1957, contra Senn 1966b). Obviously, any sufficiently idiosyncratic innovation common to Baltic and Slavic, as against the rest of IE, should tentatively be ascribed to the PBS1. stage. Conversely, if a phenomenon in Baltic and/or Slavic finds parallels in non-BSl. IE languages, we should seriously consider the possibility that it continues an older, (post-)PIE feature. To be sure, many individual components of the PIE system inherited by PBS1. have been obscured by the accretion of a series of prehistoric innovations — not surprising, considering the relatively late date of attestation of accented Baltic and Slavic texts (cf. §4.2.1). Yet as I hope to demonstrate below and in chapter 5, Balto-Slavic, like Greek, has retained far more of its PIE accentual inheritance than ordinarily assumed, by either BaltoSlavicists or Indo-Europeanists. 4.2. Background It is not possible here to give even a brief overview of research to date on BaltoSlavic accentology; for concise summaries, cf. Derksen 1991, largely a recapitulation of Kortlandt's writings, and Lehfeldt 1993, which conveniently presents the major findings of the "Moscow Accentological School", represented by the Soviet scholars V. M. IllieSvityð and V. A. Dybo and their students. The latter have contributed enormously to our understanding of the PS1. system through a combination of in-depth philological analysis of data from modern dialects and medieval manuscripts, internal reconstruction, and external comparison, particularly with Lithuanian and Latvian but also with other IE languages; their findings have been accepted to various degrees among the small number
103
of Western scholars active in BS1. accentology since Stang, from Ebeling (1967) to Koch (1984) to Halle (1997:297-8, 2001).2 In this and the following chapter, I have adopted Dybo's system as a starting point for discussion. The reconstruction of the accentual evolution of BS1. proposed here differs from that of Dybo, IlliC-SvityC, Halle, and other BS1. accentologists, not so much in its evaluation of the modern languages and dialects and medieval sources as in the historical interpretation of the various accentual shifts, paradigmatic alternations, and morphological patterns which have shaped the prosodic systems of (pre-)PBSl. and PS1. As will become clear, these differences of interpretation stem largely from a difference of historical perspective, namely from an attempt to derive the Balto-Slavic accentual systems from that of PIE, as reconstructed in chapter 2 on the basis of other IE languages. 4.2.1. The evidence Within Baltic, most of our information comes from the modern Lithuanian literary language, along with the various dialects, of which some — in particular the Zemaitian dialects of the Baltic Sea coast — preserve a number of archaic features. As the first documents in Lithuanian date only from the 16th c , these do not differ from the contemporary language in phonological essentials, on which cf. §4.2.2. Our knowledge of the accentuation of Old Lithuanian comes principally from the Postilla of DaukSa in 1599; from this and other accented texts, as well as from peripheral modern dialects such as eastern AukStaitian, Zemaitian, or the extinct Lithuanian dialects of East Prussia, it is often possible to distinguish innovations in the standard language and the southern and western Aukstaitian dialects upon which it is based. Foremost among these innovations is the gradual shift of many nouns and most adjectives with underlyingly accented stems to the unaccented accentual paradigm, i.e. from classes 1/2 to classes 3/4 (cf. §4.4.1). Latvian has restricted surface stress to initial syllables, where long vowels and diphthongs contrast three possible intonations: rising (~), falling f), and "broken" or glottalized (A). The historical origins of these will be reviewed below in §4.2.3. The accentual system of Old Prussian is known only from a single document, the Enchiridion or Third Catechism of Abel Will, dating from 1561 and containing a number of Protestant lessons and prayers with parallel German translation. In this text, macrons are often, although not consistently, printed over stressed long vowels; when one of the diphthongs -In contrast, Kortlandt's conception of BS1. historical accentology has not in general found followers outside of Leiden.
104
ai, au, ei is stressed, the macron is printed over the first element if the diphthong has falling or circumflex intonation, and over the second element if the diphthong has rising or acute intonation, hence ai, au, el vs. āi, āu, ēi. In diphthongs composed of vowel + sonorant, macrons could appear only over the first vowel, hence only when the diphthong was circumflex; this is no doubt due to the limitations of Abel Will's printer, since there is no reason to doubt that OPr. also had acute aJ*, /r*, etc, continuing PBS1. *ál, *ír, etc.3 As we possess no documents in Proto-Slavic, and the oldest preserved (Old Church) Slavic manuscripts do not indicate the position of surface stress or intonation, our reconstruction of Proto-Slavic accentuation rests on the evidence of medieval manuscripts marked for accent, and the modern languages and dialects. The former category includes Old Russian documents, beginning with the Cudovskij Novyj Zavet (New Testament of the Cudov closter in the Trinity-St. Sergius Monastery outside Moscow), the oldest accented Slavic manuscript; and Middle Bulgarian texts, mostly representing the eastern Bulgarian dialect area of Veliko Tarnovo, seat of the Bulgarian Orthodox patriarchate under the Second Bulgarian Empire of Tsar Ivan Asen in the 14th c. The modern Slavic languages and dialects provide three types of information relevant to the reconstruction of the prosodic system of Proto-Slavic, namely length, position of stress, and intonation. Distinctive length is preserved in Czech-Slovak and western South Slavic (Slovene and Serbo-Croatian), and was present in Old Polish; modern Polish maintains reflexes of vowel length in the vowels ó [u], q [5] < OPol. *ō, *ņ.4 The East Slavic languages and eastern South Slavic (Bulgarian, Macedonian, and the Torlak dialects of southeastern Serbia) lack phonemic length, and apparently lost PS1. length at a relatively early date. Contrastive stress placement is found in East Slavic and most of South Slavic, the only exceptions being certain transitional dialects in and near Macedonia with fixed stress on either the final, penultimate, or antepenultimate; the best known of these are the dialects 3
The communis opinio holds that geminate consonants indicate that the preceding short vowel is stressed; in contrast, Kortlandt (1974a) has proposed that it is the following vowel which is stressed. As far as I can tell, both of these views are mistaken. Geminate spelling indicates not the position of stress, but the shortness of a preceding vowel, as one might expect from German orthographicpractice, which of course served as the model for OPr. Cf. widdewu 'widow', dat. pi. widdewumans, i.e. [videvú:], [videvú:mans], with geminate dd marking a short preceding [i]. 4 OPol. *ņ, *ņ continue PS1. *e_, * and "neoacute" *ej, *Q, respectively; the two PS1. nasal vowels merged in early Polish, but the short and long counterparts developed contrasting qualities, resulting in modern q, q [q]. Thus the two nasal vowels of contemporary Polish do not correspond directly to the two of PS1.: cf. lsg. -e < *-t¡ < PS1. *- vs. 3pl. -g < *-q < PS1. *-Qt(i), generalized from presents of AP c. See further Klemensiewicz et al. 1965:102-11.
105
of northern F.Y.R. of Macedonia around Skopje with antepultimate stress, upon which the literary Slavic Macedonian language is based. In West Slavic, only two peripheral areas direct attest contrastive stress: Slovincian, the archaic and northwestern dialect of Kashubian, extinct since the 1940s; and Polabian, spoken on the Elbe in north central Germany and documented in 18th-century texts, shortly before it too became extinct. The remainder of West Slavic, by contrast, has generalized fixed stress, on the initial syllable in Czech and Slovak and on the penult (with certain morphological and lexical exceptions) in Polish. As for intonation, the only Slavic languages to preserve surface intonational contrasts on long vowels are the conservative Carniolan dialects of Slovene and the Cakavian, kajkavian, and most štokavian dialects of Serbo-Croatian, to retain the inaccurate but still commonly cited traditional three-way classification of South Slavic dialectology. It is commonly assumed that the acute and circumflex intonations of these dialects in some way continue those reconstructed for Proto-Slavic and Proto-Balto-Slavic and attested in Lithuanian. In fact, however, neither Slovene nor the archaic northwestern SC varieties have preserved the PS1. intonations as such: the contrast of acute and circumflex on long vowels in the modern languages and dialects is of secondary, post-PSl. origin. In the case of Slovene, PS1. unaccented forms or enclinomena (cf. §4.2.3) come to stress the second vowel of the phonological word; this vowel receives circumflex intonation, e.g. in goro 'mountain (ace.)', na goro 'to the mountain' (Russ. góru, ná goru, SC goru, ná goru). All other long vowels in Slovene receive acute intonation, either on the same vowel as in PS1. or, if the stress in PS1. fell on the final syllable, the preceding long or short vowel (except /a/): cf. Garde 1976a:253-62. The modern Slovene intonations, then, do not directly continue the PS1. acute and circumflex, but rather reflect the contrast between accented and unaccented phonological words in PS1. In the ðakavian dialects of Serbo-Croatian, today confined to Istria and the islands along the Dalmatian coast of Croatia (with a few pockets on the mainland, e.g. Novi, Senj) but formerly spoken across the western part of the Serbo-Croatian-speaking territories of pre-1991 Yugoslavia, long vowels may carry either acute (') or circumflex (A) intonation: cf. the descriptions of the dialect of Novi Vinodolski, on the coast north of Senj (Belie 1909), and of Orlec on the island of Cres (Houtzagers 1985). Yet as Garde (1966:47-8, 1976a:218-20, 244-7) has noted, the contrast between the two is not underlying: rather, surface intonations are actually contours across moras. Compare the paradigms of Russ.
106
borodá, ðak. brādá, which belongs to the mobile class, AP c (the Cakavian data are from Novi; Belie 1909:228).5 nom. sg. ace. sg. nomVacc. pi. gen. pi. dat. pi.
Russian borod-(a borod-u borod-y borod-(0 borod-(amO
borod-á bórod-u bórod-y boród-0 borod-ám
ðakavian brðd-(a brðd-u brād-e brad-(a brad-an(a
brād-áa brád-u brád-e brád brād-án
In unaccented forms such as the a-stem ace. sg. or nom./acc. pi., we find circumflex intonation in Cakavian: the default initial stress of Russian bórodu, bórody here falls on the first of the two moras which make up the long vowel a:, hence bráadu or brádu. Similarly for the nom./acc. pi.: Russ. bórody, ðakavian bráde, i.e. bráade. By contrast, when stress is retracted from the underlyingly accented final jer in the gen. pi., it is retracted to the preceding vowel in Russian, hence boród, but to the final mora in Cakavian, hence braád or brád. Similarly, Stang's Law, which retracts stress from a word-internal circumflex long vowel (i.e. one not in the final syllable) to the preceding vowel, yields rising intonation in Cakavian if the preceding vowel is long: hence the presents of moc and pīsát are morēn, mores, etc. (with -r- for -2-) and pfífēn, píšeš, etc., respectively (Belie 1909:241, 247-8).6 All intonations in Cakavian thus appear to be surface realizations of stress placement on a word-initial or word-final mora; one never finds, say, a contrast between acute and circumflex on internal long vowels, such as one would expect if intonation were in fact phonemically distinctive in the language (on the "neo-circumflex" see fn. 12). This contrast on long vowels has been lost in the majority of ¡Stokavian dialects (except for Posavian, see below). The famous system of rising and falling intonations on both long and short vowels in the standard stokavian Serbo-Croatian language(s) is the ^Cf. §§1.2.4, 5.1 for the computation of surface stress in these forms. The nominal stems in these mobile paradigms are underlyingly unaccented: cf. §4.2.3 for a discussion of accentual classes in Slavic. On the underlying forms of the Russian and cAakavian endings, cf. §5.1. Note that standard SC has remodeled the gen. pi. to /-s-ā/, with an -ā of uncertain origin (cf. the discussions in Leskien 1914:434-7, Stankiewicz 1978), and generalized -a-ma in the dat., instr., and loc. pi.; the older endings, dat. -am, instr. -ami, loc. -ah, are preserved in the kajkavian dialects of northern Croatia proper (vicinity of Zagreb) and in Cakavian, as well as in the Dalmatian lectionaries of the 15th and 16th c. (cf. Leskien 1914:418ff., Browne 1993:382-4). 6 This retraction produces an innovative stress alternation in postaccenting verbs, i.e. verbs of AP b, whose presents originally had columnar stress on the thematic vowel, e.g. Russ. mogú, mótel' 'I, you can', piSii, píšeš' 'I, you are writing', standard (stokavian) SC mogu, moieš, (pišēn), pišēš (cf. §§4.3.2, 6.6). Retraction with rising intonation on a long vowel is also found in "neo-mobile" plurals of AP b, i.e. formerly postaccenting nouns (§5.1.1). On the question of neo-acute intonation on short vowels, cf. §§4.2.2, 6.3.1.
107
result of a subsequent retraction of stress one syllable to the left in the east Hercegovinian dialect native to Vuk Karadzic, upon which the standard Serbo-Croatian language(s) is/are based. This so-called "štokavski shift", which is in fact a spreading of high tone to the preceding mora (as phonetic investigations of recent decades have demonstrated; for phonological analyses of Serbo-Croatian pitch and stress, cf. Ivic 1976, Garde 1976a: 17, 250-1, Inkelas and Zee 1988, Bethin 1994, 1998:162-8, 239-43, Langston 1997), is responsible for the so-called "rising" and "falling" intonations on short vowels, which in fact are determined by whether the following syllable head has high or low tone. Nevertheless, the underlying placement of stress remains unaffected and the morphological stress alternations of SC may be modeled in the same way as, e.g., those of Russian, as Browne and McCawley showed in their still invaluable 1965 article.7 To repeat, the intonational contrasts of SC have no direct connection to the reconstructed PS1. intonations, but rather continue the distinction between rising "neoacute", on long vowels which have received stress as a result of the loss of a following stressed weak jer or word-internal circumflex long vowel (Stang's Law), and falling "circumflex" elsewhere. This contrast, which may be analyzed synchronically as a difference in stress placement on the first vs. second mora of long vowels, is preserved in the ðakavian dialects (e.g. of Nov! and Cres) as well as in the conservative Posavian dialects of Slavonia, traditionally classified as Stokavian. In the remaining štokavian dialects, the contrast of neo-acute and circumflex has been lost, so that the intonations of standard SC produced by the leftward spread of tone are, so to speak, two degrees
7
The secondary nature of the SC intonations has been known since the 19th century and is beyond controversy, yet one not infrequently comes across statements in the scholarly literature on Slavic accentuation such as "the acute (rising) tone (")•..its reflex in Serbo-Croatian is short falling accent (kráva)" or "Falling tone on a long syllable—the Slavic circumflex (A)—is rematined as such in Serbo-Croatian" (IlliC-SvityC 1979:75-6). Cf. also Kortlandt 1988:304: "On the analogy of the endstressed forms, the laryngeals were eliminated from the barytone forms in paradigms with mobile stress, e.g. SCr. ace. sg. gldvu 'head', sin 'son', where the circumflex points to the absence of a laryngeal, as opposed to Lith. gálvq, su'/iy, where the acute tone reflects its original presence (Meillet's law. cf. Stang 1957: 10)", where it is surely the vowel length in SC. not the circumflex, that reveals the neutralization of tonal oppositions in AP c in pre-Proto-Slavic. As one might expect, the "Stokavski shift" has occurred to varying degrees throughout the former Yugoslavia, some dialects having escaped it altogether (e.g. the Zeta-Lovcen dialects of Old Montenegro), some having carried it through in full (e.g. eastern Hercegovinian, the native dialect of Vuk Karadzic and the basis for all varieties of modern standard Serbo-Croatian). Elsewhere it is conditioned by the syllable structure and vowel length: retraction is more likely from a vowel in absolute word-final position, i.e. from an open final syllable, and is favored if the preceding vowel is long; for details cf. Alexander 1993. Cf. Slovene, where retraction from word-final syllables to long *e, *o precedes retraction to short *e, *o (Garde 1976a:255-60, 448).
108
removed from those of Proto-Slavic.8 As we shall see — and as has been long been known — acute, circumflex, and "neo-acute" intonations on PS1. long vowels are reflected in vowel quantity in Serbo-Croatian, i.e. as length vs. shortness; for details see §4.2.2. The underlying position of stress in standard SC, Stokavian, as the underlying (and surface) position of stress in Cakavian and those stokavian dialects which have not undergone leftward tone spreading or exhibit it only partially, continues the placement of stress in Proto-Slavic, with of course certain innovations and alterations. Note finally that the East Slavic languages are generally considered to have preserved a reflex of PS1. intonation in the accentuation of the so-called "ToRT" or polnoglasie sequences. In Russian, Ukrainian, and Belarusian, the PS1. diphthongs *er, *or, *el, *ol develop into disyllabic ere, oro, olo, olo, respectively. If the diphthong was originally acute, this is reflected as accent on the second of the two identical vowels, e.g. Russ. moróz 'frost' (cf. morótenoe 'ice cream') < PS1. *mórzú (SC mráz), Russ. vorótit' < PS1. *vrátiti (SC vrātiti). Circumflex *er, etc., on the other hand, develop into *ére, etc., with accent on the first vowel, e.g. Russ. górod, Ukr. hórod < PS1. (unacc.) *gordu, (cf. SC grād, ú grād, Cz. hrad). This would then be the only case in which modern Slavic languages continue PS1. intonations as a contrast of position of surface stress, rather than of vowel length. More recently, however, it has been argued that the divergent stresses of moróz and górod do not require recourse to PS1. intonations (cf. Kurytowicz 1962:35ff., Jakobson 1963, Garde 1976a:264-7, 428, 449): if one assumes that acute accented PS1. diphthongs developed into East Slavic polnoglasie sequences with stress on the second vowel, e.g. PS1. *(mórz-í > Russ. moróz, the stress on the first vowel in górod need not directly reflect a PS1. circumflex, but is rather a default initial stress in unaccented PS1. *gðrd-u (cf. ná gorod). To summarize, the various modern Slavic languages provide the following kinds of evidence for the reconstruction of PS1. accent: • •
Slovene (conservative Carniolan dialects): distinctive length on final vowels only; intonations non-distinctive (circumflex on PS1. unaccented forms, otherwise acute) Serbo-Croatian (Cakavian, kajkavian, Posavian): distinctive length, surface intonations are contours across moras (retraction from a weak jer or non-final circumflex long vowel produces acute; otherwise circumflex);
^The conservative Posavian dialects of Slavonia differ from the rest of Stokavian in that they preserve the post-PSl. rising and falling intonations, like the Cakavian of Nov! or Cres. As a result of the Stokavian leftward spreading of high tone, these dialects now possess three contrastive intonations on long vowels: circumflex or falling (A), "old" rising from (post-)PSl. neoacute (~), and "new" rising from Stokavian retraction ('); cf. Garde 1976a:248.
109
• • • • • •
Serbo-Croatian (other non-Torlak štokavian dialects): distinctive length, intonations result from stress retraction; East Slavic (Russian, Ukrainian, Belarusian): lexical accent, no length, intonations reflected only in polnoglasie sequences Eastern South Slavic (Bulgarian, most Torlak and some Macedonian dialects): lexical accent, no length or intonations Transitional South Slavic dialects (mostly in Macedonia); position of stress automatic (antepenultimate in literary Slavic Macedonian), no distinctive length; Slovincian, Polabian: distinctive stress (still poorly investigated in the case of Polabian); Rest of West Slavic: initial stress, but contrastive length (Czech, Slovak, Old Polish; penultimate stress in modern Polish, with relics of former vowel length in ó, q < *ō, *?)•
Of the modern Slavic languages, the leading role in accentological studies has traditionally been assigned to Russian and Serbo-Croatian (Sto- and ðakavian), with supporting or supplementary data from Ukrainian and Slovene, respectively. Although Bulgarian preserves lexical stress, intraparadigmatic alternations have been sharply reduced in comparison with Russian or Serbo-Croat; furthermore, various stress shifts, whether leftward from final syllables or rightward towards a postposed definite article (< PS1. enclitic demonstrative pronoun *ji, *ja, *je), have led to extensive remodeling and partial merger of the inherited accentual types, e.g. the three classes of nominal stems (cf. §4.2.3). Thus Bulgarian data have typically been adduced, and will be cited below, only where it clearly continues a PS1. pattern, e.g. in the stress alternation between sg. mjásto 'place' and pi. mestá, or unstressed /-stem noC 'night', rádost 'happiness, joy', def. noð-tá, radost-tá. Other evidence, e.g. from Slovincian or from vowel length alternations in West Slavic, will be adduced where appropriate, particularly in the discussion of the accentuation of the PS1. verb (chapter 5). 4.2.2. The prosodic inventories of Lithuanian and Proto-Slavic As noted above, the phonemic inventory of vowels and diphthongs in Lithuanian, and their possible intonations, are especially clear. Standard Lithuanian possesses four short vowels, /', e, u, and a, the latter of which continues PBS1. *a < PIE *a, *o. These may carry the stress but, being short, do not distinguish intonations; in most cases, especially in open syllables, a stressed Id or/a/ is lengthened (Senn 1966a:70-l).9 On the
^Exceptions: the posssessive adjectives máno, tāvo, sávo 'my, your, his/her/its own', tie 'not', comparatives of adjectives, e.g. gerésnis 'better' to gēras 'good', and infinitives to roots of the
110
other hand, the long vowels y /i:/, é /&:/, o /o:/, and tī/u:/, as well as the various diphthongs, ai, ei, au, ie, uo, and combinations of a, e, i, u with r, /, m, n, may be underlyingly marked for acute or circumflex intonation. Thus the vowel system of Lithuanian, including intonational contrasts, is as below: 1:
*
e:
é:
u: ó:
fr
ír er íl el
úr ár úl ál ún án úm ám
1:
ér fl él in én im ém
in
én im
ēm
u: ð:
úo íe
úo íe
úr ár úl
éi ái áu
éi ai
au
āl ún
ān um
ām
Note that the historically nasalized long vowels q, f, 1, y, which have been denasalized in most dialects and the standard language, may continue to be treated as underlying sequences of vowel + nasal, at least in word-internal position. This is especially clear in cases such as pres. lsg. trlsii < *tri-n-s- to inf. tristi 'become restless', or pa-Zl-stu < *fin-st- to inf. pa-Hn-ti 'know (a person)'; in the latter, the pres. and inf. reflect *VN before a consonant and before a vowel, respectively (i.e. tautosyllabic *VN vs. *V.N). Like other vowel + nasal diphthongs, they contrast acute vs. circumflex intonation, e.g. ¿qsis 'goose' < *ghans- (Gr. khé:n, gen. khε:n-ós, Lat. āns-er, PGmc. *gans- > OHG gans, OE gas) vs. pa-iļ-sta 's/he knows' < *Zin- < *gjjh3- (cf. Bammesberger 1993). Word-finally, nasal vowels occur only in the ace. sg. of nouns and adjectives (of all stem classes) and in the nom. pi. masc. of the present and perfect active participles, e.g. dirbq '(who are) working', dirbq 'having worked, who worked' (sg. dirbqs,dirbqs,
stems dtrbant-,
dirbus-).
Considering that there is no contrast between e.g. q and an or am in final position,10 and that unstressed long [a:] and [ε:] do not otherwise occur in Lithuanian, I treat these endings as underlying /aN/, /eN/, /iN/, and /uN/ as well. Although it is not always acknowledged, the Proto-Slavic vowel inventory appears to have been extremely similar to that of Lithuanian. Like Lithuanian, PS1. had four short shape CaC-, CeC-, e.g. ras-ti 'to find', deg-ti 'to burn'. Final short -a(s) and -e(s) in the nom. sg. of āstems and the instr. sg. and ace. pi. of ā- and ē-stems is regular by Leskien's Law (§4.2.3). 10 The only exception is the dat. du. in -ám, -im, -urn and instr. du. in -am, -im, -um, on which see §5.2.
HI
vowels, *í, *e, *ú, and *o, with the latter continuing PBS1. *a < PIE *a, *o. n The long vowels and diphthongs, namely *i (< *i), *€ (< *ē), *y (< *u), *a (< *ā < PBS1. *ā, *ð), *u (< PBS1. *au, *eu < PIE *au, *ou, *eu), *e. (< *in, *en, *im, *em), * (< *un, *on, *um, *om), *ír, *er, *úr, *or, *I1, *el, *úl, *ol, could be underlying either acute or circumflex. The PS1. vowel system, with all intonational possibilities, was therefore *i
*é
*á
*ir
*ir
*ér
*ér
*íl
*51
*él
*él
*ur *ór *ul *ól
*ur *ðr *ul *ðl
This system has been somewhat disturbed by two probably late innovations:
the
shortening of historically long vowels in word-final position, and the introduction of "neoacute" intonation on long vowels and diphthongs as well as on short vowels. Although some scholars have posited a distinction between acute and circumflex intonation on long word-final vowels on the basis of language-specific phenomena, in particular the "neocircumflex" of Slovene and neighboring Serbo-Croatian dialects (cf. fn. 12), the evidence for such a distinction is less than clear. I therefore tentatively follow Garde (1976a:207-8) in assuming that contrasts of length, and therefore of intonation, were neutralized in wordfinal position. In cases where the long vowel has been preserved in a non-final syllable, the inherited intonation remains underlying: in d-stem nouns, for instance, nom. sg. *-a (short, and thus marked with grave accent below where relevant) alternates with acute *-á-, preserved in pi. dat. *-á-mú, instr. *-á-mi, *-á-sú (see §5.1.4). Otherwise, word-final *-a, *-y, *-§, etc. must be regarded as allophones of underlyingly long *a, *y, *^, indeterminate as to intonation. As noted in §4.2.1, Stang's Law retracts stress from a word-internal circumflex long vowel; if the preceding vowel is long, it receives "neo-acute" intonation, which is realized as a long rising vowel in Cakavian and yields a long vowel in all contemporary "Evidence that [a] was the original pronounciation and that this survived quite late into the prehistory of Slavic comes from the early medieval Greek spelling Sklabznoi 'Slavs', and from the OCS name for Salonica: Greek (T''es)salon(íkε:) must have been borrowed into pre-PSl. as *Salunu. whence PSI. and OCS Solunii (still called Solun by the South Slavs today). Cf. Goth. Rumoneis 'Romans', which demonstrates that the merger product of post-PIE *a and *o, traditionally reconstructed as *[5], must have been *[ā]: Lat. Rōmān-ī—> PGmc. *Rumān-lz > Goth. Rumoneis.
112
Slavic languages which preserve length distinctions. Whether preceding short vowels likewise had neo-acute intonation has been a topic of controversy among Slavic accentologists: Stang (1957:22-3) believes in a contrast of neo-acute intonation on short vowels in PS1., whereas Vaillant (1950:265-8) and Garde (1976a:221, 232-3) argue that "néo-aigu sur breve" has arisen by analogy to long neo-acute vowels in the individual (mostly West Slavic) languages. To be sure, short vowels which have acquired stress as a result of Stang's Law do not become long in all Slavic languages with distinctive length: in eakavian, the root vowels of pres. lsg. morēn, 2 mores are short, just as in inf. moc (cf. standard SC mógu, moleš, mole). This, however, is no argument against neo-acute short vowels in PS1.: it is entirely possible that the reflexes of short vowels with neo-acute intonation — however that intonation was realized phonetically — became phonemically identified with length in West Slavic (Czech, Slovak, Old Polish, pre-Sorbian), but not in western South Slavic, where they remained short. Thus the long vowels in Czech pres. 2sg. múleš, 3 mule (lsg. mohu, colloquial mú¿u) need not be somehow "analogical" to e.g. vále 'ties' (inf. vázati for *vazati with long vowel from pres., Vaillant 1950:260-1; otherwise Garde 1976a:232), but may directly reflect post-PSl. *mōZeš\ *mōzet < pre-PSl. *moZéši, *mo2éti (lsg. *mogq, cf. Russ. mogú, móleš, mólet); in any case, such an analogy, as Stang (1957:23) correctly observes, will not account for forms such as Cz. vule 'will' < PS1. *v51'a < *vo!ija (Russ. vólja, dial, vuólja; SC volja). This hypothesis is supported by the evolution of the simple thematic presents in eakavian and Slovak, both of which require a neo-acute on the short thematic vowel *-e- that cannot have arisen by any reasonable analogy (cf. §6.3.1).12 Aside from these relatively late developments, and the typically Slavic monophthongization of PBS1. diphthongs with glide or nasal as second element (*ai > *e\ *ei > *i, *au, *eu > *u, *i/*e + *n/*m > *^, *u/*o + *n/*m > *q), one sees that PS1. has retained the fundamental characteristics of the PBS1. system which are so faithfully reflected in Lithuanian: a contrast between acute and circumflex in long vocalics, and a lack of such contrast in short nuclei. To be sure, none of the modern Slavic languages preserves the PBS1. intonational contrasts as such: the rising and falling intonations attested in contemporary Slovene and eakavian, kajkavian, and Posavian Serbo-Croatian 12
By contrast, the "neo-circumflex" attested in Slovene and adjoining dakavian and kajkavian dialects of Serbo-Croatian (i.e. the unexpected circumflex in certain historically accented forms, in contrast to the Slavic circumflex on PS1. enclinomena) is probably not to be projected back to Proto-Slavic: cf. Stang 1957:23-35, Garde 1976a:240-l; differently Dybo 1981:31-2, Dybo et al. 1990:16-34.
113
dialects do reflect important prosodic contrasts of PS1. or early post-PSl. date — namely vowels with vs. without "neo-acute" intonation — but do not continue PS1. acute and circumflex long vowels and diphthongs (§4.2.1). Instead, the contrast between acute and circumflex, in underlyingly accented (or postaccenting) stems only, is reflected in the presence or absence of vowel length in those languages which preserve length distinctions, e.g. Serbo-Croatian or Czecho-Slovak (see below; on the East Slavic polnoglasie sequences cf. §4.2.1). The intonations themselves are an innovation of Balto-Slavic, but find their origin in the syllable structure of PIE. Specifically, it appears that PIE vowels or diphthongs followed by a tautosyllabic laryngeal, i.e. in syllables of the form *CVH, *CVRH, yield acute long vowels or diphthongs in BS1. This historical connection, first noted by de Saussure (1894) and Bezzenberger (1891), has been demonstrated by Dybo (1981:40-4) on the basis of several Lithuanian forms with good stem-cognates (although, unfortunately, these rarely form exact word equations; cf. §4.3.2 below), including the following:13 Lith. gyvas (AP 3), Latv. dzivs, PS1. *2ívú (AP c, Russ. Ziv, -á, -o) 'living, alive' < PIE *gwih3-wó-: cf. Skt. jīvá-, denom. thematic present *gwih3W-e/0- > Skt. jívati, Av.juuainti, Gr. zo:o:, Lat. vivo, TB šaim, TA šos (§6.1, fn. 2; on the failure of Hirt's Law see §4.3.2); Lith. jáunas (AP 3), Latv. jauns, PS1. *junú (AP c, Russ. jun, -á, -o) 'young' < PIE *yewH-n-o-: cf. Skt. yúvā, ace. yúvānam, gen. yénas, Av. yuua, ace. yuuānam, gen. yunō < *yuH-ōn, *-on-iņ, *yuH-n-és <— *yéwH-on- ~ *yuHn-' (amphikinetic paradigm, cf. §2.2.1.3); Lith. tgvas (AP 3), Latv. tiévs 'thin, slender' < PIE *tenh2-wo-: cf. Gr. tanaós (<— *tenawós). In contrast, PIE syllables of the form *CVR, without laryngeal, develop circumflex intonation in BS1., as the cognates below illustrate: Lith. diēvas 'god' (AP 4), Latv. dievs 'god, sky', PS1. *divu (AP b, cf. §4.3.1) < PIE *deywós: cf. Skt. devás 'god', Av. daēuuō 'demon'; Lith. miēgas (AP 4), Latv. miegs 'sleep' < PIE *h3moyghós: cf. Skt. meghás, Av. maēγa- 'cloud'; Lith. kainá (dial. AP 4), PS1. *céná (AP c) 'price' < PIE *k w oyneh 2 : cf. Gr. poinε: 'penalty, punishment, ransom' (Skt. cáyate 'punishes, exacts (debt, repentance)' < *k w éy- e / 0 -, Gr. Horn. ti:no:, Att. tino: 'pay (a debt), make amends' < *ti-nw-e/0- <— *kwi-néw- ~ *kwi-nu-', tísis 'payment, penalty' < *kwi-ti- <— *kwéy-ti-); 13
The quality of IE cognates for acute intonation is significantly worse than for circumflex intonation, until one adduces examples of Hirt's Law (see below, §§4.3.1-2).
114
Lith. šimtas (AP 4) 'hundred, Latv. simts 'sotnja' < PIE *lúņtóm: cf. Skt. šatám, Gr. hc-katón, PGmc. *hunda. > Goth. hund\ Lith. dantls 'tooth' (AP 4, gen. pi. dantif) < hjdónt-: cf. Skt. dán, acc. dántam, gen. datás, Gr. odd.-s, gen. odóntos, Lat. dēns, gen. dentis, all eventually < PIE *hid-ént- <— *hiéd-ņt- 'eating', pres. act. ptcp. of *hiēd- ~ *hied- 'eat' (§2.3.1.2); Lith. iiema (AP 4), Latv. ziema, PS1. *zima (AP c) 'winter' < *gneymeh2 <— PIE *ghey-ōm, gen. *g"hi-m-és: cf. Av. acc. ziiqm < *gny-ōm with zero-grade, whence nom. ziia; Gr. kheimð:n 'winter', khio:n 'snow' by paradigm split; Lat. hiems 'winter'; Skt. héman- 'wintry', himá- 'snow'); Lith. minus 'mind, opinion' (AP 4) < *mņ-ti- <— PIE *mén-ti- ~ *mņ-téy-: cf. Lat. mens < *m(e)n-t-, Skt. man's, mátis 'mind, thought' (§2.2.1.4); Lith. savsas, Latv. sáuss, PS1. *suxú 'dry' < PIE *saus-o-: cf. Skt. šúsyati 'dries, sticks', PS1. *súše/0-14 (—> *sux-ne/0-, Russ. sóxnet) < *sus-ye/0-, Lith. susti, Latv. sust 'dry, wither' with zero-grade *sus-. It also appears that original PIE lengthened grades, of non-laryngeal origin, yielded circumflex intonation in BS1., although the evidence is necessarily somewhat sparse. This is suggested by the intonation of word-final -é and -uo in Lith. duktē 'daughter', sesuō 'sister', akmuo 'stone', with PBS1. nom. sg. *-ē, *-ō indirectly continuing PIE r-stem *ēr, *-ōr, /j-stem *-ō (< pre-PIE *-er-s, *-or-s, *-on-s by Szemerenyi's Law). On the Slavic side, Jasanoff (1983:139-42, 146-8) explains the PS1. reflexes of these endings, namely *mat-i 'mater', *dut'-i 'daughter' (OCS dúšti) and *kam-y 'stone', *kor-e, 'root', as the result of a regular sound change raising circumflex vowels in word-final syllables; this Auslautgesetz neatly accounts for at least two other nominal endings, cf. §5.1.4.15 Further support for a circumflex treatment of PIE Dehnstufe comes from the Slavic sigmatic aorist, which continues the (post-)PIE sigmatic aorist with long-grade of the root 14 Or rather PS1. *-c- - *-q-, as it seems clear that speakers of PS1. and OCS had reinterpreted the thematic vowel as *q in lsg. and 3pl. vs. *e elsewhere. In my view, such a reanalysis could have been in some way responsible for the replacement of *-o- by *-e- in lpl. *-e-rmī (and ldu.*-e-vé, with ending remodeled after ldu. pers. pron.), usually ascribed to generalization from * - j e / 0 - presents. The elimination of inherited *-o-mu (and *-o-vV) would thus have been a late pre-PSl. innovation, subsequent to the rise of nasal vowels. 15 To be sure, the circumflex intonation of «-stem nom. sg. *-6 could instead continue a generalized *-o-h!0, with Hoffmann suffix *-h,(o)n-. The intonation of r-stem *-ē, *-ō would then be analogical to the «-stem ending. Cf. Jasanoff 1983:141, 147. — Jasanoff (pp. 147-8) cites the equation of OCS (otú)-tgdē 'thence' with Goth, þandē 'since, because' as evidence that *-l > PS1. *-é (which would require the *-i of *mat-i to be from *dut'-i, with *-i < *-€ after palatalized *t'), but it is difficult to separate this form from PS1. *kú-de 'where' (> OCS ku-de, Bulg. kadé, Cz. kde, Pol. gdzie). *sl-de 'here' (> OCS sī-dē, Cz. zde). Since *-e has been replaced by o-stem loc. sg. *-eA in more than one instance (cf. Russ. gde, zde-s' with emphatic particle *-sī <— *-se [Vasmer 1986a:400. 1986b:89-90], SC gd(j)e, ovd(j)e <— *ovi-de 'here'), it is not impossible that the final vowel of (otu)tņdé is to be explained in a similar manner.
115
(outside the 2 and 3sg., where OCS attests only a suppletive thematic aorist with ordinary full-grade). Unfortunately, this formation survives only in the oldest OCS manuscripts, e.g. the Codex Marianus; already in the late OCS Codex Suprasliensis, it begins to be replaced with the so-called "extended" or "second sigmatic aorist" {erweiterter Aorist), a specifically Slavic innovation (van Wijk 1931:220-2, Leskien 1962:131-3). Relic forms survive into Middle Bulgarian, Old Czech, and Serbo-Croatian, and are attested in various Church Slavonic texts, but as accented medieval manuscripts are almost entirely of Russian and Bulgarian origin (whether Russian and Bulgarian CS, or Old Russian and Middle Bulgarian), and those languages preserve no relevant intonational reflexes,16 they provide no information on the intonation of the stem vowel of the PS1. and OCS sigmatic aorist. Fortunately, Serbo-Croatian preserves an indirect reflex of the vocalism of one such aorist, in the inf. dó-n(ij)ē-ti 'bring', pres. lsg. do-nés-ēm. This infinitival stem, for expected "do-nes-ti", has no other conceivable source than as a backformation from lsg. *né-xíí, lpl. *né?-xomíí, 2 *né-ste, 3 *né-5e. <— PS1. sg. 1 nēs-u, 2, 3 *né (attested in medieval SC; replaced by thematic nes-e, attested in OCS), lpl. nes-omtf, 2 *ngs-te, 3 nēs-q (3du. néís-ta), with substitution of the alternating suffix *-x- ~ *-s- ~ *-§- of the productive erweiterter Aorist and reinterpretation of the stem-morpheme boundary (cf. Leskien 1914:547-8, Vaillant 1966:60).17 The long -ē-1-ije- < *-é- in dó-n(ij)ē-ti indicates that this vowel had circumflex intonation in PS1. Assuming that the sigmatic aorist of PS1. *nes-ti behaved no differently with regard to intonation from other inherited sigmatic aorists, we may therefore deduce that the long *i of the PIE sigmatic aorist receives circumflex intonation in PS1., just as in word-final position in the nom. sg. of n- and r-stems. How the acute and circumflex intonations are to be interpreted synchronically is not immediately obvious, and more than one analysis is possible. Halle and Vergnaud (1987:191-2) and Halle (1997:293) treat circumflex intonation as the marked member of the pair, and assume that the final nucleus slot of circumflex long vowels and diphthongs, 16
The polnoglasie sequences do not come into consideration here: of the verb stems with a sigmatic aorist in the oldest OCS texts, only *velk- and *vlrz- are of the shape CVRC, and their aorists (OCS vléxu, vlé'xomú, vléste, vléšq; vrés-ú, -ta, -q) have been eliminated in favor of the productive "extended" -oxv aorist in SC. 17 Note that among the verbs for which a sigmatic aorist is attested in OCS or Church Slavonic, nes-ti, trqs-ti 'shake', and ves-ti 'drive, transport' (to /vez-/) are the only ones with a root ending in -s- or -z-. Similarly, the PS1. sigmatic aorist lsg. réxā, 3pl. rēšq, medieval SC rēh, réše, has led to the creation of an inf. rijéi in the Dalmatian dialects (e.g. Dubrovnik, cf. Vaillant 1966:60), with long -ije- from circumflex *-é-. — I believe that the strange alternation of pres. *jéd-e/0- - inf. *jéx-ati 'ride' also owes its origin to a reanalysis of the suffix of the sigmatic aorist, and hope to take up this topic in the near future (pres. *-d- is from the iptv., cf. LIV:275).
116
i.e. the second mora of a long vowel or second component of a diphthong, must be specified as potentially stress-bearing (i.e. projected onto Line 0 of the metrical grid; see §4.2.3). Since acute intonation has developed from a former syllable-final laryngeal in PIE (see above), and circumflex long vocalics pattern with short vowels in many respects (see above on the lengthening of stressed a and e in Lithuanian, and §4.2.3 on de Saussure's Law), I propose that it is instead acute long vowels and diphthongs which were specified as [acute], with circumflex long vowels and diphthongs being unmarked for the feature.18 This feature in turn is likely to have corresponded to some phonetic property; given that acute vowels and diphthongs in originally pretonic syllables are continued in Latvian with the so-called "broken" tone, i.e. are pronounced with a glottal stop (cf. the Danish st0d), a number of scholars have tentatively proposed that acute long vocalics were phonetically accompanied by glottalization. It must be remembered that the contrast of acute and circumflex intonation on a long vocalic is — at least in principle — entirely independent of the underlying accentuation of the morpheme containing that long vocalic. This in itself is hardly surprising, yet it deserves explicit repetition because of the widespread and occasionally explicit assumption that Proto-Slavic acute and circumflex long vowels or diphthongs were respectively accented (with columnar/immobile stress) and unaccented (with mobile stress), e.g. in the philologically impressive articles of Koch (1984, 1985, 1986).19 In Lithuanian, of course, just as nouns with accented stems (and columnar stem stress in pre-Lithuanian, before the operation of de Saussure's and Leskien's Laws; cf. 4.2.3) may contain acute long vocalics (AP 1) or circumflex long vocalics or short vowels (AP 2, with the effects of Saussure's Law), unaccented stems may be either acute (AP 3) or circumflex/short (AP 4). Likewise, pre-PSl. had four possible combinations of underlying accentuation and stem-vowel intonation, on which see §§4.3-4: 1. accented acute 2. postaccenting circumflex
(V V(
' 8 AIso possible is that [acute] is a binary feature, with short vowels and circumflex long vocalics being marked [-acute]. As far as I am aware, however, there is no evidence which would require this assumption. 19 Thus Koch (1984:142ff., 1985:129-31) claims that P(B)S1. acute roots, which continue PIE laryngeal-final or "set" roots, follow columnar root-stressed paradigms, and correspondingly that circumflex roots, which continue PIE roots not ending in a laryngeal ("anit" roots), belong to the mobile accentual paradigm, AP c, into which some acute roots have fallen by '"analogy". As Reinhart (1992:371) correctly emphasizes, verbal roots of AP c may continue either anit or set roots, e.g. PS1. *mer- 'die' (inf. *mer-ti) < PIE *mer-, PS1. *pén- 'stretch' (inf. *p<-ti) < PIE *penH-. See further §6.6 on Koch's evaluation of unaccented acute verbal roots in Slavic.
117
3. unaccented acute 4. unaccented circumflex
V V
To be sure, the contrast between acute and circumflex intonation in unaccented stems survives only in certain cases, e.g. infinitive SC grhti 'be biting, chewing', Russ. gryzt' 'be grawing, nibbling', SC kldsti, Russ. klast' 'be laying' vs. SC trésti, Russ. trjasti 'shake' (pres. 3sg. SC grízē, kládē, írésē, Russ. gryzet, kladet, trjasēt, all AP c), where the former, containing an originally acute vowel, have retracted the suffixal accent by Hirt's Law (cf. §4.3.2). Elsewhere, the distinction has been neutralized, as first observed by Meillet in 1902 ("Meillet's Law"): it is commonly stated that circumflex intonation is generalized, but the correspondences in vowel length among the contemporary Slavic languages, in particular Serbo-Croatian and West Slavic (e.g. Slovak) reveal a different pattern from the reflexes of acute intonation in AP a and circumflex intonation in AP b: PS1. acute circumflex neutralized (AP c)
SC short long long
Slovak short long short
Building upon Garde (1976a:213ff., 1976b), who considers the correspondence of length in SC and shortness in Slovak to reflect a vowel in the "unaccentable portion" of the phonological word, I assume that vowels were neutralized with respect to intonation in unaccented forms (enclinomena, in Dybo's terminology) and when occurring before the pretonic syllable.20 The phonetic product of this neutralization was generalized to other accent forms, e.g. nom. sg. *glā:v-a, *zl:m-a, gen. sg. *glā:v-y, *zī:m-y like ace. *glā:vQ, *zI:m-Q, instr. pi. *glā:v-á:mi, *zl:m-á:mi < PBS1. *gálv-'á:, *féim-'á: (Lith. galvá, gálvq vs. ¿iemá, ¿iēmq). After the breakup of PS1., this neutralized intonation — whatever it was phonetically — merged with the circumflex in SC and with the acute in West Slavic with respect to vowel length, yielding a long vowel in SC but a short vowel in Slovak.21 20
Because he assumes Dybo's Law, Garde (1976a:213-7) considers the "zone inaccentuable" to include all syllables up to and including the pretonic syllable. The analysis proposed here is supported by the evidence of Serbo-Croatian, where long vowels, and therefore length contrasts, may occur only in the stressed syllable (i.e. the one marked for stress in standard grammars) or following syllables: because stress in standard Stokavian SC has been shifted leftwards (§4.2.1, cf. fn. 7). the stressed syllable corresponds to the pretonic syllable in PS1. 2 'Except in "pre-pretonic" syllables, where vowels with neutralized intonation — i.e. all vowels — are short in SC. In my view, this is the only reasonable explanation for the shortening of the long root vowel in the dat./instr./loc. pi. of a-stems of AP c, e.g. glāv-ama to gláv-a 'head', -é. glávi, -M, gláv-ðm, pl. gláv-e, gláv-ā, rúk-ama to rúk-a 'hand', -é, rúc-i, ríik-u, rúk-ðm, pl. rúk-e, rúk-ā
118
Note that if acute and circumflex vocalics were respectively accented or unaccented — in other words, if acuteness of a long vocalic were correlated with underlying accentuation — then the PBS1. accentual system would not have been lexically contrastive. Just as many languages assign surface stress to the leftmost or rightmost syllabic head marked for a certain property (e.g. examples), stress in PBS1. would be assigned to the first acute syllabic head, i.e. the leftmost one marked [+acute]. Such a view is inconsistent with the simple fact that in Lithuanian, and in pre-PSl. prior to the neutralization of intonations just described, monosyllabic nominal or verbal roots with acute-intoned long vocalic could be either accented or unaccented. The membership of a given root in one or the other accentual paradigm is entirely unpredictable, at least for some morphological classes, and must be represented in the lexicon: the learner of Lithuanian, for instance, has to learn that the root of 'crow' is underlyingly accented (várn- and that the root of 'head' is unaccented gálv-, just as the learner of Russian must memorize to which of the three major APs (and their various subtypes) a given word belongs. Both intonation and underlying accentuation were lexically specified properties of morphemes in pre-PBSl., and both must be taken into account in investigating the accentual history of the Baltic and Slavic languages. Finally, the combined evidence of ancient Greek, East Slavic polnoglasie sequences (at least under the traditional analysis; cf. above), and Old Prussian indicate that acute intonation corresponds to greater prominence on the second mora of a long vowel or second element of a diphthong, and correspondingly, that circumflex intonation corresponds to greater prominent on the first mora or first element. Lithuanian has famously "reversed" this distribution: acute long vowels and diphthongs are pronounced with greater intensity on their first mora, whereas circumflex long vowels and diphthongs stress the second: hence ó, 5 stand for phonetic do, ob, respectively.22 This reversal is (Leskien 1914:352). In PS1. and early SC, these forms were stressed on the ending, so that the intonation of the root vowel, which fell in the "zone inaccentuable", was neutralized and developed to a short vowel; the resulting alternation between glavama (and old instr. pi. glavami) and glāva, - ē , giāvu (na glāvu), pi. giāve {n'aglāve), etc. remained after the datVinstr. du. —> dat./instr.(/loc.) pi. shifted its accentuation to the stem vowel -a-. (The same applies to gen. pi. forms in -il, which with its long vowel must continue contracted *-aju or sim. for PS1. genVloc. du. *-u: ruk-u < *ruk-aju <— PS1. ♦rgk-u.) If the original accentuation of the du. and pi. oblique case forms had been on the stem vowel, PS1. *gólv-á-ma should have developed to SC "glávama" /gletv-ama/, exactly like nom. sg. *gólv-a > SC gláva /glðv-a/. —That this "reversal" need not reflect the original phonetic values of the acute and circumflex becomes clear from a comparison with other Lithuanian dialects and with Latvian. In the Zemaitian dialects of northwest Lithuania, the PBS1. circumflex remains falling, whereas the acute is phonetically glottalized, i.e. "broken". The central dialects of Latvian famously continue PBS1. acute intonation as sustained or rising in forms corresponding to Lith. AP 1 (i.e. initial-stressed) and "broken" in forms corresponding to Lith. AP 3 (i.e. with non-initial stress, retracted to the first syllable
119
represented in the orthographic notation of stressed diphthongs: acute ár, ll, úo vs. circumflex af, il, uo. Although the agreement of the other languages makes it almost certain that it is Lithuanian which has innovated in the realization of acute and circumflex, the precise phonetic nature of these intonations will not concern us here; what is relevant, and all that may be reconstructed by the comparative method, is the phonemic distinction between two contrasting "types" of long vocalics, embodied in the feature [acute] (cf. above). 4.2.3. Accentual classes and accentual paradigms As noted in chapter 1, Halle (1997) has proposed that BS1. languages such as Russian, Serbo-Croatian, and Lithuanian are characterized by a system of underlying accentual specifications:
every morpheme, including nominal and verbal stems,
derivational suffixes, and inflectional endings, is either accented or unaccented. In his terminology (p. 280ff.), if one of the syllabic heads of a morpheme is marked with a preceding left bracket, that morpheme is referred to as accented; if the left bracket is to the right of the last syllabic head, the morpheme is postaccenting. One of the most significant discrepancies between the Baltic and Slavic accentual systems lies in the number of contrasting accentual classes. Since Stang (1957), all scholars have agreed that Proto-Slavic distinguished three classes of nominal and verbal stems, namely accented, unaccented, and postaccenting; this three-way distinction has been preserved in most contemporary Slavic languages which have retained lexical accent, including Russian and Serbo-Croatian. We have already encountered examples of these classes in §1.2.4, in the paradigms of the Russian a-stems dúm-a 'thought', gospo¿-á 'lady', and borod-á 'beard': nom. sg. gen. dat. ace. instr. loc.
dúm-a dúm-y dúm-e dúm-u dúm-oj(u) dúm-e
gospoi-á gospoi-í gospot-é gospoi-ú gospoi-éj(u) gospoi-é
borod-á borod-y borod-é bórod-u borod-ój(u) borod-é
in Latvian). The other dialects suggests that the latter reflex is original: western Latvian merges sustained with falling, whereas eastern Latvian merges falling with broken, so that the three major dialects agree only in reflecting non-initial (i.e. retracted) acute as broken. It therefore appears that acute-intoned syllabics were originally glottalized, as suggested above, and that their subsequent phonetic development, as well as that of circumflex-intoned syllabics, has assumed various outcomes in the Lith. and Latv. dialects.
120
nom. pl. gen. dat. acc. instr. loc.
dúm-y dúm-0 dúm-am dúm-y dúm-ami dúm-ax
gospoi-í gospoí-éj gospot-ám gospot-í gospoi-ámi gospot-áx
bórod-y boród-0 borod-ám bórod-y borod-ámi borod-áx
In the treatment of Stang (1957:56ff.), these classes are denoted as a, b, and c, respectively; they will occasionally be referred to by these labels below for convenience. As one may observe, the accented stems have columnar surface stress on a stem syllable, whereas postaccenting stems, at least in pre-PSl., have columnar stress on the syllable after the stem, i.e. the first syllable of the ending. The latter pattern has been slightly disturbed on the surface by stress retraction from word-final PS1. jers, which were then lost, e.g. in PS1. *konic-i 'end', acc. *ljubuv-i 'love', etc.; this rule/constraint remains valid in the synchronic grammar of Russian (and Ukrainian, Serbo-Croatian, etc.), so that the paradigms of stol(-, ljubOv(-, etc. are underlyingly still columnar postaccenting. The same retraction has occurred in the gen. pl. of unaccented borod-á: the accented ending in underlying /borod-(0/ is realized as [0], and the stress surfaces on the preceding syllable, hence boród [barot]. Halle (1997:283-4) treats Accent Retraction and Jer Deletion within the context of metrical grid theory, deriving forms such as Russ. konéc, boród in the following manner:
*
ko nEc E
—>
* (* (*)
ko néc
cf.
ko nEc a
* *
*
bo rod
*
t*\
0
—>
—>
t*\
kon cá *
(* (*)
bo rod
*
*
* (*
cf.
(*
* r* * r*\ * * bo rod am 0 —-> bo rod am *
It is only the unaccented nominal stems which allow one — or, for that matter, the learner or speaker of Russian — to determine the underlying accentual properties of the various endings. Since a given ending should have the same underlying accentuation regardless of the accentual properties of the stem to which it is attached, we may analyze the paradigms of dúma, gospoiá, and borodá as follows:
121
nom. sg. gen. dat. acc. instr. loc.
(dúm-(a (dúm-(y (dúm-(e (dúm-u (dúm-(oj (dúm-(e
gospoZ(-(á gospo2(-(í gospo2(-(é gospoZ(-ú gospo2(-(éj gospoZ(-(é
borod-(á borod-(y borod-(é bórod-u borod-(ój borod-(é
nom. pl. gen. dat. acc. instr. ioc.
(dúm-y (dúm-(U (dúm-(am (dúm-y (dúm-(ami (dúm-(ax
gospoZ(-í gospoZ(-(éj gospoZ(-(ám gospoZ(-í gospoZ(-(ámi gospoZ(-(áx
bórod-y boród-(U borod-(ám bórod-y borod-(ámi borod-(áx
Thus in Russian, the endings of a-stem nouns are accented in the singular except for acc. sg. -u, and in the plural except for nom.(/acc.) pl. -y. To be sure, there are various exceptions to the three major accentual classes, mainly as a result of fluctuation in the underlying specification of these endings. Thus within the Russian a-stems, one finds nouns which alternate ending-stress with default initial stress in the nom./acc. pl., but not the acc. sg., e.g. sveð-á 'border', skovorod-á 'frying pan', acc. -w, but nom./acc. svéð-y (gen. sveð or sveð-éj, dat. sveð-ám, etc.), skóvorod-y (gen. skovoród, dat. skovorod-ám, etc.). In addition, the postaccenting type of gospoi-á is limited to suffixed nouns only: unsuffixed a-stems, as well as those formed with the suffixes -in-á, -ot-á, retract the stress to the final vowel of the stem in the plural.23 Some of these fluctuations reflect post-P(B)SL, language-specific innovations: thus elimination of the alternation in the sg., where only acc. /-u/ is unaccented, is an entirely natural case of faulty generalization based on imperfect language learning, and is paralleled e.g. in South Slavic (cf. §5.1.2), whereas retraction in the plurals of postaccenting a- and other stems is a relatively recent innovation of East Slavic, and likewise finds parallels in Serbo-Croatian (§5.1). In addition, individual lexical items have shifted from one accentual class to another, for reasons that may not be recoverable: this has occurred on a large scale in some languages, especially Bulgarian, but less extensively in East Slavic and Western South 23
A handful of nouns exhibit combinations of these variations: for instance, zimá 'winter' and zemljá 'earth, land', historically unaccented (acc. zímu, zémlju, cf. ná zimu, zá zimu, ná zemlju), have generalized retracted stem-final stress in the pi. from the postaccenting type: zimy, zim, -am, -ami, -ax, zémli, zemél', zémljam, -ami, -ax. (The stress of zeniél' indicates that the underlying accent is zem(El-, hence that these nouns have not generalized default initial stress from the nom./acc. pl.; cf. §5.1, fn. 2). On the other hand, dolja 'fate', with columnar stem stress in the sg., alternates in the pl. between nom./acc. dóli and gen. doléj, dat. doljám, etc.
122
Slavic, so that the number of items which correspond in e.g. Russian, Ukrainian, SerboCroat, and Slovene is fairly large.24 The computation of surface stress in Russian nouns proceeds straightforwardly from the application of the rules for prosodic computation given in § 1.2.4. I illustrate this below with metrical grids for the instr. pi. and ace. sg. of our three a-stems; as seen in the previous table, the endings are respectively accented -(ami and unaccented -u. * r* /*
* /■**■)
(durn (a mi *
(dúm u
*
*
*
*
(*
(*
/*
*\
bo rod (á mi *
*
*(
s*
*\
gos poZ( (á mi *
bó rod u
gos po2( ú
The same algorithm applies to the stress systems of other languages such as SerboCroatian, where leftward spread of high tone has resulted in rising and falling intonations on the pretonic syllable (cf. above, §4.2.1). As we have seen in chapter 2, Halle's extension of this analysis to other IE languages such as Lithuanian and Sanskrit is also fundamentally sound, permitting a number of insights into the workings of the PIE system of accent. Particularly significant for the study of Slavic accentology and its relation to PIE is the notion of underlyingly unaccented forms or "enclinomena." We have seen above that such forms have no accentual specification of their own, but instead surface with default initial stress in the phonological word as a result of the parameter settings described in §1.2.4. In Proto-Balto-Slavic, enclinomena were distinguished by their ability to "retract" stress to a preceding proclitic, or the first proclitic if more than one was present; underlyingly, of course, the entire prepositional phrase is treated as a single prosodic domain, which receives default initial stress. This property has been entirely lost in Lithuanian, where originally unaccented forms such as ā-stem ace. sg. gálvq, íiēmq have restricted default initial stress to the noun itself: hence } gálvq 'into the head', pefíiēmq 'during the winter', with no "retraction" to preposition (Garde 1976a: 189-91). In Old 24
Another innovation, which has aided the process of confusion and/or transfer of lexical items, has to do with the influence of the accentual pattern and alternations of one accentual class on those of another. One especially striking example of this is the retraction of stress in the pi. of originally oxytone a-steins: see §5.1. 123
Russian and Middle Bulgarian, however, and in Serbo-Croatian (at least the conservative norm), this system is fully productive (cf. ORuss. i volk 'and the wolf, OBulg. í more 'and the sea', SC i bag 'and God', with proclitic i; Dybo 1981:45-7); modern Russian has preserved some examples in prepositional phrases, e.g. ná goru 'to the mountain'. In addition, Dolobko in his classic article of 1927 demonstrated that enclitics following an enclinomenon regularly attracted the stress in PS1. This pattern, often referred to as the "Law of Vasil'ev-Dolobko", is amply attested in Old Russian and Middle Bulgarian (Dybo 1981:47-54) and survives today in Russ. preterite masc. sg. reflexive forms such as zanjal-sjá 'occupied oneself with, engaged in' < PS1. *za-ne,-lú se. (f. zanjalá-s'; non-reflexive zánjal, zanjal-á), in SC forms such as nocas 'tonight' < PS1. *not'ī si 'night-this' (cf. archaic Russ. notes'), and in the stressed Bulgarian postposed definite article with nouns of AP c, e.g. grad-it 'the city' < PS1. *gordii ru. The same phenomenon occurs in archaic Lithuanian illatives to unaccented nouns, e.g. galvon < *galvo-na 'into the head', pi. galvos-na (Stang 1957:65-7, Garde 1976a:9, 11), so one may assume that underlying accent was generalized to all enclitics at some point in pre-PBSl. Returning to the accentual classes, how does the three-way contrast reconstructed for Proto-Slavic compare with the evidence of Lithuanian? Descriptive grammars of standard literary Lithuanian agree in distinguishing four accentual paradigms (APs): paradigm 1 is columnar on the stem, and paradigms 2, 3, and 4 are "mobile", i.e. exhibit intraparadigmatic alternations. These are illustrated below with the classic examples liepa 'linden tree', ranka 'arm, hand', galvá 'head', and iiema 'winter', once again from the largest class of feminine nouns, the ō-stems: AP 1 liepa Uepos liepai liepq liepa liepoje liepa
AP2 ranka rahkos rahkai rankq ranka rankoje ranka
galvá galvðs gálvai gálvq gálva galvojé gálva
nom./acc. du. liepi liepom dat. liepom instr.
ranki rankom rankom
gálvi galvóm galvom
tiemi īiemóm íiemom
nom./voc. pi. Uepos liepy gen. liepoms dat.
rahkos rankya rahkoms
gálvos galviļ galvóms
íiēmos iiemíļ iiemóms
nom. sg. gen. dat. ace. instr. loc. voc.
124
AP3
AP4 iiema iiemōs
īiēmai íiēmq īiema íiemojé íiēma
acc. instr. loc.
líepas líepomis líepose
rankás rankomis rankose
gálvas galvomis galvosé
íiemās tiemomls íiemosé
In fact, however, both paradigms 1 and 2 are historically and synchronically columnar on the stem syllable, so that it is only paradigms 3 and 4 which are underlyingly alternating. The attested distribution of stress alternations is the result of de Saussure's Law, first formulated by the great Swiss linguist in 1894: stress, or rather high tone, is attracted from a non-acute syllable head (i.e. a circumflex long vocalic, or a short vowel) to a following acute long vocalic (cf. also de Saussure 1896). A subsequent sound change, Leskien's Law (cf. Leskien 1881), shortens all acute long vowels and diphthongs in word-final syllables. Thus underlying /-íe/, l-úol, l-ól, l-él surface as -i, -ú, -á, -, respectively; the underlying long vocalics are preserved when followed by an additional syllable, giving rise to alternations such as those observed between the endings of indefinite (short) and definite (long) forms of adjectives and between non-reflexive and reflexive present forms of verbs (Leskien 1881:189-90, cf. Dybo 1981:11 -12): instr. sg. masc. nom./acc. du. nom. pi. acc. pi.
indefinite ger-u 'good' ger-u ger-l ger-us
definite ger-úo-ju ger-úo-ju ger-íe-ji ger-úos-ius
nom. sg. fern. instr. nom./acc. du. acc. pi.
ger-a ger-a ger-i ger-as
ger-ó-ji ger-q-ja ger-íe-ji ger-qs-jas
lsg. pres. 2
non-reflexive suk-u 'turn (tr.)' suk-l
reflexive suk-úo-s(i) 'turn around, spin' suk-íe-s(i)
Both de Saussure's and Leskien's Laws remain operative, and may be considered as part of the synchronic grammar of contemporary Lithuanian. Thus the paradigms of the four ō-stem nouns cited above may be considered to belong to two underlying accentual classes, one columnar barytone, with accented stem syllable (> APs 1 and 2), the other mobile and unstressed (> APs 3 and 4). The underlying accentual properties and intonations of these stems and endings are listed below:
125
AP 1 (líp-(ó (líp-(ðs (líp-ai (líp-f (líp-ó (líp-o(jé (líp-a
AP2 (rank-(ó (rank-(ōs (rank-al (rank-% (rafik-ó (rank-o(jé (rank-a
AP3 gálv-(ó gálv-(ōs gálv-ai gálv-3 gálv-ó gálv-o(jé gálv-a
AP4 2iēm-(ó 2iēm-(5s fiēm-(aí Ziēm-4 ziēm-ó 2iēm-o(jé 2iēm-a
nomVacc. du. (líp-íe (líp-(óm dat. (líp-(om instr.
(rank-íe (rank-(óm (rank-(om
gálv-íe gálv-(óm gálv-(om
2iēm-íe 2iēm-(óm 2iēm-(om
(rank-os (rank-(ij (rank-(óms (rank-ós (rank-o(mís (rank-o(sé
gálv-ōs gálv-(iļ gálv-(óms gálv-ós gálv-o(mís gálv-o(sé
2iēm-ōs 2iēm-(y 2iēm-(óms Ziēm-ós 2iēm-o(mís 2iēm-o(sé
nom. sg. gen. dat. acc. instr. loc. voc.
nom./voc. pl. gen. dat. acc. instr. loc.
(líp-ōs (lip-(i (líp-(oms (líp-ós (líp-o(mis (líp-o(sé
In nominal stems of more than one syllable, any of the syllables may be underlyingly accented — although, as we shall discover, this was not the case at an earlier stage (cf. §§5.1.1, 5.2-3). De Saussure's Law, on the other hand, applies only to a syllable containing a circumflex long vocalic or short vowel followed immediately by a syllable containing an acute long vocalic. It does not operate across intervening syllables: hence the instr. sgs. to kātil-as 'kettle' and pasiuntin-ys 'messenger, envoy', acc. pāsiuntin-ļ, with underlyingly unaccented stems /katil-/, /pasjuntin-/, are kātil-u, pāsiuntin-iu, as the acute ending l-úol (> -u by Leskien's Law) is separated from the word-initial syllable by one or more intervening syllables. As a result, nominal stems are distributed among the four accentual classes of Lithuanian according to the following table: 1
2
/*' * i*
4
(• r*'
*
f*
*
*
*
f*'
*
t*
*
*
*
/•*
3
*
*
*
*
*
*
126
*
Note that class 4 includes only monosyllabic stems: since unaccented stems of two or more syllables will surface with default stress on its initial syllable, which is not affected by an acute vowel in the ending, those must all belong to class 3. The four accentual classes of Lithuanian find their correspondents in the intonations on long vowels and diphthongs Latvian initial syllables. As noted in §4.2.1, these may be either rising (~), falling O , or "broken" (~). Endzellns (1899, 1913, 1922) has demonstrated that falling syllables correspond to Lithuanian nouns with circumflex intonation on the stem vowel, i.e. nouns belonging to class 2 or 4. To Lithuanian nouns with acute stem vowel, Latvian cognates exhibit one of two possible intonations: rising, if the noun in Lithuanian belongs -(or originally belonged) to AP 1; "broken", if the noun belongs to AP 3. Cf. the correspondences below: Lithuanian vyras 'man, husband' brólis 'brother' liepa 'linden tree' dúona 'bread' sáulé 'sun'
Latvian vīrs brālis liēpa duðna saúle
AP3 galvá 'head', ace. gálvq
galva kaīns sifds
AP2 rankā 'hand', ace. rankq
ruoka llek
kálnas 'mountain' sirdis 'heart', ace. slrdļ
pres. 3sg. liēka 'remains, leaves over/behind' AP4 iiema 'winter', ace. tiēmq dravgas 'friend' ¿qsis 'goose', ace. iqsļ
zlema dráugs zúoss
The intonations on Latvian initial syllables may thus be derived from the underlying twofold contrast of accentual classes in Lithuanian. The contrast between barytone and mobile has been eliminated in stems with Proto-(East-)BaItic circumflex stem vowel, but continued in stems with acute stem vowel as rising vs. broken, respectively. Thus whereas Slavic languages exhibit three contrasting accentual classes, Lithuanian, once one subtracts the effects of de Saussure's and Leskien's Laws, contrasts only two, columnar and mobile; the evidence of Latvian intonations supports the reconstruction of the two-way contrast of Lithuanian for East Baltic.
127
It was noted above in §2.2.2 that Sanskrit and ancient Greek contrast only two accentual types in o- and easterns, the two most productive classes of nouns in late and post-PIE: stem-stressed or barytone, and theme-stressed or oxytone. Barytone and oxytone o-stems are probably to be analyzed as underlyingly unaccented and postaccenting, respectively, whereas barytone and oxytone e^-stems have underlyingly accented and unaccented stems, followed by accented *-(eh2 (cf. §2.2.2): o-stems e/12-stems
barytone R o (R (efi2
oxytone R( o R (eh.2
The differences between this reconstructed system and that of Lithuanian and Slavic are obvious. First, the mobile type of Balto-Slavic finds no parallels in the reflexes of o- and easterns in older, accentually archaic IE languages, which always exhibit columnar stress: cf. barytone Hitt. nēwaš 'new', Skt. vfka- 'wolf, sénā- 'army', Gr. logos 'word', kho:ra: 'country', oxytone Hitt. išhāš 'lord', Skt. dhumá- 'smoke', chāyā 'shadow', Gr. thu:mós 'violent emotion', si:gé: 'silence'. As reconstructed in §2.2.1, mobile stress is characteristic of consonant- and semivowel-stems in PIE (including i- and «-stems), yet Lithuanian accentual classes 3/4, and Slavic AP c, contain many important o- and ō-stems. In addition, o- and e/i2 _stems a r e divisible into only two accentually contrasting classes in PIE: unaccented (barytone) vs. postaccenting (oxytone) for o-stems, vs. stem-accented (barytone) vs. unaccented (oxytone) for easterns. In Slavic, on the other hand, one finds a three-way contrast of accent among o- and tf-stems: cf. PS1. accented *d'ym-ú 'smoke', *l'ip-a 'linden tree' vs. postaccenting *dvor-'u 'door', *vidov-'a 'window' vs. unaccented *gðrd-u 'enclosed area, fortification, town', *zim-a 'winter'. The three-way contrast of Slavic therefore raises a number of historical questions. Is it a Slavic-specific innovation? Or was it rather common to Baltic and Slavic, i.e., did Proto-Balto-Slavic possess three accentual classes, with Baltic subsequently losing the third type or merging it with one or both of the others? Although it is generally agreed that Baltic and Slavic have undergone a period of common development in their accentual systems, the prehistory of these accentual classes, and their relation to the system reconstructible for Proto-Indo-European (PIE), has not yet been fully clarified.
128
4.3. "Dybo's Law" and the origin of accented and postaccenting stems In an article published in 1962, Dybo first proposed that the postaccenting class of Russian, Serbo-Croatian, etc. was not inherited from PIE or PBS1, but is instead a specifically Slavic innovation. According to his analysis, pre-Proto-Slavic stems which were underlyingly accented on a non-acute syllable, i.e. which contained either a long vowel or diphthong with circumflex intonation or a short vowel, advanced their stress to the following syllable, thus resulting in the attested postaccenting type with its columnar stress on the syllable following the stem (cf. IlliC-SvityC 1979:140-4). Thus this forward stress shift — now generally referred to as "Dybo's Law" or the law of Dybo and illiðSvityð — is the immediate cause of, and directly responsible for, the postaccenting nominal stems found in Slavic languages such as Russian and Serbo-Croatian. This hypothesis has since become one of the foundations of Slavic historical accentology as established by the "Moscow Accentological School": cf. Dybo 1981:18ff., Dybo et al. 1990:38ff. Today, most other researchers in the area of BS1. also subscribe to Dybo's Law, e.g. Lehfeldt (1993:44ff.), Jasanoff (p.c, class notes); cf. Collinge (1985:313) for additional references. Other scholars have gone even further than Dybo and illiðSvityC in restricting postaccentuation to certain branches of Slavic: cf. Garde (1973), who argues on the basis of Slovincian evidence that the postaccenting accentual class never existed in West Slavic, but peculiar to East and South Slaivc only; therefore Dybo's Law took place after the breakup of Proto-Slavic. Halle (1997:297-8) originally endorsed the latter view, but has most recently opted for a PS1. date of the oxytonic class on the basis of a new interpretation of length correspondences in Czech, Slovak, and Sorbian (cf. Halle 2001). Aside from its validity for West Slavic, however, there is general consensus today that Dybo's Law underlies the postaccenting class of the Slavic languages. To be sure, Dybo's Law neatly explains the complementary distribution of intonation in Proto-Slavic accented and postaccenting stems: the former retain their accent on an acute syllable head, whereas the latter have advanced it from a circumflex or short stem vowel to the following syllable. Yet the forward shift postulated by Dybo fails to account for a curious and hitherto insufficiently acknowledged peculiarity of BS1. accentual reconstruction: both in Lithuanian and in reconstructed Proto-Slavic, underlyingly accented simple polysyllabic stems are always accented on the stem-final syllable. In modern Slavic languages such as Russian or Serbo-Croatian, of course, stems with underlying accent on a syllable other than the last are legion. Yet most of these are demonstrably derivatives or
129
compounds of simple nominal stems, deverbal nouns containing preverb and verbal stem, or loanwords; on evidence for their secondary origin, see §4.3.3. In PIE, nominal and verbal roots consist of only one syllable; the same holds true to a considerable extent in (pre-)Proto-Slavic, so that the number of bases with a non-final syllable is not large. So far as I have been able to discover, however, there are no disyllabic (or longer) nominal or verbal bases for which one must reconstruct underlying accent on a penultimate or earlier stem syllable, e.g. (V V or V (V V or (V V V. Such a distributional restriction is both unexpected and diachronically unmotivated if one supposes, with Dybo and Illið-Svityð, that a rightward stress shift took place in the prehistory of Proto-Slavic and was responsible for dividing the single class of accented stems into accented and postaccenting stems, i.e. APs a and b. 4.3.1. Hirt's Law: Dybo's Law reversed? Dybo's and Illið-Svityð's belief that the Slavic postaccenting stems are an innovation rests on a relatively small set of alternating (mobile) forms with oxytone cognates in other IE languages. Most of the extra-BSl. etymologies adduced by illiðSvityc" (1979) and Dybo (1981:24-6), however, are of dubious merit at best, and a number are entirely impossible.25 The reason for this deficiency is not far to seek. A comparison of a list of lexical items securely reconstructible for PIE with their equivalents in Baltic and Slavic immediately reveals the extensive lexical replacement which has occurred within the prehistory of Proto-Balto-Slavic. To a perhaps even greater degree than Germanic, where PGmc. *geBan^ 'give', *drirjkwana, 'drink', with no good cognates elsewhere in IE, have completely replaced PIE *deh3- and *hjeh2gwh- ~ *peh3-(i-), respectively, the vocabulary of BS1. is replete with forms that find no good equivalents in other IE branches, or whose etymologies are questionable and controversial. Even worse, many words with good PIE etymologies have undergone morphological remodeling, so that they no longer form word equations with their extraBSl. cognates. To take only the most obvious examples, PS1. *bratru, Lith. brólis (< *brotere.~lis) 'brother' and PS1. *sestra 'sister' clearly continue their PIE preforms *bréh2tēr (Skt. brātā, Gr. phrá:tε:r 'member of a phratry', Lai.frater, Olr. bráthair, PGmc. *brōþēr, TB procer) and *méli2tēr (Skt. mātá, Gr. mé:tε:r, Lat. māter, Olr. 25
E.g. PS1. *bhixa 'flea" (Lith. bluso) < PBS1. *blusá: with Gr. psúlla; PS1. *kosa 'scythe* with Skt. káksā 'armpit'; PSI. *véCer-ú 'evening' (cf. Lith. vākaras) < *wek(w,er-os with Gr. hésperos, Lat. vesper < *wesper-os.
130
máthair, PGmc. *mōðēr, TB mācer for *mocer). Yet both have been "thematized", i.e. morphologically adapted to more productive nominal classes, to the o- and d-slems in Slavic, and to the y'o-stems in Lith. (cf. gen. brólió), where the form has in addition been contracted, probably through hypocoristic use. These forms, therefore, are of no probative value for the development of PIE accent in BS1.: just as noted above in connection with the reconstruction of the archaic ablaut-accent classes of PIE (§2.2.1), only perfect or nearperfect word equations can shed light on the evolution of the BS1. accentual system and its diachronic relationship to that of PIE.26 In addition to the paucity of word equations, a further weakness of the comparative support for Dybo's Law rests on a misconception of the accentual system reconstruct!ble for PIE on the basis of other IE languages, namely Sanskrit, ancient Greek, Anatolian, and Germanic. Many Slavists have assumed, on the basis of the two-way division of Lithuanian (classes 1/2 vs. 3/4) and Slavic (APs alb, i.e. accented and postaccenting, vs. AP c, unaccented), that PIE had a similar dichotomy between "barytone" paradigms and "oxytone-mobile" paradigms. The latter, however, is a fiction that conflates disparate layers of the reconstructible PIE grammar: as argued in §2.2, columnar barytone and oxytone thematic nouns, i.e. o- and e/12-stems, belong to a historically later component and must be analyzed separately from the archaic stratum of ablauting roots, suffixes, and endings. As relatively few consonant stems survive in the BS1. languages, the evaluation of Dybo's Law must be based upon a comparison of thematic stems in BS1. and the other IE languages.
26
The same criticism may be leveled against Garde's (1976a:45) two Paradebeispiele of PIE barytone stems with non-acute intonation of the stem vowel corresponding to BS1. oxytone stems, which he cites as evidence for IlliC-SvityC's Law: PIE *ékwos (cf. §2.2.2) - Lith. ašvá (AP 4). PIE *widhéweh2 (Skt. vidhávā) - PS1. *v!dov-a, ace. *-q. The first of these may in fact continue fern. *eéwéh2 or the like, with the originally collective accented suffix *-eh2 receiving stress after an unaccented root (§2.2.3); the second was originally ablauting in PIE, and Indo-Aryan and Balto-Slavic can easily have generalized the stress of the nom./acc. and obi., respectively. A number of items listed by IlliC-SvityC (1979) and Dybo (1981:20-6), Dybo et al. (1990:40-6) are remodelings of PIE root nouns and thus offer little evidence for the PIE prehistory of Balto-Slavic nominal accent: PS1. *mf so, pi. *me.sa 'meat' <— PIE *mf ms- - *méms- (Skt. māmsám, PGmc. *mimzao > Goth, inimz); PS1. *zvéri, Lith. ¿vérís, Latv. zvérs 'beast, wild animal' <— PIE *ghwer- (Gr. /V:r, -as); PSI. *g9sl, Lith. Zqsis 'goose' <— PIE *ghans- (Gr. khé:n, -ós\ cf. Skt. hamsás). Unlike e.g. Indo-Aryan, which has developed oxytone stress in e.g. māmsám, hamsás, BS1. appears to have maintained the root as unaccented.
131
According to Dybo and IlliC-SvityC, there should be no Slavic postaccenting stems with oxytone cognates in other IE languages. Yet the following word equation is, at best, difficult to dismiss: PS1. *div-u, gen. *-'a 'demon' (SC div/di:v-'0/ 'giant, demon', gen. dív-aláv.v-al) < PIE *deyw-ós 'god' (vrddhi formation to *dyéw- ~ *diw- 'sky, heavens'): cf. Skt. devá-, OPr. deiws, gen. deiwas, ace. deiwan, etc., never with a macron over the ei, suggesting a columnar oxytone paradigm (§4.4.2); cf. Lith. Dievas, pi. dievdi, AP 4. Here we have a thematic noun of impeccable IE pedigree, clearly postaccenting in Sanskrit (and almost certainly in Old Prussian); in SC, the only modern Slavic language to have preserved the base word (rather than derivatives or likely secondary formations such as Russ. divnyj 'wonderful', divo 'wonder, miracle'), it belongs to the oxytonic class.27 Is coincidence excluded? Other nouns may likewise continue PIE preforms with columnar oxytonesis: cf. especially the following candidates, where the loss (and in many Slavic languages, restiitution) of weak jer makes it impossible to determine whether the stem was unaccented or postaccenting in PS1.: PS1. *mígl-'á, ace. *migl-Q 'mist' (cf. SC Cak. [Novi] magl-á, ace. -u) < PIE *h3inigh-léh2: Gr. omikHe: with shifted stress? (Lith. miglā, AP 2 in Dzuk dialects —> AP 4); PSI. *snúx-'á, ace. *snúx-'o. 'daughter-in-law' (SC Cak. [Novi] snah-á, ace. -u, pi. -e, Russ. snox-á, -ú, pi. [17th c ] -i, mod. snóx-i) < *snus-éh2: Skt. snusā, OE snoru < PGmc. *snuzō (replacement of PIE *snus-ós, cf. Gr. twos; Lat. nurus, secondary «-stem?). In addition, two important PSI. postaccenting tf-stems may plausibly be explained as remodelingsof PIE ablauting stems, with generalized columnar stress on the suffix *-eh2-: PSI. *2en-a, ace. -Q 'woman, wife' (cf. Russ. ¿en-á, -ú, SC ien-a, -u) < PBS1. *gen-á: <— PIE *gwén-h2- ~ *gwn-éh2-, with a proterokinetic paradigm reflected directly by Olr. neut. bé < *gwēn < *gwénh2 (by Szemerenyi's Law, cf. Jasanoff 1989), fern, ben < *bena < *gweneh2 <— *gwenh2, gen. mná < *bnās < *gwneh2-s, and indirectly by Gr. Boiot. band: < *gwņ(n)ā vs. Att.-Ion. gunε: < *g w enā, GAv. Ja'ni- < *gwenh2- (voc. pi. jānaiiō) vs. LAv. γ3nā- < *gwneh2- (gen. pi. γ^nanam), Ved.jani- vs. gnās(pati-) '(husband) of a divine 27
The meaning of div, along with *bogú 'god', betrays the influence on the early Slavic tribes of Zoroastrianism, almost certainly via the nomads of the steppes to the north of the Black Sea: cf. Av. bagō, OP baga 'god', Av. daeuuō, OP daiva 'demon' <— '(false) god'.
132
wife'. Here Balto-Slavic, like both Attic and Boiotian Greek, has generalized the suffix stress of the weak stem; Indo-Iranian, on the other hand, has rebuilt 'woman' as an /-stem with columnar stem stress, starting from nom. sg. GAv. Ja'ni-, Ved.jani-. PS1. *v!dov-'a, ace. *-'Q 'widow' (cf. Russ. vdová, -ú, pi. vdovy—> vdóvy) < PBS1. *widew-á: <— PIE *widhéw-h2 ~ *widhw-éh2-, e.g. nom. *widhéw-h2, gen. *widhw-éh2-s (for the ablauting suffix, cf. Skt. vidhávā, PBS1. *widewá:, Greek ε:(w)íthe(w)os < *-ew- vs. Goth, widuwō, Lat. vidua < *-(u)w-). Sanskrit has generalized the strong stem stress of *-éw-, whereas Balto-Slavic has generalized the suffixal stress of the weak alternant; on OPr. widdewu, dat. pi. widdewumans, cf. §4.4.2. Based on these etymologies — and keeping in mind that the massive morphological and derivational remodeling of inherited PIE lexical stock in BS1. considerably reduces the number of potentially probative word equations 28 — I propose that the Slavic postaccenting class is not an innovation, but directly continues the inherited oxytonic type of Sanskrit devá- 'god' or Greek zugón 'yoke'. The complementary distribution of APs a and b in PS1., discovered by Dybo and attributed by him to a rightward shift from a nonacute vowel, may therefore be considered as the result of retraction of post-stem stress to a preceding acute vowel, i.e. one originally containing a laryngeal in (post-)PIE. Such a hypothesis immediately accounts for the synchronically unmotivated restriction of underlying accent to the final syllable of polysyllabic stems in Slavic: these stems were postaccenting in pre-PSl., with surface stress and underlying accent retracted to a stem-final syllable with acute intonation by "reversed Dybo's Law". This retraction is hardly an original suggestion in the study of BS1. historical accentology. Indeed, it immediately recalls Hirt's observation of over a century ago that a number of Lithuanian nouns with "Stosston", i.e. acute intonation of the root, correspond to oxytone forms in other IE languages (Hirt 1895:94). Although Hirt does not explicitly formulate a sound change to account for this discrepancy, later scholars expanded Hirt's 28
Note that although PS1. *jigo (cf. Russ. igo from Church Slavonic) clearly continues PIE *yug-ó-m 'yoke' (Vedic yugá-, Greek zugón; cf. Hitt. iugan, never "iuigan"), Bulg. igo, -to may reflect PS1. s-stem *j!go, gen. *jlf -es-e: cf. PS1. *oko, gen. *oð-es-e 'eye' (Slov. ok-ó, oð-és-a. with ci stern endings) < *h3ékw-os, *-es- alongside du. *oCi < *h}ékw-ihi to PIE root noun *h3ekw-; PS1. *kolo. gen. *kol-es-e 'circle, wheel' (Slov. kol-ó, kol-és-a) < *kwékwl-os, *-es- <— PIE o-stem *kwék*l-o-s (with rounding of *e between two labiovelars? cf. TB kokále 'chariot' < PT *kwákwlé <— *kwékwIos). Outside of Slovene, j-stems have for the most part become ordinary neuter o-stems, e.g. PS1. *slov-o 'word, speech, song', pi. *slov-es-a —> Russ. slová, SC slova, Bulg. slová, Pol. stowa, Cz. slova (Cz. sloveso 'verb' backformed to slovesa; Slov. slovó, slovésa), *kolo 'circle, wheel', *kol-es-a —> pi. tant. SC kola, Bulg. kolá 'wagon, car' (Russ. kolesó from Church Slavonic). The discrepancy between Bulg. igo and expected PS1. *jIgo is thus only apparent: the initial stress of the Bulg. form would be regular from a proterokinetic i-stem, cf. §2.2.1.4 (root underlyingly unaccented).
133
original statement into what is now generally known as "Hirt's Law": at some stage common to both Baltic and Slavic, i.e. in pre-PBSl., PIE oxytonic stems retracted their stress to a preceding long vowel. In terms of our current conception of PIE phonology, the long vowels which trigger Hirt's Law are actually sequences of vowel or diphthong plus tautosyllabic laryngeal, i.e. *VH (including *R.H) or *VRH. The following are among the columnar stem-accented BS1. forms with indisputable oxytone cognates in Sanskrit and/or Greek (cf. Dlie-SvityC 1979:57-64, 138-9): Lith. dúmai (AP 1), Latv. dvmi (pi. tant.), Russ. dym, -a 'smoke' < PBS1. *d'ú:mas < PIE *dhuHmos 'breath, smoke' (Skt. dhvmá-, Gr. thu:mós 'violent emotion'); Lith. dúona (AP 1), Latv. duðna '(hunk of) bread' < *PBS1. *dó:n-a: < PIE *dhoHnéh2 (Ved. dhānās 'grain', pi. tant.); Latv. grlva 'mouth of a river' < PBS1. *gri:v-a: < PIE *griHwéri2? (Ved. grlvā 'nape of the neck'); Latv. ilgs, SC dug 'long' < PBS1. *dilg-as < PIE *dļh1ghós (Skt. dīrghá-, Gr. dolikhós (?); Lith. ilgas, AP 3 with relics of AP 1, but why no d- in Baltic?); Latv. pilns, SC pun, -a 'full' < PBS1. *p fln-as < PIE *pļhinós (Skt. purņá-; Lith. pilnas, AP 3 with barytone forms in OLith.); Lith. tiltas (AP 1 in standard language and most dialects), Latv. tilts 'bridge' (Finn. silta <— neut. *tilta) < PBS1. *firt-an < PIE *trHth2óm? (Skt. tīrthám 'ford, path to watering place') Lith. vyras (AP 1), Latv. vīrs 'man, husband' < PBS1. *wir-as < PIE *wiHro(Ved. virá-); Lith. ¿irnas 'pea', SC zrno 'corn' < PBS1. *Zirn-an < PIE *gŗh2nó- (Skt. jlrņá-). As there is general agreement that tautosyllabic laryngeals are the major source of the BS1. "acute" (cf. §4.2.2), it follows at once that Hirt's Law accounts for the complementary distribution of acute vs. circumflex intonation in accented vs. postaccenting stems observed by Dybo. Although all major researchers have acknowledged the validity of Hirt's Law (see the references in Collinge 1985:81-3), it has played an amazingly minor role in historical studies of BS1. accentology and seems to be regarded by most as an afterthought, of interest only within the wider picture of the PIE background of BS1. accent or as an explanation of isolated diachronic peculiarities: cf. Garde 1976a:333-4, Dybo 1981:17-8, 33-9, 233ff., Derksen 1985:77-8, Dybo et al. 1990:50. Each of these studies first introduces the various accent shifts which may be reconstructed from internal comparison of the BS1. facts, e.g. Dybo's Law, Stang's Law, etc., then mentions Hirt's Law, but
134
without placing it chronologically with respect to the other proposed laws or assessing its significance for the accentual paradigms of BS1. Yet if Hirt's Law depended on the presence of a pretonic laryngeal (i.e. vowel length, in pre-laryngeal terms) in PIE oxytonic stems, it must have been a very early change — perhaps the earliest recoverable accent shift in the post-PIE history of BaltoSlavic.29 Of course, the retraction in question could have taken place after PIE laryngeals had been lost, after the contrast between syllables of the form *VH, *VRH and *VR, *VHV had been reinterpreted as acute and circumflex intonation, respectively, in ProtoBalto-Slavic. Nevertheless, the existence of examples such as those listed above in both Slavic and Baltic makes it clear that Hirt's Law was an innovation of the pre-PBSl. period. I therefore conclude with Dybo that pre-PSl. had a single immobile accentual paradigm from which PS1. accentual paradigms a and b are descended, but that the prePSl. paradigm was postaccenting and, at least in its core, directly continues the columnar oxytonic type of PIE. The Lithuanian cognates adduced above suggest that this retraction is of pre-Proto-Balto-Slavic date, as already suggested.30 Putting together the Hirt's Law examples with the PS1. and (OPr.) postaccenting stems of PIE antiquity, we obtain the list below of BS1. nouns which continue PIE stems with columnar stress on the thematic vowel: OPr. deiws, deiwas, deiwan, etc. (*[deiw-as, -an]), PS1. *dív-ú, gen. *-a 'demon' (SC div 'giant, demon', gen. div-a) < PIE *deywós 'god' (Lith. diēvas, pl. dievai, AP 4); Lith. dúmai (AP 1), Latv. dvmi (pl. tant.), Russ. dym, -a 'smoke' < PBS1. *d'ú:mas < PIE *dhuHmós 'breath, smoke'; Lith. dúona (AP 1), Latv. duōna '(hunk of) bread' < *PBS1. *do:n-a: < PIE *dhoHnéh2; Latv. grīva 'mouth of a river' < PBS1. *gfi:v-a: < PIE *griHwéri2 (?); Latv. ilgs, SC dug 'long' < PBS1. *dflg-as < PIE *dļhighós (Lith. llgas, AP 3 with relics of AP 1); PS1. *mígl-á, ace. *migl-'o. 'mist' (SC oak. [Nov!] maglá, ace. maglu) < PIE *h3mighléh2 (Lith. miglá, AP 2 in Dzuk dialects —> AP 4); Latv. pilns, SC pun, -a 'full' < PBS1. *p'iln-as < PIE *pļh1nós (Skt. purņá-; Lith. pllnas, AP 3 with barytone forms in OLith.); Lith. tlltas (AP 1), Latv. tilts 'bridge' < PBS1. *tirt-an < PIE *trHth2om (?); 29
Some scholars, particularly Kortlandt, have supposed that PIE laryngeals survived well into the prehistory of Baltic and Slavic, being lost at different times in different positions. Support for such a hypothesis is lacking, however; as I hope to show elsewhere, the BS1. facts are consistent with an early loss of laryngeals as segments, the contrast between syllables with and without laryngeal being continued by, or reinterpreted as, the distinction between acute and circumflex intonation in pre-PBSl. 30 On the disappearance of the columnar postaccenting paradigm in Lithuanian, see §4.4.3.
135
OPr. widdewú(dat. pi. widdewúmans), PS1. *vídov-a, ace. *-q 'widow' (cf. Russ. vdová, -ú, pi. vdovy —> vdóvy) < PBS1. *widew-á: <— PIE *widhéw-h2 ~ *widhw-éh2-; Lith. vyras (AP 1), Latv. vīrs 'man, husband' < PBS1. *wlr-as < PIE *wiHró(Ved. vīrá-); PS1. *2en-'a, ace. - Q 'woman, wife' (cf. Russ. ien-á, -ú, SC ién-a, -u) < PBS1. *gen-á: <— PIE *gwén-h2- ~ *gwn-éh2-; Lith. iirnas 'pea', SC zrno 'corn' < PBS1. *2 irn-an < PIE *grh2nó-. If the Slavic APs a and b continue PIE oxytone o- and easterns (with and without Hirt's Law retraction, respectively), it should follow that nouns of AP c, i.e. underlyingly unaccented stems with mobile stress, continue PIE barytone stems. Recall from §2.2.2 that the latter are themselves unaccented, with default initial stress; this stress has been reanalyzed as underlying initial accent in Sanskrit and Greek, neither of which offers evidence for underlyingly unaccented nominal forms or "enclinomena" of the Slavic type. The following impeccable PIE word equations suggest that the unaccented class of o-stems has survived as such into Slavic, complete with the ability to "retract" stress to a proclitic: PS1. *goji 'abundance' (SC goj, gen. goj-a 'peace') < PIE *góyH-os: Ved. gáyas 'house and home'; PS1. *logii 'flat place, lair' (SC log, gen. log-a 'lair, den', Russ. log, gen. lóg-a 'ravine') < PIE *lóghos: Gr. lókhos 'ambush'; PS1. *sunu 'sleep' (Bulg. def. san-at < *sunu t'u; transferred to AP b in SC sdn, gen. sn-áa, Slov. sen, gen. sn-a) < PIE *supnos: Gr. húpnos, alongside *swépnos > TB spam (Ringe 1987:258ff., 1991:149-50, R. Kim 1999:164-5), OE svefn, *svópnos > Arm. k'un (Ved. svápnas Av. xvafnō, Lat. somnus from either), cf. Schindler 1966; PS1. *vilku 'wolf (SC vuk, gen. vuk-a, Slov. vdlk, Bulg. def. valk-at < *vllku t'u, Russ. volk, gen. vólk-a, cf. ná volka, ná volky in 1647 Sbornik) < PIE *vļkwos: cf. Ved. vŗkas, Gr. lúkos (Lith. vilkas, AP 2 —> mod. AP 4); PS1. *vozu 'cart' (SC voz, gen. vóz-a, Kri2anic ná vozu, Slov. voz, Russ. voz, gen. vóz-a (16th c. ná vozy, dial, pó vozu) < PIE *woghos, deverbal noun to *wegh'convey': cf. Gr. 6khos 'chariot'; PS1. *zqbu 'tooth' (SC zub, gen. zub-a, Bulg. def. zab-ét < *z
136
The following are also potentially cognate with PIE barytone o-stems, but their probative value is decreased by the likelihood of borrowing or the weakness of the comparative evidence:31 PS1. *bogu 'god' (SC bog, gen. bog-a, Slov. bog, Bulg. dial, bogé < *bogú t'ú, Russ. bog, gen. bóg-a, 16th c. ná boga, ná bogy, pó bozēa) < *b h ógos 'distributor of blessings': Ved. bhágas 'distributor of blessings, god', GAv. bagō 'god' (but the meaning is clearly due to Iranian influence, cf. fn. 27, so a borrowing from Iranian is probable); PS1. *gromú 'thunder' (SC dial, gróm, gen. grom-a, Slov. grom, Russ. grom, gen. gróm-a, cf. ORuss. ót gromu, ó gromex) < *ghrómos 'crash': Gr. Hesykh. khrómos ■ psóphos poiós 'some kind of sound'; PSI. *snégú 'snow' (SC snég, gen. sneg-a, Slov. sneg, Bulg. sneg-ét < *sné"gú fu, Russ. sneg, gen. snég-á) < PIE *snóygwhos: Prakrit siņeha-, Goth, snáiws (Lith. sniēgas, AP 2 in northwest Zemaitian —> AP 4 in modern standard and most dialects; OPr. snaygis, Elbing Vocabulary) — but the accentuation of the Indo-Aryan form is not attested, and the o-stem may be a replacement for the root noun *snigwh- > Lat. nix, nivis, Gr. ace. nipha (agá-nniphos 'very snowy'), if not deverbal to *sneygwh- 'stick (of snow)' (LIV:521-2; cf. Hoffmann 1965); PSI. *vðlsú 'hair' (SC dial, vlās, gen. vlas-a, Slov. las, dial, vlás, Bulg. dial. def. vlas-ó, etc. < *vlasii fu, Russ. vólos, gen. vólos-a, zá volosy) < *wóllcos: Skt. válšas 'sprout, branch'. Thus in direct opposition to Dybo and Illie-SvityC's hypothesis that the Balto-Slavic barytone (> Slavic accented and postaccenting) and mobile paradigms continue PIE barytone and "mobile-oxytone" stems, respectively, I conclude that PIE barytone o-stems, underlyingly unaccented and postaccenting, are continued by unaccented stems of AP c in Slavic. On the other hand, oxytone o-stems, which are underlyingly postaccenting, undergo a conditioned split: they are continued by postaccenting stems (AP b) if the nucleus of the syllable preceding the thematic vowel does not end in a laryngeal, in which case the . If, however, the nucleus of the final stem syllable does end in a laryngeal, stress is retracted by Hirt's Law, resulting in "pre-accented" stems in BS1. (PSI. AP a, Lith. AP
31
1 omit forms such as PSI. *vðrmī 'raven', *gðlsú 'voice', for which the comparative evidence cited by IlliC-SvityC is limited to Lithuanian and/or Germanic (respectively Lith. vafnas and ON kail 'call' < PGmc. *kalza.). Slavic forms with only Baltic cognates, of course, tell us nothing about the relationship of the (Balto-)Slavic APs to those of PIE. The Gmc. evidence for stress, which is limited to Verner's Law reflexes (cf. §1.3), is not always clear or reliable; in addition, Baltic and especially Slavic possess many Germanic loanwords, and linguistic contacts between Gmc. and BS1. undeniably date far back into prehistory, so that for any individual lexical item, one cannot exclude the possibility of parallel accentual innovations shared by both branches.
137
1); the stem-final syllable receives acute intonation, which is preserved in Lithuanian and whose reflexes in the various Slavic languages have been described in §§4.2.1-2.32 PIE barytone (unacc.) oxytone (postacc.)
PBS1. unaccented accented postaccenting
PS1. c a b
Lith. 3/4 1 (see §4.4.3)
4.3.2. A restatement of Hirt's Law within the brackets-and-edges framework So far, the only support for a leftward shift to an acute syllabic head (i.e. one closed by a laryngeal in PIE) rather than Dybo and Ulið-Svityð's rightward shift from a non-acute syllable has been the existence of Slavic and Old Prussian postaccenting stems continuing PIE or post-PIE oxytone forms, and the undeniable examples of retraction which have traditionally been cited for Hirt's Law. The set of Slavic and OPr. oxytone o- and 5-stems with absolutely certain PIE derivations, however, is extremely small, if difficult to dismiss. It is therefore fair to ask whether Balto-Slavic offers any other empirical verification for Hirt's Law in place of Dybo's Law. Any defense of a retraction in postaccenting stems must first reckon with the alleged counterexamples which have led to its gradual abandonment or deemphasis over the last century. The following nouns, which appeared above as examples of acute intonation from a PIE syllable-final laryngeal, do not exhibit the expected retraction: we find AP 3 in Lithuanian and AP c in Slavic, instead of AP 1 and a, respectively. Lith. gyvas (AP 3), Latv. dzivs, PS1. *Zivu (AP c, Russ. Ziv, -á, -o) 'living, alive' < PIE *gwih3-wó- (cf. Skt. jlvá-) Lith. tqvas (AP 3), Latv. tiévs 'thin, slender' < PIE *tenh2-wo- (cf. Gr. tanaós <— *tenawós) I can offer no new explanation for these exceptions at present, other than to speculate that Hirt's Law may have been subject to certain restrictions, the exact nature of which will require further investigation. Interestingly, Ebeling (1967:582) has observed that «-stems with acute stem vowel are apparently not subject to Hirt's Law, most important *su:n-u-. In these nouns, the stem syllable contains an acute long vowel and therefore must descend from a PIE sequence *VH. Yet the Lithuanian and Slavic forms do not exhibit the columnar stem
32
The development of e/i2-stems differs in certain respects from that of o-stems: see §5.1.2.
138
stress of e.g. Lith. démai, Russ. gen. dúm-a 'smoke' < PBS1. *dú:m-a- < PIE *dnuHm-o, Lith. pilnas, SC pún 'full' < PBS1. *pilnas < PIE *pļhi-nó-s; instead they fall under class 3 in Lithuanian, with mobile accent, and the mobile class in Slavic. If Hirt's Law applies to thematic nouns and adjectives such as 'smoke' or 'full', why then does it fail to occur in the above forms? A ready explanation for this contrast is available within the brackets-and-edges model, which allows a more precise statement of the conditions under which Hirt's Law applies. As traditionally formulated, Hirt's Law retracts final stress in any oxytone form to a preceding acute long vowel. I propose rather than the retraction in question crucially depends on underlying accent, and not merely on surface stress. Specifically, an accentual specification (denoted by a left bracket) is retracted by one syllable if the preceding syllabic head is marked [+acute], i.e. consists of an acute long vowel. [+acute]
[+acute]
Thus, only sequences of the form * (* are subject to retraction and become (* *, while unaccented sequences * * remain unaffected.
Because «-stems in pre-PBSl. were
presumably unaccented — like proterokinetic w-stem roots in PIE (§2.2.1.3) — they were not affected by Hirt's Law: hence PIE *sun-u- 'son' gives Lithuanian sún-ús (<— *sdnus), sún-aús, sún-ui, sén-y, etc., with the mobile stress of AP 3. 33 So long as our investigation is confined to simple nominal stems, the internal BS1. evidence may be reconciled with a stress shift in one direction or the other. The system of BS1. verbal accentuation, however, offers at least two examples of suffixes whose accentual behavior proceeds straightforwardly from Hirt's Law, but suffers difficulties under the hypothesis of Dybo's Law. Like nominal stems, verbal stems are distributed among three APs a, b, and c, in which a and b continue a single columnar paradigm (with stem-final stress and Dybo's Law advancement from non-acute syllables according to Dybo, or postaccenting with Hirt's Law retraction to acute stem-final syllables, as argued here).
The stress patterns of these in the present are as follows.
For details of
reconstruction, see chapter 6, passim (especially §§6.3.1, 6.6); on Stang's Law see §§4.2.1, 6.6. 33
Contra Kortlandt (1978:275fn.5), there is no need to assume that stress had become fixed columnar on the second syllable of *sun-us in pre-Lith. Note that western Lith. (Zemaitian) su'n-us may reflect an extension of Hirt's Law, rather than leveling from an originally end-stressed paradigm as per Derksen 1991:59, 78.
139
1 2 3
*s^d-e-ši *s^d-e-ti
pl. *s^d-e-mú *s^d-e-te *s^d-q-fí
APb
1 2 3
*mog-9 *mo2-é-ši > *moZ-e-ši *moZ-é-tí > *moZ-e-tí
*moZ-é-mi *moZ-é-te *mog-Q-tí
APc
1 2 3
*nés-Q *nes-e-ší *nes-e-ti
*nes-e-mu *nes-e-té *nes-ņ-ti
sgAPa
> *mðZ-e-mú > *móZ-e-te > *mog-Q-fí
The PS1. infinitive suffix *-ti is accented, so one would expect the unsuffixed infinitive to be stressed on the root for verbs of AP a, and on *-ti for verbs of AP b and c. This is not what we find, however: a number of verbs belonging to AP c exhibit rootstressed infinitives. APa
SC sedem, sédeš Russ. sjádu, sjádeš'
sēsti sest'34
APb
SC mógu, moieš (Russ. mogú, móieš'
móci moð')
APc
SC nésēm, lpl. nesémo Russ. nesú, nesēš';
nésti nestí
SC trésēm, trésēmo Russ. trjasú, trjaséš'
trésti trjastí
SC grízēm, grízēmo Russ. gryzú, gryzeš'
gristi gryst'
SC kládēm, kládēmo Russ. kladú, kladeš'
klásti klast'
but
Garde (1976a: 121-4) concludes that the infinitive suffix is "dominant négatif', i.e. that all infinitives are unaccented with default initial stress, regardless of the underlying
^4In Russian, the vowel of the PSI. infinitive ending *-ti is apocopated everywhere except when stressed and preceded by [s], e.g. nesti 'carry', vezti 'drive, transport', mesti 'sweep', vesti 'lead' to the roots /nes-/, /vez-/, /met-/, and Ned-/, respectively. Thus Russian is probative for determining the accentuation of the PSI. unsuffixed infinitive only for roots ending in a coronal obstruent.
140
accentuation of the root. As a result of Dybo's Law, infinitives to roots of AP b and to circumflex roots of AP c have advanced the stress to the suffix, e.g. pre-PSl. *mog-ti > PS1. *mot'i (AP b), *n'es-ti, *trfs-ti > *nes-t'i, *tr|s-fi (AP c). Other than the vocative, however, "dominant recessive" suffixes or morphological categories do not appear to be reconstructible for PIE (cf. chapter 2), and many such suffixes in Baltic and Slavic are clearly secondary and language-specific, as Garde himself admits (ibid., 91, 186-7 and passim). As Dybo (1981:233-44) has demonstrated, the split of AP c by root intonation is simply the result of Hirt's Law, i.e. it is acute unaccented roots which have retracted the stress from the inf. suffix *-ti:35 AP a (V AP¿? V(
(sés mog
(ti (ti
APc
klád tr^s
(ti > (klád ti > PS1. *klásti by Hirt's Law (ti > PS1. *tr?sti
V V
> PS1. *sēsti > PS1. *mot'i
Where two or more suffixes follow in succession, Hirt's and Dybo's Laws can be tested against each other for their explanatory capacity. Consider the case of the productive class of Slavic verbs in *-ov-a-ti, pres. *-u-j e / 0 -. The preconsonantal form of the suffix *ov- must be *-ú- with acute intonation, for its SC reflex is uniformly short -u-, as in the examples below. Garde (1976a: 171-3) analyzes the suffixes in question as inf. *-(ov-(a-, pres. *-(u-(j e / 0 -. Thus if the root is postaccenting, stress should fall on *-ov- in the infinitive but on *-u- in the present, e.g. *(med'-(ov-(a-ti > *med'(-ov(-(a-ti > *med'ovati, *(med'-(ú-JQ > *med'(-(ú-jņ > *med',ujo. 'limit' (to postaccenting *med'a 'boundary' > Russ. meíá, SC méðá), with both stems undergoing advancement of stress from the nonacute root syllable by Dybo's Law. Yet alongside e.g. Russian meiúju, -úješ', SC gréhujēm,
-ēš 'sin' (to gr(ij)éh, gen. gr(ij)éha
35
'sin'), we find not
"meZovat"',
Note that the intonation of the root vowel would have been neutralized in all forms of the pres. of AP c: in the Isg., which is unaccented, and in the other forms, where it was pre-pretonic (cf. §4.2.2, fn. 21). Thus in all verbs of AP c, the root vowel is long in SC and short in e.g. Slovak. Contra Kortlandt 1974b:298, 1975:3, 81 (cf. Schrijver 1991:228-9), the contrast between Russ. pret. fern, pilá, SC pila /pll-a/ 'she drank' and Russ. bfla, SC bila /b'il-a/ 'she was' is not probative evidence that PIE *Hi, *Hu did not trigger Hirt's Law, only later merging with *iH, *uH as *I, *u\ As SC inf. phi < PS1. *piti < *pi:- demonstrates, the stem vowel did trigger retraction and so was acute when Hirt's Law operated. The stress of Russ. pilá, SC pila (and SC pi. neut. pila) must be analogical to PS1. masc. *'pi-hl, neut. *pi-l-o, pi. masc. *pi-l-i, fern. *pi-l-y, which remained unstressed and unaffected by Hirt's Law (cf. Dybo 1981:245, Dybo et al. 1990:74); in many Stokavian SC dialects and the standard language, the other non-masc. sg. forms later also acquired ending-stress, hence sg. neut. pilo, pi. masc. pili, fern. pile. The merger of *CIHC and *CHIC (*I = *i, *u) was already PIE: cf. Mayrhofer 1986:173-5.
141
"gréhovati", but meZovat', grehbvati.
Garde appeals to the analogy of infinitives to
unaccented roots such as *zimovati, *zimuje/o- 'winter, hibernate' (> Russ. zimovát', zimúju, SC zimóvati, zlmujēm), which under his analysis are underlyingly *zim-(ov-(a-ti > *zim-ov(-(a-ti > *zimovati, *zim-(ú-jo > *zim'uJQ. It is less than obvious, however, why the stress of the latter, admittedly larger group of verbs in *-ovati should have been generalized so completely to derivatives of postaccenting roots that no trace of the expected accentuation *-'ov-a-ti survives. The solution lies first in recognizing that the inf. suffix *-a- is unaccented, as I hope to argue elsewhere. Given the accentuation of e.g. Russ. zimovát', SC zimovati, however, the suffix *-a- after unaccented roots must be stressed; therefore the preceding suffix *-ovmust be postaccenting. In the preconsonantal allomorph of *-ov(-, namely *-ú(-, the operation of Hirt's Law would have produced accented *-(u-. As a result, the stress in the present surfaces one syllable to the left of the stress in the infinitive: inf. *meZ-ov(-a-(ti > *meZovati > Russ. meZovat', with no accentual displacement, vs. pres. *meZ-ú(-(JQ > *meZ-(ú-(JQ. > *meí'ujq > Russ. meZuju, with stress retracted to the acute suffix *-u- by Hirt's Law.36 I therefore conclude that Hirt's Law, far from a marginal footnote in BS1. accent, was a — if not the — pivotal shift in the evolution of PBS1. accent from PIE. Not only does the assumption of a leftward shift to syllabic heads of the shape *VH, *VRH help to connect the distribution of BS1. accentual paradigms of o- and eh2-sierc\s with the twofold division of barytone vs. oxytone thematic stems in PEE, but its superior explanatory power accounts for the stress patterns of PS1. infinitives and derived verbs, which otherwise require recourse to ad hoc analogies or accentual properties of suffixes. Dybo's Law, then, is an unnecessary addition to the multitude of accentual "laws" for which BS1. accentology is so infamous: in its place, one should posit a "reverse Dybo's Law', i.e. Hirt's Law, which, although long established and generally accepted, only now assumes a much more prominent role in the prehistory of BS1. accent than before. If Dybo's Law is really just a special case of Hirt's Law, it follows that, unlike Dybo and Illic-Svityc, who posit their rightward shift for Proto-Slavic, or Garde, who limits it to South and East Slavic only, the retraction is of Proto-Balto-Slavic date and therefore common to the Baltic languages as well. It remains to be seen whether the other
36
0n the accentuation of the pres. suffix *-j e / 0 -, see §6.6. 142
accentual properties of the Lithuanian noun and verb are likewise consistent with this hypothesis. 4.3.3. Nominal compounds and the rise of accented stems This chronological difference between accented stems on the one hand and unaccented and postaccenting stems on the other finds unexpected and, to my knowledge, previously unnoticed confirmation in the nominal accent of Serbo-Croatian. Just as all vocatives in Sanskrit receive initial stress regardless of underlying accent, SC exhibits intial syllable stress not only in unaccented nouns such as bog 'god', voc. boī-e, d(j)évēr 'husband's brother', voc. d(j)ever-e, but also in postaccenting nouns: Dalmatinac 'Dalmatian', gen. Dalmatinc-a, voc. Dálmatīnð-e; vójnīk 'soldier', gen. vojnik-a, voc. vojnīð-e; ¿ivot 'life', gen. ¿ivot-a, voc. ¿lvot-e, fern, séstra 'sister', voc. séstro (Leskien 1914:348-9).37 Strangely, however, this rule does not apply to o-stems with underlying accent: the vocative of Posavac 'Posavian, native of the Sava river valley' is Pdsāvð-e., with the same stress placement as the rest of the paradigm.38 One could of course attribute the stress of these vocatives to "analogy" from the other case forms, but why would postaccenting stems not have been subject to such analogy? Clearly, the restriction of the "vocative initial stress" rule to postaccenting o-stems in SC is synchronically unmotivated. For that reason, it is likely to preserve some peculiarity of an earlier stage of the grammar, and to find its origin within the history of the language. Resetar (1931a:20-2), in his study of alternations of stress and vowel length in standard SC (and in the dialects of his native Dubrovnik and vicinity, on the accentual system of which he published a description in 1900), offered an explanation for the failure of initial stress in vocatives of accented o-stems which has major consequences for our understanding of accented stems in Slavic as a whole. He noted that many accented stems are in origin compounds of preverbs and deverbal nouns (or, alternatively, deverbal nouns to compound verbs) or of two simple nominal stems, e.g. zázor 'shame' < *za- + *zoru. If the base noun was postaccenting in PS1., the compound would be as well (assuming that the first member of the compound was unaccented). If the base noun was unaccented, however, it would have received default initial stress, e.g. *zor, voc. *zor-e, which could 37
The same retraction also occurs in the pi., at least in the conservative 19th-century norm of Budmani: sokóli 'falcons', voc. sokoli\junáci 'heroes', voc. júnāci, ¿éne 'women', voc. iéne (Leskien 1914:349). 38 The exceptional retraction of Cbv(j)ek 'person', ¿ov(j)eðe, d(ij)éte 'child', voc. d(¡j)éta. djévōjka 'girl', djévōjko need not be old; cf. ReSetar 1931a:22, 1931b:187, 194.
143
in turn influence compounds such as unaccented *za-zoru to become zázor, voc. zázore, i.e. underlying *zā-zor, *zā-zor-e. Originally unaccented compounds such as SC *nastavak 'addition, continuation' (to na-stati 'add, continue'), *Po-savac thus acquired underlying accent and joined the ranks of AP a, but apparently after the rule of initial stress in vocatives had ceased to be productive: hence nástavak, voc. nástāvð-e, Posavac, voc. PbsavC-e. So far as I can see, the above is the only plausible historical hypothesis for the failure of initial stress in vocatives to accented stems in SC. In addition to compounds, nominal derivatives must also have played an important rule in increasing the number of accented stems: even among nouns composed of a base and a single derivative, a large proportion will be accented on some syllable (cf. Dybo 1981:55-196, summary table 199; Dybo et al. 1990:54-61, 84-108, summary table 55-7, 59): base+ (* (* (*
derivative (* *( *
accented on 1st accented on 1st accented on 1 st
e.g. *(,stár-(íc-(a, acc. *(stár-(íc-Q *('brát-istv(-o *(Ták-ov-ú,fern. *(rák-ov-(a
*( *( *(
(* *( *
accented on 2nd accented on 2nd accented on 2nd
*tráv(-('íc-(a, acc. *tráv(-(TC-Q *vídov(-,ístv(-o *dvor(-'ov-ii, fern. *dvor(-ov-(a
* * *
(* *( *
accented on 2nd postaccenting unaccented
*dérv-('ín-(a, acc. *derv-(in-Q *bo2-Istv(-o *'snég-ov-ú, fern. *snēg-ov-(a
Since it is not unusual for derivatives to become morphologically and phonologically dissociated from their bases, through semantic specialization or loss of the base, such forms may easily have been reanalyzed as independent accented stems, e.g. zázor 'shame' above.39 Most of all, it was the unceasing, ever growing influx of loanwords from other languages, whether Greek, German, Turkish, Latin, French, or English, which contributed to and then massively bolstered the preponderance of underlyingly accented nominal stems, to the point where they now constitute some 92% of all nouns in Zaliznjak's 1967 Russian dictionary (Halle 1997:281). Such words must have been incorporated into the native grammar with columnar stem stress, which of course allows for maximal ease of 39
Many of these nouns, to be sure, would hardly ever be used in the vocative, but Posavd-e and the like speak for the reality of the facts presented here.
144
acquisition for child learners (not to speak of second-language learners!). In contemporary Russian, as in Ukrainian, Bulgarian, and Serbo-Croatian, the vast majority of nouns have fixed stress on one of the stem syllables throughout the paradigm. 4.4. From b and c to a and b: the prehistory of Baltic accentuation 4.4.1. Proto-Baltic or Pre-Lithuanian? As mentioned above, it is common in studies of Balto-Slavic accent to cite Lithuanian data as representative of Baltic in general. The reason for this, of course, is obvious: of the two modern (East) Baltic languages, only Lithuanian has preserved mobility of surface stress and exhibits a full contrast between underlying acute and circumflex intonation in long vowels and diphthongs, which is neutralized in unstressed syllables in the standard language and many AukStaitian dialects but survives in the Zemaitian dialects. In Latvian, which has restricted surface stress to the initial syllable, the contrast of falling, circumflex, and broken tones on long vowels and diphthongs by and large confirms the evidence of Lithuanian: in many cases, where an originally barytone (i.e. accented) noun or adjective has become mobile (i.e. unaccented) in modern standard Lithuanian, the original accentual paradigm, which survives in Old Lithuanian documents (particularly that of Dauksa) and/or in contemporary dialects (especially the Zemaitian dialects) is confirmed by the Latvian evidence. As a result, we now know that, say, the predominance of mobile accentuation (classes 3 and 4) in the adjective is a relatively recent innovation in Lithuanian (cf. Illið-Svityð 1979:15-8). Nevertheless, scholars of Balto-Slavic accent have by and large reconstructed Proto-Slavic forms and paradigms and compared them with those of Lithuanian, so that "Baltic" in practice becomes an extrapolation of pre-Lithuanian. The primacy of Lithuanian for our knowledge of Baltic accentuation has thus often led to a tendency to cite pre-Lithuanian forms in lieu of Proto-Baltic reconstructions. Although such an approach is understandable, one runs the serious risk of projecting Lithuanian-specific innovations backward in time to Proto-Baltic without sufficient consideration of other Baltic data, in particular that of Old Prussian. To be sure, Latvian does provide a check on e.g. the distribution of accentual paradigms in Lithuanian, as just noted; yet it is well known that Lithuanian and Latvian, as East Baltic languages, share a number of isoglosses and innovations which together sharply distinguish them from Old Prussian, the sole West Baltic language for which more than onomastic evidence survives. In §4.4.2,1 consider the evidence of the Third Catechism (Enchiridion), the sole accented 145
Old Prussian text, which suggests that the oxytone paradigm of PIE was in fact continued as such in Proto-Baltic. The elimination of columnar postaccentuation in East Baltic, and evolution of the Lithuanian(-Latvian) system of accentual classes, will be explored in §4.4.3. 4.4.2. The evidence of Old Prussian In his comparative grammar of Baltic, Stang (1966:172-3, 300) observes that the Old Prussian word for 'god', which occurs over 100 times in the Enchiridion, is never printed with a macron over either the e or / of the stem diphthong ei: nom. deiws (34x), deiwas (lx), gen. deiwas (48x), ace. deiwan (30x), voc. deiwa (2x), deiwe (lx), ace. pi. deiwans (1 x). Although the far from consistent accentual markings of that text render any argument from silence less than probative, one would expect at least some of these to be written dēiw- or deiw- if the stress were on the stem syllable (with acute or circumflex intonation, respectively). The consistent absence of such spellings throughout over 100 attestions, then, strongly suggests that deiw- had columnar oxytonic stress throughout the paradigm, hence gen. *deiwás, dat. *deiwán, etc.40 Similarly, the noun widdewu 'woman' also occurs in the dat. pi. as widdewumans. The agreement of both forms in stressing the thematic vowel -u- (< *-ā- after *w) calls to mind the PS1. cognate *vidov-a, which belongs to AP b and so likewise exhibits columnar postaccentuation, cf. §4.3.1. This empirical observation is crucial not only for establishing the synchronic accentual system of Old Prussian — a formidable task, given the paucity of evidence — but also for distinguishing archaisms from innovations in Lithuanian accentuation. It has repeatedly been claimed that the lack of a columnar oxytone paradigm in Lithuanian reflects the PBS1. situation, i.e. that PIE oxytonic o- and eh2-stems was eliminated in pre-PBSl.; the postaccenting class (AP b) of Slavic owes its existence to a rightward shift from a nonacute stem-final syllable, i.e. Dybo's Law. If, however, Old Prussian agrees with Slavic in possessing nouns with columnar thematic stress, and in particular two nouns whose oxytonesis either goes back to PIE (*deiw-as, SCr. gen. div-a < PIE *deyw-ós) or to a 40
One might expect nom. deiws < *deiwas, with syncopated o-stem nom. sg. ending -s < *-as (-as preserved in deiwas (lx) in the Enchiridion and in deywis in the Elbing Vocabulary of 1400), to have been stressed on the stem vowel as a result of a retraction not dissimilar to that which produced the "neoacute" in (post-)Proto-Slavic, e.g. *kōrljī < *korlji. The lack of a single spelling "deiws" (with circumflex ēi, cf. Lith. diēvas, Latv. dievs) out of 34 occurrences, however, points to unstressed ei. Perhaps the nom. has become unaccented as a result of the loss of syllabic a, just as the lsg. of simple thematic presents became unaccented upon apocope of *-ōmi > *-ōm (§6.3)?
146
post-PIE remodeling of a PIE ablauting paradigm (*widew-á:, PS1. *vídov-á < *widhewéh2 <— *widhéw-h2 ~ *widhw-éh2-), then it must be Lithuanian which has innovated in this regard. The limited data of the Enchiridion precludes more far-reaching conclusions on the accentuation evolution of Baltic. Stang (1957:60-1), noting the alternation of ace. sg. mērgan 'girl' vs. dat. pi. mergvmans (cf. also nom. sg. fern, antra 'second' vs. ace. āntran, spigsnā 'bath' vs. spīgsnan, probably not due to de Saussure's Law; Stang 1966:173-4) with the consistent absence of marking in ace. sg. deinan (i.e. *deinan; cf. Lith. diená, AP 4), tentatively concludes that Old Prussian had a mobile paradigm in addition to a columnar postaccenting class of d-stems; if so, the language may have possessed a three-way contrast of APs similar to that of Slavic. 4.4.3. How did postaccenting nouns become unaccented? If the PIE postaccenting type of Ved. devá-, Gr. misthós, and Russ. stol, stol-á did indeed survive unaltered into Proto-Baltic and down into Old Prussian, how was it eliminated in Lithuanian — or, given that Latvian likewise reflects only a two-way contrast of accentual classes, East Baltic? Let us first consider how PIE nominal stems would have evolved in Proto-Balto-Slavic under the assumption of Hirt's Law, beginning with ostems. As argued in §2.2.2 and reiterated above, o-stem nouns in PIE were underlyingly either unaccented or postaccenting. Figuring in the contrast of acute and circumflex intonation, we see that the stem vowels of pre-PBSl. o-stems would have been distributed among the following four possibilities: 1. postaccenting acute *'( 2. postaccenting circumflex 3. unaccented acute 4. unaccented circumflex
*A( *' *A
By Hirt's Law, the first of these would have become (*', i.e. postaccenting o-stems with acute stem vowels or diphthongs retracted the stress to the stem, and the stem syllable became accented. 1. accented acute 2. postaccenting circumflex 3. unaccented acute 4. unaccented circumflex
(*' *A( *' *A
147
It follows that class number 1 should give Lithuanian nouns of AP 1, and that classes 3 and 4 are continued by Lithuanian nouns of AP 3 and 4, respectively. But what of class number 2, postaccenting nouns which did not retract their stress by Hirt's Law? And what is the origin of AP 2 in Lithuanian? Without pretending to be able to answer these questions fully at present, I wish to note the following points related to the o-stems, the focus of discussion above. Once a number of o-stem case endings had become underlyingly accented under the influence of the /- and «-stems (and a:-stems, cf. §5.1.2), oblique du. and pi. forms with stressed ending could have been interpreted as belonging to either AP b or c, i.e. to postaccenting or unaccented stems. By themselves, these case forms would not be expected to have led to a full-scale reanalysis of postaccenting o-stems as unaccented. Stang (1966:171), however, has suggested that the remaining forms may have undergone a retraction from word-final syllables, producing the columnar stem stress of e.g. Lith. Dievas 'God', -o, -ui, -q, -u, -e; cf. also the contrast between indefinite and definite adjectives, e.g. báltas 'white' (AP 3), gēras 'good', def. baltas-is, geras-is, with retraction in the former but not the latter. Although the precise conditioning of this retraction is not clear, some such shift must have played a role in the transfer of PIE postaccenting (circumflex) o-stems to the unaccented class, i.e. to Lithuanian AP 4. In conclusion, it appears that the oxytonic class of PIE o- and ā-stems has been preserved in Proto-Balto-Slavic, and survives as such in Slavic (with especially clear reflexes in Russian and Serbo-Croatian) and probably Old Prussian (deiw-, widdewv). Its elimination in Lithuanian and Latvian is due to a number of retractions from the ending in (pre-)Proto-East-Baltic, which led speakers to reanalyze these postaccenting stems as underlyingly unaccented with mobile stress. Nominal stems with underlying accent on the stem-final syllable are historically postaccenting (oxytone) and have undergone "reverse Dybo's Law", i.e. Hirt's Law. Certainly, not all of these retractions have been definitively established, and there is no denying the necessity of further research in the prehistory of BS1. accentuation, particularly in the fate of nominal paradigms. The validity of Hirt's Law, however, is confirmed by its ability to explain not only an isolated set of columnar stem-stressed nouns with oxytone cognates (the traditional basis for Hirt's Law), but also to derive the distribution of BS1. nominal and verbal APs from securely reconstructible facts of PIE accent: in particular, the accentuation of PS1. *-ye/0- presents and of tthe infinitive in *-ti receives a simple and straightforward analysis. I therefore propose that Dybo's Law, one 148
of the cornerstones of the last few decades of Balto-Slavic accentology, should be set aside in favor of a pre-PBSl. retraction of stress to a preceding acute vowel or diphthong.
Chapter 5. Making historical sense of synchronic complexity: The Proto-Indo-European origins of Balto-Slavic nominal accentuation 5.1. Reconstructing the accentuation of Proto-Slavic nominal inflection The following chapter treats the reconstruction of Proto-Slavic and Proto-Baltic accentual alternations in the noun, with a view to reconstructing the accentual properties of case-number endings for each nominal class and determining their relationship to the PIE alternations described in §2.2.1. It will be shown that such salient features of the PIE nominal system as the columnar stress of o-stem nouns or the fundamental divide between "strong" (i.e. nom. and ace.) and "weak" (oblique, e.g. gen., dat., instr.) cases have been continued far more faithfully in the prehistory of Balto-Slavic, and in particular of Slavic, than has hitherto been recognized. As any second-language learner of a Slavic language with lexical stress will attest, the noun in, say, Russian or Serbo-Croatian exhibits a seemingly bewiildering number of intricate paradigmatic accent shifts, with many apparent idiosyncrasies and exceptions. Yet this picture of surface complexity masks some very strong patterns of regularity, which follow from the accentual properties of case-number inflectional endings. Taking Russian as an example, one finds that •
•
Russian masculine and neuter o-stem nouns never alternate stress in the singular: the only exceptions are the several dozen so-called "second locatives" in -ú, found in certain fixed expression with v and na 'in, on', e.g. v sadú 'in the garden', na beregú 'on the shore'; there are almost no Russian nouns which alternate stress among the four oblique cases of the plural: the sole exceptions are déti 'children', with gen. detéj, instr. det'mi vs. dat. détjam, loc. détjax; ljúdi 'people', with gen. ljudéj, instr. ljud'mívs. dat. ljúdjam, loc. ljúdjax (but older ljudjáx); and a number of originally oxytone feminine a-stem nouns which have retracted their stress one syllable in the plural, e.g. sesīrá 'sister', pi. nom. séstry, dat. sestram, instr. sestrami, loc. sestrax vs. gen./acc. sester, on which cf. below.
Not surprisingly, Serbo-Croatian exhibits similar regularities in its system of nominal accentuation: •
not including the neoacute, stress alternations in the singular masculine and neuter nouns are limited to two kinds: locatives in /-u/ of unaccented inanimate masculine
150
•
nouns, which contrast with datives in unaccented /-u/, e.g. u grádu /grād-u/ 'in the city' vs. dat. grádu /grād-u/; and vocatives of postaccenting stems, e.g. inostránac 'foreigner', gen. inostránca, voc. ¡nostrdnce (Leskien 1914:348-50, ReSetar 1931a:202). in most nouns, the accentuation of the dat./instr./loc. pi. in -ma (historically the ending of the dat./instr. dual) agrees with that of the gen. pi. Exceptions, which are for the most part confined to masculine o-stems, are due to later accent shifts in the gen. pi., mostly conditioned by the interaction of accent and vowel length (for details cf. Leskien 1914:355-60, ReSetar 193 la:29-35,38-40, 1931b:190-l, 195-6).
These patterns, of course, are likely to be inherited from Proto-SIavic, the reconstruction of which will preoccupy us for the remainder of this section. As pointed out in §4.2.2, it is only underlying unaccented stems, i.e. those with mobiile surface stress, which allow the learner or linguist to determine which nominal endings are underlying accented or unaccented. The accentual specifications of the various case-number endings for Russian and Serbo-Croatian are listed by Halle (1997:281, 288, tables 1 and 3); these are reprinted below for reference. Note that 101 and /E/ denote the Russian "jer" vowels, which surface as o and e, respectively, when followed by another 101 or ¡El in the following syllable, otherwise as 0; similarly, Serbo-Croatian fa/ denotes so-called "movable a" and is realized as a before a syllable containing another fa/, otherwise as 0. Russian1 nom. sg. gen. dat. ace. instr. loc. nom. pi. gen. dat. ace. instr. loc.
masc. neut. o-stems -0 -0 -a -u * -o -omO
-(é-
-u, -(u -(ojE, -(oju -(é-
-£§
-y. -(y -(ovO -(ejE
d-stems -(a -(y
-(a -(0
-y> -(y -(0
/-stems -E -i -i -E -Eju -i -i
-(ejEi
-(amO *
*
*
-(ami -(axO
neut. n-stems -ja -en-i -en-i -ja -en-emO± -en-i -en-(a -en-(0 -en-(amO
*
-E(mi**
*
-en-(ami -en-(axO
T h e ace. sg. of masc. o-stem nouns and the ace. pi. of all nouns is identical to the nom. or gen. if the noun denotes an inanimate or animate object, respectively. ^Historically -enEmO, -(EjE.
151
§Several dozen nouns contrast an innovative "second locative" with v, na, ending in dominant accented -(u, hence always stressed -ú. **Found only in 5 relic forms: det'mí, doðer'mí, dver'mí, lošad'mí, ljud'mí.
Serbo-Croatian
masc. neut. o-stems
nom. sg.* gen. dat. acc. instr. loc.
-9
nom. pl. gen. DIL acc.
-i -(a -(s-a: -(a-a: -(ima -e -(a
-0
-a -u *
-o -oma -(u, -u
cí-stems -(a -(e: -i,-(i -u, -(u -(o:m -(i
i'-stems -a -i -i
cons.-stems -e,-i
-3
-e
-e, -(e -(a-a: -(ama -e,-(e
-i -(i: -(ima -i
-sju -¡
*A number of masculine nouns, mostly proper names, take the nom. sg. in -o or -e, e.g. Marko, Tito, Dimitrije, Pavle, hypocoristic ho, Ditto, Savo, Boro (the latter are more common e.g. in Bosnia, whereas Sava, Bora prevail in Montenegro and Serbia). These inflect like masc. o-stems or (for hypocoristics) a-stems: hence Zoran voli Tita 'Zoran loves Tito', pjesma Dine Medina 'Dino Merlin's song'. *The acc. sg. of masc. o-stem nouns is identical to the nom. or gen. if the noun denotes an inanimate or animate object, respectively.
The nominal accentuation of Ukrainian and Belarusian in general confirms the pattern of the Russian standard language and dialects, with, as expected, certain differences in archaisms and innovations. Similarly, the various dialects of South Slavic agree for the most part with the pattern of the standard štokavian Serbo-Croatian noun, although certain areas, in particular Slovenian and its dialects and the ðakavian and Posavian dialects of Serbo-Croatian, preserve interesting accentual archaisms and pecularities.1 We have seen in §1.2.4 and §4.2.3 that PS1. nouns could belong to one of three accentual paradigms (APs), namely accented, postaccenting, and unaccented, commonly 1
The Torlak dialects of southeastern Serbia and neighboring northern Macedonia and western Bulgaria, whose accentuation has been investigated by Alexander (1975), preserve a number of important archaisms of PS1. date, but most of these concern the verb — not surprisingly, since these dialects have reduced their nominal inflection to two (or, in some transitional dialects of Kosovo bordering the Kosovo-Resava dialect area, three) cases. Note that the "sAtokavski shift" of the East Hercegovinian dialect underlying the standard language(s) does not affect the underlying placement of stress, so that sAtokavian dialects which have not undergone this shift, e.g. Torlak or Montenegrin, or have undergone it only in part, do not necessarily provide additional information for a historicalcomparative study of accent — although they are of course of great interest for South Slavic dialectology and the phonetic and phonological study of stress retraction and/or tone spreading (cf. Alexander 1993).
152
denoted a, b, and c. This three-way contrast is continued in all Slavic languages which have preserved lexical accent, although in some languages and dialects, especially Slovincian, (eastern) Bulgarian, and to some extent Slovene, the original distribution of lexical items has been significantly altered as a result of phonetic and or analogical shifts: in particular, stress retractions have led to the virtual elimination of the postaccenting class. In addition to the three main APs, the modern Slavic languages exhibit a number of variants which have arisen individually in their separate histories. Most important of these is the shift of postaccenting stems from columnar oxytone stress in the pi. to a pattern in which the nom.(-acc) pi. has stress retracted to the preceding stem-final syllable (or to the preceding mora in Cakavian SC, yielding rising intonation on a stem-final long vowel): the result is a "neo-mobile" paradigm. In many cases, the nom.(-acc) stress then spreads to the oblique cases, yielding columnar retracted stress in the pi. vs. oxytone stress in the sg. Russian and Serbo-Croatian, along with other Slavic languages with mobile stress, have generalized this pattern to virtually all postaccenting neuter nouns (where it finds parallels in other IE languages and may be inherited, cf. §2.2.3; Garde 1976a:237-8, 271-2). In ci stern feminines, SC has introduced the "neo-mobile" paradigm, whereas Russian has gone one step further and generalized the retracted stress of the nom./acc. Unlike Russian, however, SC has also introduced retracted stress to a number of masculine o-stems. neuter: Russ. seló 'village', pi. sēla, sel, -am, -ami, -ax polotnó 'linen', pi. polótna, polóten, polótnam, -ami, -ax SC sélo, séla (Cak. selo, séla) vlákno 'fiber', vlákna (Cak. vlākno, vlákna) feminine: SC tráva 'grass', ace. trávu but pi. nom./acc. trove vs. dat./instr./loc. trávama Russ. tend 'woman', ace. íenú but pi. ieny, ten, -am, -ami, -ax dolgotá 'length', ace. dolgotú but dolgóty, dolgót, -am, -ami, -ax2
-It is commonly claimed that gen. pi. forms such as sestér, zemél' (vs. sēstr-y, -am, -ami, -ax, zémli, zémlj-am, -ami, -ax) to sestrá 'sister', zemljá 'land' are relics of an earlier stress pattern according to which nom. sestry, zémli contrasted with oblique sestēr, sestr-ám, -ámi, -áx, zemél', zemlj-ám, -ámi, -áx. In fact, this assumption is unnecessary: if the underlying forms in modern Russian are /sestOr-/, /zemEl'-/, with "secondary jers" as required by the gen. pi., the stress is underlyingly on the jer in pi., with retraction in all forms in which the jer is not vocalized, i.e. all nongen, forms: hence sg. sestOr(-, pi. sest(Or-y, -amO, -ami, -axO > sestr-y, -am, -ami, -ax vs. sest(Or-0 > sester. Similarly for neuter gen. pi. forms such as okón: sg. okn(- in okn-ó, pi. ok(On- in nomVacc. ok(On-a > ókna but gen. ok(On-0 > okón. See R. Kim, in progress a.
153
masculine: SC kbnac 'end', gen. kónca but pi. nom. kdnci, ace. konce The gradual spread of retracted stress in the pi. of non-neuter nouns of AP b may be traced in the historical record of East Slavic as well as Serbo-Croatian: forms such as Russ. pi. nom. vdov-y, dat. vdov-ám, etc. survive into the 18th or even 19th c. (Illið-Svityð 1979:83, Garde 1976a:238, 272), and the older ending-stressed obi. pi. forms are preserved to the present in idioms such as govorít' po dušám 'speak from the heart', idík svin'jám 'go to the pigs', vsem sestrám po ser'gám 'to all the sisters by the earrings', for modern dúšam, svín 'jam, séstram. Within East Slavic, retracted stress in a-stem plurals has spread to the majority of Russian and Ukrainian dialects, so that columnar oxytone stress in the pi. is preserved only in the peripheral dialects of Transcarpathia (i.e. Rusyn) and northern Russia, e.g. Tot'ma iony, s'os'try (Stang 1957:59; IlliC-Svityd 1979:83-4).3 Other variant APs are obviously offshoots of the mobile AP c, e.g. a-stems with generalization of ending-stress to the dat. or dat. and ace. sg. (cf. §5.1.2). On the innovative Russian o-stem masc. nom. pi. in stressed -á, cf. §5.2. In seeking to reconstruct the nominal accentuation of Proto-Slavic, one must focus above all on three points of interest: 1) the oblique cases of the plural, which have considerably influenced and/or merged with each other in most Slavic languages; 2) the accentual alternations of the a-stem nouns, whose prehistory has never, as far as I am aware, been satisfactory addressed; and 3) the original accentuation of the consonant-stems, which have mostly been remodeled after the o-, a-, or /'-stems in the modern languages and only scattered traces of whose original endings survive in medieval documents and the modern languages and dialects. Let us consider each of these in turn. 5.1.1. Oblique plural and dual endings, and the accentuation of o-stems in ProtoSlavic In most Slavic languages, the dat., instr., and loc. pi. have generalized an ending originally proper to a single stem class: cf. Russian -am, -ami, -ax < post-PSl. a-stem *armi, *-ami, *-axu (<— *-asu), Polish instr. -ami, loc. -ach, but dat. -om < PS1. o-stem *omii. The accentual alternations of the modern languages, and in particular the accentual 3
This innovation has led to a merger of APs b and c in Belarusian and southern Russian dialects: both now share ending-stress in the sg. and stem-stress in the pi., e.g. Bel. vadá 'water', ace. -ii, pi. nom./acc. vódy (orig. AP c, Russ. vodá, vódu [ná vodu], vódy) like sjastrá, acc. -ú, pl. nom. séstry (orig. AP b, Russ. sesírá, -ú, sestry —> séstry). Cf. IlliC-SvityC 1979:83-4.
154
"curve" of the mobile paradigm of unaccented nominal stems, thus reflect the underlying accentual properties of these generalized endings. In order to reconstruct the accentuation of the various endings which have been lost, one must turn to data of three sorts: 1) relic forms; 2) internal reconstruction; and 3) archaic dialects which have preserved the oblique plural endings to a greater extent, especially Slovene and the kajkavian and Cakavian dialects of (Serbo-)Croatian. Crucial for the accentual reconstruction of PS1. nominal inflection are endings consisting of more than one syllable, i.e. endings in which the PIE stem vowel and the case-number ending proper have not phonetically fused into a single unanalyzable stemspecific ending. In contemporary Russian, disyllabic nominal endings are always accented on their first syllable: a-stem instr. sg. -(oju, alongside -(ojE, or pi. dat. -(amO, instr. -(ami, loc. -(axO (cf. Stankiewicz 1993:183, 185). Yet a number of relic forms in Slavic, and the evidence of modern dialects (especially Cakavian) and medieval accented texts, indicate that this is a comparatively recent development, and that at least certain disyllabic endings in PS1. and PBS1. were accented on the absolute final syllable. In Russian, /-stem feminine nouns ending in a palatalized consonant (underlyingly -E) normally form their instr. pi.in -ami, e.g. noð-ámi, povestj-ámi, svjazj-ámi to noð' 'night', póvest'
'story, tale', svjaz' 'tie, bond'. A small set of exceptions, however,
preserve the original PS1. /-stem ending -Emi < *-imi, namely det'mí, doðer'mí, dver'mí, lošad'mí, and ljud'míto déti 'children', doð' (gen. doðer-i) 'daughter', dver' 'door', lóšad' 'horse', ljúdi 'people' (cf. table in §5.1), to which one may add kost'mi in the idiom led' kost'mi 'fall in battle' (lit. 'lie down with (one's) bones') for modern kostjámi. These forms are the only reflexes of PS1. non-a-stem inflection in the dat., instr., or loc. pi. to survive in the language. Elsewhere in East Slavic, PS1. *-Imi survives in Ukrainian forms such as ljud'my, kist'my, dit'my/dit'my (to ljúdy 'people', kist', gen. kóst-y 'bone', dit-y 'children'); cf. also archaic and dialectal Serbo-Croatian ljúdma < *ljudmá (modern standard ljúd-ima, with regular /-stem dat./instr./loc. pi. -(ima), which can only continue PS1. du. dat./instr. *ljud-Ima (cf. konjma, kónjma < *konj'ima, Stankiewicz 1986a:332-3; for other old SC /- and w-stem forms in -mi, -ma, cf. Leskien 1914:424-5,428,443). The accentual alternation of these nouns, e.g. pi. nom. déti, ljúdi vs. gen.(/acc.) detéj, ljudéj, dat. detjám, ljudjám, loc. detjáx, ljudjáx, makes it clear that they are underlyingly unaccented. Thus the relic instr. pi. in -mi, underlyingly -Emi, must be
155
accented — and on the second syllable, for Accent Retraction and Jer Deletion (Halle 1997:283-4; cf. §4.2.3) would result in the wrong stress if it were underlyingly -(Emi:4
*
(*
*)
det
E
mi
* (* (*
—>
*)
"dét' mi"
Similarly in the other modern East Slavic languages, one must likewise posit underlying -E(mi: thus e.g. Ukr. ljud'my, kist'my, underlyingly ljud-E(my, kost-E(my. The survival of absolute word-final accent in the PS1. i-stem instr. pi. in East Slavic naturally leads one to consider whether other PS1. disyllabic oblique pi. endings were also accented on their second syllable. Under this hypothesis, the phonological development of /- and w-stems would have been as follows: gen. dat. instr. loc.
i-stem *-íj-(í *-i-(mú *-í-(mi *-i-(xú
w-stem *-ov-(ú *-ú-(mú *-u-(mi *-ú-(xú
As a result of retraction of stress from a weak word-final jer, stress in the gen., dat., and loc. would have come to fall on the first syllable of the ending in unaccented stems, e.g. post-PSl. *noð-íjI, *-imú, *-ixu, *syn-óvú, *-umú\ *-úxú. It is not difficult to imagine that these endings could have been reanalyzed by speakers and/or child learners as underlyingly *-(íj-i, *-(í-mu, *-(í-xú and *-(ov-ú, *-(ú-rmá, *-(ú-xú, with accented stem vowel (*í or *ú) followed by the ending proper. On their model, the instr. could have been remodeled to *-(i-mi and *-(u-mi, but the final stress of Russ. ljud'mi, Ukr. ljud'my indicates that the accentuation of PS1. *-i(mi has been preserved in East Slavic; similarly, archaic and dialectal SC ljúdma and Slovincian /-stem instr. pis. such as vosmi to vitos 'axle' attest to the survival of disyllabic endings with underlying accent on the second syllable (or ending proper) in South and West Slavic as well. The evidence of relic forms with reflexes of PS1. i-stem instr. pi. *-imi therefore makes it likely that PS1. disyllabic endings such as those of the i- and a-stem oblique plural were composed of stem vowel *-í-, *-ú- (or ablaut variants thereof, e.g. *-ov- in the u4
Exactly such a retraction is obsersved in ORuss. (Cudov NT) instr. pi. pútmi < *put(-ī(mi to postaccenting put' 'road, way' (cf. ace. pi. puti; Stang 1957:85).
156
stem gen. pi.) and accented ending proper. We shall see that this pattern is inherited from Proto-Balto-Slavic, and indirectly continues the accentuation of oblique case endings reconstructible for PIE root nouns and other consonant stems (cf. §2.2.1). What of the accentuation of the o- and a-stems? The standard treatments of Slavic accentology all agree in reconstructing underlying accent for the oblique plural of these stem classes, but disagree as to whether these endings were accented on the final vowel, i.e. on the ending proper as in /- and «-stems, or on the stem vowel. Garde (1976a:27-8, 30-1) represents the former view, whereas Stang (1957:62-4, 72-7, 84-5) reconstructs endingstress in the dat. and loc. pi. of o-stems of AP c but stressed stem vowel in a-stems of AP c, i.e. *-omu, *-éx'ú vs. *-ámú, *-'ámi, *-'ásn (similarly du. dat./instr. *-éma vs. *ama).5 Dybo (1981:26-39) and Dybo et al. (1990:47-54), who follow Stang on this question, correctly argue that pre-PBSl. *-á:-(mas, *-á:-(mi:s, *á:-(su would have become PS1. *-(á-nrm, *-(á-mi, *-(á-sú by Hirt's Law. But since Hirt's Law has also taken place in Baltic (§4.3), and Lithuanian clearly exhibits ending-stress in tf-stem instr. pi. -omis (§5.2), I consider it more probable that the sound-change outcome of Hirt's Law was undone on the model of the other stem classes; see immediately below for details. This hypothesis is supported by two sets of facts (Garde 1976a:30): 1. the Cakavian dat. -án, loc. -áh, e.g. in Orlec ofcán, ofcáh, goráh (Houtzagers 1985:85-7), can only continue PS1. *-am,ii, *-ax'u, with retraction from a wordfinal jer producing a long rising neo-acute;6 2. Slovincian a-stem instr. pis. in -ami, e.g. oufcamí, gluovamí (gluova, gen. glúovē), which have spread to other stem classes, e.g. bfegami to bfég 'shore, bank', kolamito kuolo 'circle, wheel';7
Loc. pi. *-éxú is from the PIE pronominal inflection, with *-é- < *-oy- (Skt. -esu, Av. -aēšú, Gr. Horn, -oisi <— *-oisu). On PS1. *-y < PIE *-ōis (Skt. -āis; Gr. Myc. -o /-ois/, Horn, -ois probably with shortening by OsthofFs Law, cf. §2.2.1.2, fn. 11 and §3.3.3, fn. 15), cf. Tedesco 1951:172, Jasanoff 1983:144-5. The PS1. a-stem loc. pi. was actually *-asu, preserved in Old Czech but replaced in OCS by -axu after -éxú, -vcú, -uxv. 6 In Orlec, stressed a, o, and e (not including /a/ from strong jer, which has merged phonetically with e; cf. a in the dialects of northern and southern Cres, a in Slovene) are lengthened in non-final syllables only; see fn. 13. The vowel length of loc. pi. -áh (-āh in Houtzagers's notation) therefore must continue a neo-acute, unless it has been influenced by both dat. -án < *-amú (with lengthening before syllable-final sonorant, see below) and instr. -ami (e.g. ofcámi, glavámi) < *- ami with lengthening in a non-final syllable. Belie (1909:226-8) gives gordn, gorih for NovT, which would seem to suggest that at least in this (generally archaic) dialect, the dat. and loc. pi. were *gor-amú, *gor-axu, whence gorán (with lengthening before syllable-final sonorant), gordh with short -a- < acute 7
Stang's attempt (1957:63) to explain this away as an innovation is not convincing. Is the Russian dial, form slezami cited by him (182fn.45, to slezá 'tear') perhaps also old?
157
Once word-final jers had been lost, of course, it would have been simple for dat. *-ām, loc. *-āh, with neo-acute *-ā-, to be reanalyzed as underlyingly accented stem vowel *-(a- plus desinence *-mu, *-xu, and consequently for the instr. pi. to be remade to *-(a-mi; this has occurred in East Slavic (Russ. glav-ámi, gor-ámi), standard SC (gláv-ama, gór-ama; on the quantity of the root vowel in SC see §4.2.2, fn. 21), and ðakavian (gorámi, dobrotámi to dobrotá 'kindness')- Thus all disyllabic oblique plural case endings, of o- and a-stems as well as /- and «-stems, were accented on their second syllable in PS1., i.e. on the ending proper. This conclusion stands in direct contradiction to one of the most established features of the system of nominal accent reconstructed for PIE principally on the basis of Sanskrit and Greek, namely that, whereas the /- and w-stems were marked by alternating ablaut and accent (with the proterokinetic type in *R(e)-i- ~ *R(0)-éy-, *R(e)-u-, R(0)-ewprevailing in the majority of daughter branches, including Balto-Slavic), the thematic o- and eh2-stems had fixed columnar stress on either stem or ending; cf. §2.2.2. Since the thematic vowel *-o- ~ *-e- was to all appearances unaccented in PIE, one would therefore expect unaccented o-stems in Proto-Slavic to be unaccented in all case/number forms throughout their paradigm. In contrast, the originally collective suffix *-eh2- was accented, so that unaccented PS1. ō-stems should have had columnar stress on the stem vowel. How then did unaccented o- and ō-stems acquire word-final accent (i.e. accent on the ending proper) in the dat., instr., and loc. pi.? Furthermore, how did unaccented a:-stems develop underlyingly unaccented ("enclinomenon") ace. and dat. sg. forms, as in the well-known alternation of Russ. nom. borodá vs. ace. bórodu (cf. §4.2.3), gorá vs. góru, ná goru, SC nom. rúka, gen. rúkē, instr. rúkōm vs. dat. rúci, acc. ruku, ná ríikul The origin of the unaccented ō-stem acc. and dat. sg. will be addressed in the next section. As for pi. dat. *-á:-(mú, instr. *-á:-(mi, loc. *-á:-(sú, du. dat./instr. *-á:-(ma in place of *-(á:-(mú, *-(á:-(mi, *-(á:-(sú, *-(á:-(ma with columnar accent on the stem vowel *-á:- < *-eh2-, one must conclude that these endings have been influenced by those of iand «-stems. In the latter, pi. dat. *-ī-(mú, *-ú-(mú, instr. *-ī-(mi, *-u-(mi, loc. *-i-(xu, *u-(xu, du. dat./instr. *-i-(ma, *-u-(ma, as well as pi. gen. *-Ij-(I, *-ov-(u, were easily analyzable as underlying combinations of unaccented stem vowel (or ablaut variant *-ovin the «-stem gen. pi.) and accented ending proper, and probably had been so analyzable since the pre-PBSl. remodeling of those endings on the basis of root and hysterokinetic consonant-stem nouns (cf. above). Thus pre-PBSl. pi. dat. *-(á:-(mas, instr. *-(á:-(mi:s or
158
*-(á:-(mins,8 loc. *-(á:-(su, with originally accented stem vowel *-(á:- and accented obi. pl./du. ending, were replaced by unaccented *-á:- and accented ending: the i- and «-stem pi. oblique case desinences were understood as "unaccented stem vowel + accented ending", and that pattern was generalized totf-stemsregardless of the historically accented nature of the stem vowel *-(á:- < PIE *-(eli2-. In contrast to the ending-stress of unaccented d-stems in the pi. (and du.) oblique, closer examination of the available evidence for masculine o-stems casts significant doubt on the reconstruction of an original alternation between unaccented nom.-acc. and endstressed oblique.9 To be sure, Old Russian and Middle Bulgarian, as well as contemporary Russian, do attest just such an alternation for unaccented masculine o-stems, entirely parallel to the pattern of i- and «-stems reconstructed above for Proto-Slavic. Cf. the following examples from Stang 1957:72-3, Neweklowsky 1986:166-8, Hinrichs 1985:34ff.: Middle Bulgarian 10 mírú, grad síi míra, í b(og)a bēsove, mg¿íe
nom. sggen. nom. pi.
Old/Middle Russian vólokú ná b[o]ga vary, vó vški, mu¿i ¿é
gen. dat. instr. loc.
vēkóvú, rodóvú vrágú, rcjdóvīa bésomú, bšsómú bēscómī z<$by górody, z bēsy grobšxú, v dómēxv}' vcoéxú, víbco3éxī
Russian kóren', nógot' kórnja, nógotja kórni, nógoti
kornéj, nogotéj kornjám, nogotjám kornjámi, nogotjámi kornjáx, nogotjáx
Many of these forms, however, have undergone morphological remodeling and so are of no value for determining the PS1. accentual paradigm. In Middle Bulgarian, for example, 8
The discrepancy in vowel length in the instr. pi. ending between PS1. *-mi and Lith. -nils may be reconciled if the latter continues pre-Lith. *-mi:s, with shortening by Leskien's Law. Mikkola (1950:34, 48; cf. Tedesco 1951:171) has proposed that the instr. sg. ending, PS1. *-mī (*-o-mī, *-I-mI, *-ú-mī) and Lith. -mi (-imi, -umi), actually continues *-min rather than *-mi (cf. Gr. -phin alongside -phi), and that the instr. pi. *-mis (cf. Skt. -bhis, Av. -bJS, Olr. /-v7) was remade to *-min-s, whence PS1. *-mi, pre-Lith. *-mi:s with the same acute intonation as in the ace. pi. endings (-us, -úos- < *-ons, -as, -qs- < *-ans, etc.). Attractive as this hypothesis is, it requires separating the instr. sg. forms in PS1. *-ml, Lith. -mi from d-stem PS1. *-OJQ <— *-q, Lith. -á, -cf- < *-ām < *-āmi (§5.1.2, fn. 17), as the latter probably cannot reflect *-ā-min. 9 On neuter o-stems, see the end of this section. 10 The acute, grave, and other accentual markings of MidBulg. manuscripts all appear to indicate stress; on their use and distribution, which has been influenced by Greek scribal practice, cf. Hinrichs 1985:19-21. 1 'May be for *doméxu, as the accent sign is dispreferred with the vowel jat' (probably owing to its shape, cf. Neweklowsky 1986:163, 166-7).
159
the nom. pi. is already largely formed with /-stem and especially «-stem endings, i.e. *-Ije (mnZie, mod. maíé) and *-ove (bēsove 'demons'; extended to virtually all monosyllabic masculine nouns in the modern language); as in many other Slaivc languages, the «-stem gen. pi. in *-ovu also occurs frequently. This is even truer of the modern Russian paradigm, in which all of the oblique plural endings were originally particular to other stem classes and only spread to the o-stems within the attested post-PSl. history of Russian: gen. -ov and -ej, underlyingly /-ovO/ and /-ejE/, continue PS1. u-stem *-ovu and /-stem *lji, respectively, whereas -am, -ami, and -ax are of course the old a-stem endings.12 The stress of MidBulg. gen. pi. rcodóvī or Russ. nogot-éj, -jám, -jámi, -jáx therefore tells us nothing about the accentual properties of PS1. unaccented o-stems in the oblique cases of the plural. Indeed, the distribution of stress in the plural of Russian unaccented nouns strongly suggests that ending-stress in the oblique was not original to masculine o-stems. Stankiewicz (1993:190-1) points out that the /-stems, for the most part a recessive declension (the only productive /-stems being derivatives in -(n)ost'), still make up a large proportion of nouns exhibiting the alternation between pi. nom./acc. and obi. Furthermore, most of the masc. o-stems with alternating stress in the pi. end in a soft (palatalized) consonant, and the majority of these have in turn been transferred from /-stem inflection, e.g. gost' 'guest', gus' 'goose', kóren' 'root' (/-stem <— PS1. n-stem *koreJ, zver' 'beast'. Taken together, these two facts strongly suggest that the alternation, i.e. endingstress in the pi. oblique, spread from /-stems to these nouns, probably by lexical diffusion. Data from other Slavic languages confirm these deductions. In Serbo-Croatian, the /-stems consistenly alternate between stem-stress in the (originally unaccented) pi. nom./acc. and ending-stress in the pi. obi., whereas only a few Stokavian dialects (e.g. Ragusan, the dialect of Dubrovnik) have a similarly widespread alternation in masculine ci sterns (Resetar 1931a:30-l); in plurals with -ov-, mostly to monosyllabic nouns, the alternation between e.g. nom. drug-ov-i, ace. drvg-ov-e and gen. drug-óv-ā, dat./instr./loc. drug-ov-ima to drug 'friend, comrade' is surely from the PS1. «-stem inflection, the source of the infix -ov- (Leskien 1914:350-1, 356-7, Resetar 1931a:40). Note especially that the alternation is completely absent from neuter o-stems. 1
similarly for Slovincian gen. pi. -on (< w-stem *-ovu, cf. Pol. -ów, /'-stem -ej in masc. gen. pi.), dat. -ami, loc. -āx (
160
Most important, however, is the direct evidence of those modern Slavic languages and dialects which best preserve the PS1. endings of o-stem nouns. In the ðakavian of Novi, masculine o-stem nouns which correspond to unaccented o-stems elsewhere in Slavic exhibit an almost entirely columnar paradigm, with default initial stress on all forms except (optionally) the gen. and loc. pi. in -īh; by comparison, the dialect of Orlec appears to have innovated somewhat, but stem-stressed oblique pi. forms to masc. o-stems of AP c are attested. Cf. the following forms from Belie (1909:209-11) and Houtzagers (1985:82-5):
nom. gen. dat. ace. instr. loc. voc.
Novi sgvlás 'hair' vlasa vlásu vlás vlásom vláse vlðse
pi. vlási vlásīh (ylāsíh) vlásōn vlási vlási vlásīh (ylāsíh) vlási
Orlec sgbrót 'ship' brōda]ļ
pl. brōdi brōdi
brōdon brðdi brōdu/brodv brōde
Traces of initial stress in masculine o-stems of AP c may also be found elsewhere, e.g. Slovene mqi 'man, husband', dat. pl. motem < *mQ2-emu, whose circumflex intonation continues an unaccented preform (vs. loc. motéh < *-éxií, with stressed ending; cf. Stang 1957:73-4), or MidBulg. forms such as instr. pl. z<$by 'with teeth', loc. víbcoļēxī 'in the gods' (see above). I therefore tentatively conclude that unaccented masculine o-stems in Proto-Slavic, before any remodeling on the basis of other stem classes took place, were originally unaccented throughout all case/number forms of the paradigm. Given that masculine oand ¿/-stems already were in the process of morphological merger in OCS, the «-stem endings must have played the primary role in introducing ending-stress in the oblique cases of the plural, and so creating an accentual alternation between nom.(/acc.) and oblique pl.: this is unquestionably the case in štokavian SC, where the -ov- infix found with all but a handful of monosyllabic masculine nouns is obviously of «-stem origin. Even without such drastic morphological remodeling, the o-stems could have acquired such an 13
In these northwest Cakavian dialects, e, o, and a (but not the reflex of a strong jer, which has merged with e in Orlec and with a in northern and southern Cres, but remains distinct a in Slovene) are lengthened in non-final syllables, e.g. Cres, gen. Cres-a, uskot 'doorstep', gen. uskōd-a, brát 'brother', gen. brā-ta (Houtzagers 1982:126, 1985:17-8). Cf. Slovene, where i and u are also lengthened in non-final syllables.
161
alternation under the combined influence of the u-, /'-, and a-stems, as in Old Russian and Middle Bulgarian.14 Modern Russian, of course, has entirely replaced the original o-stem endings with the reflexes of other classes, but the skewed distribution of nouns with alternating pi. stress reveals the secondary nature of this alternation in old PS1. o-stems (see above). In unaccented neuter o-stems, the situation was somewhat different. Like masculines, they must have been unaccented throughout the sg., and probably also in the oblique cases of the pi., given the identity of oblique case endings for masculine and neuter nouns in all stem classes in PS1. (a pattern inherited from PIE). By the time of the accented medieval Russian and Bulgarian documents, however, the accentuation of the nom.-acc. pi. in *-(á, which continues accented PIE collective *-(eh2 (cf. §2.2.3), has spread to the rest of the pi., resulting in columnar ending-stress. If this analogy is of PS1. date, it follows that unaccented neuter o-stem nouns would have contrasted a uniformly unaccented sg. with a uniformly postaccenting pi., in marked contrast to the behavior of masculine o-stems. 5.1.2. Unstressability and reanalysis in a-stems Let us return to the problem of accentual alternation in the sg. of unaccented dstems. As emphasized in the previous section, the combined evidence of Sanskrit and Greek favors the reconstruction of columnar stress on the stem or ending for PIE e a sterns, as for o-stems; neither language offers the slightest trace of accentual mobility. What to make, then, of the stress alternations within ō-stem paradigms in Baltic and Slavic, the famous contrast of Lith. ¿iemá, iiēmq and PS1. *zimá, *zímņ > Russ. zimá, zímu, SC zíma, zimu, ná zimu0. It is of course thinkable that BS1. has here preserved an archaism of PIE date which has been lost in the ancient classical languages; such a hypothesis, although relatively unlikely, should not be dismissed out of hand.15 I believe, however, that it is in fact 14
Cf. also mod. Russ. vólos 'hair', gen. pi. volós < *vōlsi5, or Cakavian dialect forms such as vlás 'id.', ztib 'tooth', gen. pi. vlás,zúb < *vðlsú, *ZQbú (Stang 1957:73 with refs.) The fluctuation between initial and ending-stress lasted well into the modern period in Russian: witness Pushkin's variant bésovú for bésóvu, gen. pi. of bésú 'demon' (Neweklowsky 1986:166 with refs.). 15 Note for instance that East Baltic is absolutely crucial to the correct reconstruction of the PIE o-stem abl. sg. ending: whereas Skt. -āt, Av. -St, OLat. -ad, and PS1. *-a suggest PIE *-ōd < **-oed, Lith. -o (underlyingly circumflex, since it does not effect de Saussure's Law), Latv. -a can only continue PBS1. *-a: < PIE *-e-h2ed — i.e., the endingless (pre-)PIE loc. in *-e followed by the postposition *hied (> Lat. ad 'to, at', OE cet). Latin has not surprisingly replaced the vocalism with *ō-, as has Greek, cf. fossilized oiko: 'from home'; Indo-Iranian and Slavic are indeterminate.
162
possible to derive the intraparadigmatic alternations of BS1. d-stems from a columnar pattern of the Sanskrit or Greek type, and that this innovation may be explained within the "brackets-and-edges" model with the help of an additional assumption with good parallels in other languages. In a classic article published in 1965, Stang argued on the basis of a comparison of attested forms in IE languages that the expected word-final sequences *-eh2m and *-eh2ns in the ace. sg. and pi. of e/12-stems, respectively, had already become *-ām and *-ās in PIE. Ace. sg. *-ām, of course, underlies Skt. -ām, Av. -q, Gr. -ānl-εn, Lat. -am, Osc. -am, PGmc. *-q (bimoraic, cf. Goth. -0, OE -ce, OHG -a); ace. pi. *-ds, continued by Skt. -ās, Av. -á < PIr. *-āh, PGmc. *-ōs (bimoraic, cf. OE -ce > -e vs. nom. pi. -e < PGmc. *ōs; the original distribution is recoverable from the statistical distribution of endings in Kentish and early West Saxon texts, cf. Stiles 1988:130-1; Goth, -ōs is indeterminate, cf. Wright 1910:38), has been remade to *-ē-ns by analogy to o-stem *-o-ns in the prehistory of Greek (PGr. *-ans > Cretan -ans V- ~ -as C-, Att. -ās, EAiol. -ais). Importantly, neither Vedic Sanskrit nor Gatha Avestan contain any forms in which these endings must be read disyllabically, i.e. as -aam <— *-aa < *-eh2iņ or -aas < *-eh2Jjs, with "laryngeal hiatus" between *e and syllabic nasal. It thus appears that the development of *-eh2m, *eh2ns > *-ām, *-ās had taken place already by PIE. To be sure, ace. sg. *-ām can easily have arisen by analogy to other non-neuter ace. sg. forms at any stage in post-PIE, or in the prehistories of the individual languages. The lack of *n in the d-stem ace. pis. of Sanskrit and Germanic, however, can hardly be dismissed by appeal to analogy: rather PIE *-ās has been remade with the ending *-ns of e>-stems and other stem classes in most IE languages. I propose that these cases of early non-laryngeal-conditioned *ā have been of utmost, and previous unsuspected, importance for the fate of the accentuation of PIE d-stems in Balto-Slavic. We have seen in § 1.2.3 that languages vary as to which segments, or phonemes, are projected onto line 0 of the metrical grid as potentially stress-bearing elements. The parameters of Line 0 Syllable Boundary Projection allow for a great number of possibilities, not only in which types of segments are regularly projected (vowels only, or vowels and syllabic sonorants, etc.), but also in specifying exceptions. Thus, the antepenultimate stress in Polish uniwersytet or republika, which stands in contrast to the normally penultimate stress of the language, follows immediately if the vowel of the penultimate syllable l\l (realized as y [b] or i [i] after nonpalatalized and palatalized consonants, respectively) is not projected; this assumption immediately explains why y (/) 163
is unstressed throughout the paradigms of these words: uniwersyfecie, republikami.
instr. pi. uniwersytet'ami;
uniwersytet,
loc. sg.
rep'ublika, gen. pi. republik,
instr. pi.
In such forms, the vowel in question is unstressable: because it does not
project an element on line 0, it can never receive surface stress by the rules of stress computation of the language. I propose that at some stage of pre-Proto-Balto-Slavic, the long vowel *ā became unstressable, i.e. either lost the ability to or was barred from Line 0 Projection.16 As a result, previously accented ace. sg. *-(ām and ace. pi. *-(ās were reinterpreted as unaccented endings suffixed to an unaccented stem: * (* *
(*
*\
*gálw ām
*
*
*ln
galw
ām 'onto the head'
*
*
(* *
*\
*\
(*
*
*
*feīm ās
*dá
Zeim
*\
ās 'until the winter'
The same unstressability seems to have affected the ō-stem dat. sg. ending, which suggests that loss of intervocalic laryngeal in PIE *-éh2-ey and contraction of vowels occurred early enough in pre-PBSl. to yield *-āi, with the vowel *ā of the ace. sg. and pi. In the gen. sg., however, PBS1. *-á:s is still clearly stressable and underlyingly accented (cf. Lith. galv-ōs, iiem-ðs, PS1. *gor-y > Russ. gory, —> SC gore), so that laryngeal loss and contraction in PIE *-éh.2-es must have been of later date: perhaps the PBS1. ending replaced an earlier *éh2-s, which is continued by PGmc. bimoraic *-ōs > OE gief-e [< -ce\, OHG geb-a), at some time following the contraction of dat. *-éh2-ey. As a result, pre-PBSl. came to contrast sg. nom. *-(á: < *-(eho, gen. *-(a:s < *(eh2-es, pi. nom. *-(á:s < *-(éh2-es with sg. dat. *-á:i < *-(éh2-ey, acc. *-á:n < *-(ām (< pre-PIE *-eh2-m), pi. acc. *-á:s < *-(ās (< pre-PIE *-eh2-ns). 17 These accentual I6
Such an innovation easily lends itself to an analysis under Optimality Theory, by means of a high-ranking, inviolable surface constraint against stressed *ā. 17 And instr. sg. *-an < *-ām < *-ā-(mi (cf. Lith. AP 3 gálva, pāmoka to galvá 'head', pamokā "lesson"; AP 4 iiemā < *2iēma to iiemá 'winter' by de Saussure's Law), which became unaccented as a result of the apocope of accented *-i after a sonorant. Exactly the same development has occurred in the pres. lsg. in Slavic, where PBS1. *-o: —> pre-PSl. *-<5:-mi —> *-ó:-(mi (with accented ending from the du. and pi.) became unaccented as a result of apocope of *-i. For details, cf. §§6.3. 6.5 — Slavic has replaced the PBS1. cT-stem nominal ending *-dm > *-q with pronominal *-oj-o. which is stressed on the first syllable: cf. Russ. golov-ój(u), Ukr. holov-óju to golov-á, holov-á 'head'.
164
specifications have been well preserved in Lithuanian, where they are responsible for the mobile stress of unaccented nouns of paradigms 3 and 4 (see further below): the one exception among the endings discussed so far, nom. pi. -a:s for expected -(a:s, may easily have arisen by analogy to the i- and «-stems, where unaccented nom. and ace. contrast with accented endings in the oblique cases of the pi. The virtually complete correspondence of the Lithuanian and PS1. d-stem endings is illustrated in the table below (cf. §5.2 on the reconstruction of the pre-Lithuanian endings): nom. sg. gen. dat. ace. instr. loc. nomVacc. du. gen./loc. dat./instr. nom. pi. gen. dat. ace. instr. loc.
pre-Lith. -(á: -(a:s -a:I -an -án
PS1. -(a -(y
-ē
-o. -(ojo. -(£
-ie
-é
-á:-(mV
-(u -á-(ma
-ā:s
-y
-(ō:n
-(Ú
-á:-(mus -ans -á:-(mins -á:-(su
-á-(mú -y -á-(mi -á-(xú
As a result of the unstressability of pre-PBSl. *ā, what was originally a columnar postaccenting paradigm with stress on the stem vowel *ā becomes a mobile paradigm. I therefore argue that inherited oxytonic d-stems were reanalyzed as unaccented stems, followed by accented (e.g. nom. *-(á:, gen. sg. *-(á:s) as well as unaccented endings (i.e. those containing *ā). At some point, the accentuation of the /- and «-stem plural oblique cases spread to the d-stems, so that original dat. *-(eh2-mas, instr. *-(eh2-mins, loc. *(eh2-su became PBS1. *-á:-(mas, *-á:-(mins, *-á:-(su (cf. §5.1.1). This innovation expanded the mobility of unaccented d-stems, which now possessed forms stressed on the post-stem syllable and on the ending proper (in the dat., instr., and loc. pi.), as well as unaccented forms or enclinomena. Nevertheless, we have seen that Slavic, and probably also Old Prussian, possess oxytone d"-stems (there being of course no oxytone paradigm in Lithuanian), e.g. X (cf.
165
§4.X). Although I am not able to determine at present the conditions governing the appearance of oxytone or mobile stress in a given d-stem noun (a problem complicated by the paucity of word equations with extra-BSl. cognates), the changes outlined above had the effect of creating a distinction between accented and unaccented endings within the āstem paradigm, thus introducing the potential for mobility into a class of nouns which had entirely columnar stress in PIE. 5.1.3. Proto-SIavic consonant-stems: a lack of evidence? Before summarizing the findings of the previous two sections, a few comments on the vestiges of consonant stems in Slavic are in order. In Proto-SIavic, as represented by OCS, the only surviving consonant-stem nouns are: • •
• • • •
two feminine r-stem nouns of relationship with nom./acc. sg. *-i, stem *-er-: *mati 'mother', gen. *mat-er-e <— PIE *méh2tēr (or *meh2ter, cf. §2.2.1.2), *dut'i (OCS dúšti), gen. *dut'-er-e <— *dhugh2tēr (cf. §2.2.1.4) several neuter j-stem nouns with nom./acc. sg. *-o < PIE *-os, stem *-es- < PEE *-es, e.g. *slov-o 'word', gen. *slov-es-e, pi. nom./acc. *slov-es-a < PIE *£lew-os, *-es-, *neb-o 'cloud', gen. *neb-es-e, pi. nom./acc. *neb-es-a 'heavens' < PIE *nébh-os-, *es- (cf. Hitt. nēpiš, Skt. šrávas, nábhas, Gr. kléos, néphos, gen. klé-o:s < *-eos < *-esos; Lat. genus, gen. gener-is 'kind, type, race', Goth, riqis, gen. riqiz-is 'darkness') several masculine n-stem nouns with nom. sg. *-y or *-%, stem *-en-, e.g. *kam-y 'stone', ace. *kam-en-i, gen. *kam-en-e, *kor-e. 'root', ace. *kor-en-i, gen. *kor-en-e; cf. also *dírií 'day', gen. *dln-e several neuter n-stem nouns with nom. sg. *-?, stem *-en-, e.g. *jlm-e, 'name', gen. *jim-en-e < post-PIE *hiņh3m-en-, *verme, 'time', gen. *verm-en-e diminutive/hypocoristic suffix in modern languages neuter nf-stems, productive in diminutives of animals, personal names, etc., e.g. Russ. pi. tel-ját-a 'calves', orl-ját-a 'eaglets' (sg. tel-ēnok, orl-enok, regular morphological correlation of sg. -'énok ~ pi. -játa; cf. Ukr. sg. teljá), Pol. ciel-ŗ 'calf, pi. ciel-qt-a, SC tán-e 'ball', pi. tan-ét-a, d(ij)ét-e 'child', gen. d{j)et-et-a,x% Bulg. ágn-e 'lamb', kebápCe 'piece of kebab', pi. ágn-et-a, kebáp-ðe-ta
The accentual reflexes of these in present-day Slavic languages and dialects have been the subject of a study by Stankiewicz (1986b), who concludes that PS1. contrasted endingstress in the nom. (or neut. nom./acc.) sg. with initial stress in the oblique cases. Such an alternation not only disagrees with those reconstructed for the other nominal classes in PS1., but is exactly the opposite of the pattern of consonant-stems in Sanskrit, Greek, and '^For animate «f-stems, a collective formation, usually in -ād, serves as pi.: d(ij)éte 'child', pi. d(j)éca < PS1. *défica (cf. Bulg. deté, decá); tele 'calf, BugarCe 'young Bulgarian', pi. tél-ād, Bhgare-ad (Leskien 1914:342-3).
166
Lithuanian (on the latter cf. §5.2). If this reconstruction is correct, one must explain how the unaccented "strong" case endings became accented in PS1., while the accented "weak" oblique endings became unaccented In fact, the alternation posited by Stankiewicz is directly attested in only a handful of forms, i.e. Russian (northern dialects) doC-i 'daughter' (ace. dóð-er', doí-ér', gen. dóðer-i), SC (dial.) kc! (for standard káī, coll. (k)cérka), vr(ij)ém-e 'time', gen. vrem-en-a, (dial.) plem-e 'tribe', gen. plem-en-a. As is well known, the vast majority of (medieval) and modern Slavic reflexes of the consonant-stems have undergone morphological remodeling and been assimilated to one of the major declensional classes, the traditional o-, a-, and /-stems. Following the principle enunciated in §4.1 and §4.3.1 that only word equations can serve as a sound basis for accentual reconstruction, it follows that forms such as Bulg. dašterjá 'daughter', dial. (Banat) dašteráor Cakavian (Novi) svekrvá 'wife's mother-in-law', ace. svékrvu, dat. svékrvi are of no value for determining the accentual paradigm of consonant-stems in Proto-Slavic. As for Russ. dial. doCi, the accentuation of hysterokinetic r-stem nom. sg. -i < PSI. *-i < PBS1. *-e: (cf. §§4.2.2, 5.1.4) may be connected with that of Lith. dukte: the traditional view, going back to de Saussure (1896:163), holds that PBS1. *dukte: directly continues PIE *dhugh2tē(r).19 The historical interpretation of neuter «-stems such as SC vr(ij)ém-e, plém-e, on the other hand, is hampered by uncertainty as to the origin of the nom.-acc. sg. ending. If PSI. *-% continues PIE hysterokinetic n-stem nom./acc. sg. *-ēn (Vaillant 1950:214, 1958:195; cf. Schindler 1975a:9), the stressed ending of PSI. *verm| would be exactly parallel to that of hysterokinetic r-stem *dut'i; initial-stressed oblique forms such as SC vremen-a, -u, -om or Russ. vrémen-i, -em result from the addition of unaccented o-stem (in SC) or /'-stem (in Russ.) endings to an unaccented stem, which has probably arisen in the same way as that of 'daughter'. As examples of hysterokinetic inflection in neuter nouns are otherwise rare in ancient IE languages, however, one might entertain the alternative derivation of PSI. *-% < *-én <— PIE obi. *-én- in proterokinetic n-stems:20 the nom./acc. would then have preserved the accented ending. Following the 19
Contra de Saussure (1896:161-5), who proposes PIE ace. sg. *dhugh2tériņ > PBS1. *dukterin with retraction from an internal stressed syllable, there is no evidence that the preform of Lith. diikterl, PSI. *dut'eri was not an enclinomenon with default initial stress: cf. SC zd kóēr 'for the daughter' (admittedly less than probative, as káī has been assimilated to the fern, /-stems except for the nom. sg.). 20 In favor of this hypothesis, note that all the neuter n-stems of Slavic are old PIE deverbal nomina actionis in *'-mn - *-mén-, e.g. PSI. *ver-me, 'time', *ber-me, 'burden, load' <— PIE *wert-mņ, *bher-mņ to *wert- 'turn', *bher- 'carry' (cf. §2.2.1.4). The one exception, aerostatic PIE 'name', has
167
breakup of PS1., the nomVacc. sg. of unaccented r- and n-stems for the most part took over the initial stress of the oblique forms, e.g. Russ. nom./acc. dóe', gen. dóíer-i (with regular /-stem alternation: nom. pi. dóðer-i, genVacc. doCer-ej), imja 'name', vrémja 'time', plémja 'tribe, race', gen. ímen-i, vrémen-i, plémen-i (pl. imen-á, vremen-á, plemen-á), SC ¡me, pleme, gen. lmen-a, plémen-a (pl. imen-a, plemen-a), leaving only isolated relics of the earlier alternation. In any case, the near-complete replacement of the distinctive PS1. consonant-stem endings, in particular gen., loc. sg. and non-neut. nom. pl. *-e (OCS -e), in the modern Slavic languages which have preserved lexical accent robs us of any information as to the accentual properties of those endings.21 One may reasonably suppose that gen. sg. *-e was accented, like OLith. -es (e.g. vanden-es, dukter-es; §5.1.2) and the PEE ending *-(es, and that gen./loc. du. *-u and gen. pl. *-i were accented as well, but the absence of direct evidence makes all such deductions necessarily speculative, however probable. 5.1.4. Synthesis On the strength of the deductions arrived at above, we may propose the following underlying accentuation of nominal inflections for Proto-Slavic.22
nom. sg. gen. dat. ace.
o-stems masc. neut. -u -o -a -u -u -o
ā-stems -(á -(y -é -
i-stems
«-stems
cons.-stems m./f. neut. -yl-\
-l
-ú -(u -ovi -u
-i
-%l-o
-(e -i -e/-o
been assimilated to proterokinetic inflection in Balto-Slavic, as in Anatolian, Celtic, and Greek; cf. §2.2.1.1. 2 'The only significant exception in East and South Slavic is Russ. nom. pl. -e to ethnonyms in -an-in, e.g. graid-an-in 'citizen', Anglid-an-in 'Englishman', pl. graīd-án-e, AngliC-an-e (gen./acc. pl. in - 0 < *-ú: gratd-án, AngliC-an). Gen. sg. *-e survives in West Slavic in the Czech «f-stems, e.g. jehn-ē 'lamb', gen. jehn-£t-e. The word for 'day', PS1. *dīnī, gen. *dīn-e, has also preserved consonant-stem variants in West Slavic, e.g. Cz. gen. dne, loc. ve dne 'in the day' (otherwise dni, dnu), Pol. we dnie 'id.' (otherwise dniu), pl. nom. dnie (or dni); on the array of attested forms in earlier SC, cf. Leskien 1914:444-7. —The accentual evidence for the gen., dat., and instr. sg. of /- and «-stems is conflicting, but it appears that the gen. and instr. were originally accented, the dat. unaccented: cf. Stang 1957:77-82. 86-90, Dybo 1981:28-30, Dybo et al. 1990:48-9. Following Tedesco (1951:175), I assume that the anomalous stress pattern of the numerals ' 5 ' through '10', e.g. Russ. désjat' '10', gen., loc. desjatf (whence also dat. desjati), preserves the accentuation of the PS1. /-stem gen. and loc. I am not entirely convinced by the evidence for initial stress in the ¿- and «-stem dat. pl. (Dybo 1981:29-30), and hope to take up this problem at a later time. As per §5.1.3, the accentuation of PS1. consonantstem endings is unknown; the values in the table below have been extrapolated from the pattern of /'and //-stems.
168
instr.
-o-rrií
loc.
-ē
-ē nom./acc. du. -a -(u gen./loc. -o-(ma dat./instr. nom. pl. gen. dat. acc. instr. loc.
-i -ú
-(á
-I-(mí, -I-0Q
-u-(ml
-(£
-(i
-(u
-ē
-i -īj-(u -í-(ma
-y -ov-(u -ú-(ma
-i
-lje, -i -lj-d -i-(mu -i -í-(mi -I-(xú
-ove -ov-(u -u-(mií -u -í-(mi -í-(xí
-e
-(á
-i
-í-(mú -(á -i-(mi -i-(xu
-(u -a-(ma -(y
-U
-o-mú -y -(á -y -é-xú
-i-(rrií
-(0JQ
-á-(m\í -y -á-(mi -á-(xú
-(e -ē -(u -í-(ma
The majority of vowels in these endings are short and so are not marked for intonational contrasts. For endings containing a long vowel such as /-stem gen. *-i or consonant-stem neut. nom./acc. pl. *-a, it is generally difficult to determine the original PS1. intonation, as all word-final syllables are shortened in the daughter languages which preserve phonemic length, e.g. Serbo-Croatian or Czech. In the dat., instr. and loc. pl. of d-stems, *-a:- is followed by another syllable (the ending proper); based on its modern reflexes, Cz. -ám, -ami, -ách, ðak. -án (< *-amu), -ami, -ah, this stem vowel carried acute intonation in PS1., which is consistent with its derivation from PIE *-eh2-. On the basis of these forms and Lithuanian -ā < *-á: (on which see §5.2), I give the PS1. nom. sg. as acute *-(á, as well as the nom.-acc. neut. pl. *-(á < PIE *-eh2. As for o-stem dat. sg. *-u:, n-stem masc. nom. sg. *-y1, and r-stem fern. nom. sg. *-i:, I follow Jasanoff (1983) in deriving these from *ú:i < *-ð:i < (pre-)PIE *-o-ey, *-u: < *-ð: < PIE *-ō, and *-e: <— PIE *-ēr, respectively, under the assumption of a sound change raising mid circumflex long vowels in word-final syllables.23
The nom. sg. ending *-i < PIE *-ih2 of PS1. //Jo-stems, which continue (late) PIE feminines in *-ih2- ~ *-yeh2- (§2.2.1.4), occurred in the nominal suffixes *-yn'i, *-lji, e.g. OCS bogyn'i 'goddess'. ladii 'boat' (acc. bogyn'q, ladijg), and in pres. act. ptcp. *-Qt'-i (*-e,t'-i), pret. act. ptcp. *-ú5-i (*-Iš-i), e.g. OCS nes-qSt-i 'carrying', nes-úš-i 'having carried' (cf. Lith. nēš-ant-i, nēš-us-i). Although this ending has been replaced by *-ja everywhere in modern Slavic, PS1. *-i, like its Lith. cognate -i < *i:, appears to have been accented, if Garde (1976a: 127-30) is correct in seeing Ukr. pres. act. ptcps. such as nesuCy, leiaiy 'lying down' as continuations of fern. sg. forms. 23The same change is likely responsible for o-stem gen. pl. *-ú < *-ú:n < *-ð:n < PIE *-o-Hom (cf. Ved. -aam, GAv. -aqn, frequently scanned as two syllables with laryngeal hiatus).
169
5.2. Lithuanian, Old Prussian, and Proto-Balto-Slavic nominal accentuation Having reconstructed the accentuation of PS1. nominal inflections, let us next determine to what extent this system may be projected backwards to the Proto-Balto-Slavic stage. The underlying intonation and accentuation of the inflectional endings of the noun in standard Lithuanian are given in the table below. nom. gen. dat. ace. instr. loc. voc.
(ī-stems o-stems -(é sg- -as -(ys* -(ó -ō -(ðs -(és -el -ul -āī -? -% -% -do -é -4 -é-(jé -(é§ -y-(jé -o-(jé -e -(y -e -a
n./a. du. dat. instr.
-UO
-(ám -(am
nom. pi. -(ai gen. -tt dat. -(á-ms -úos ace. -(aís instr. -uos-(é loc.
/-stems -(is -(iēs
«-stems -(us -(aus
n-, r-stems -(uð, -(é -(ens, -(efs
**
**
**
-i
-u -u-(mi -u-(jé -(aí¡
X
-i-(mi** -y-(jé -(iē
X X
χ**
-íe -(óm -(om
-íe -(ém -(ém
-y -(ím -(im
-ú -(úm -(um
-iúo, -íe X X
-ōs
-és
-ys
-ús
X -(y X X X X
-tt
-(ó-ms -(é-ms -és -*s -o-(mis -é-(mis -os-(é -és-(é
-m
-(i-ms
-fs -i-(mis -ys-(é
-(í
-(u-ms -ús -u-(mis -uos-(é
♦shortened to -is when unstressed, i.e. in nouns of classes 1 and 2 **Dsg. o-stem -/«/, -ui for masc, tT-stem -iai, -ai for fern.; also Isg. -iu, -iá beside -imi in /'- and consonant-stems §old o-stem loc. sg. ending -(iē survives only in namiē 'at home' (to nomas 'house, home') told consonant-stem gen. pi. -(tj survives in some (but not all) old consonant-stems, e.g. dantij to dantis 'tooth', otherwise -(hj with palatalization of preceding consonant X = endings identical to those of ¿-stems
Lithuanian, of course, has undergone a number of well-known innovations, e.g. M-stem nom. pi. -us (for *-avas < *-ewes by analogy to /-stem -ys < *-eyes) and loc. pi. -uosé for *-ōsé (by analogy to the o-stems: pi. ace. -us : loc. -uosé :: -us : X, X = -uosé),24 or the locative, which with the exception of o-stem -(e (and the isolated namiēa) is formed from the ace. (sg. or pi.) by suffixation of accented -(é (-é by Leskien's Law; from postposed *é:n 'in'); the PIE and PBS1. loc. pi. in -su is attested in several OLith. sources, where it is the exclusive or predominant ending, and survives in the easternmost Aukstaitian dialects, 24
Both -use and -use occur in dialects, cf. Senn 1966a:133.
170
e.g. akysu (< *akysú), ¿monésu to akis 'eye', tmóné 'man' (Senn 1966a:95-6, cf. Stang 1966:186-7; also -sa in southern Lith. dialects; -se in DaukSa and other OLith. texts). On the whole, however, the language has preserved both the intonations and underlying accentual specifications inherited from Proto-Balto-Slavic. Although the intonation of most nominal inflectional endings in Proto-Slavic cannot be determined (cf. §5.1.4), the accentuation of the great majority of endings in Lithuanian agrees with that of their ProtoSlavic equivalents. A few cases of noncorrespondence between Lithuanian and Proto-Slavic call for some comment. We have seen that PS1. /'-, u-, andff-stemplural and dual oblique case endings are composed of sequences of unaccented stem vowel plus accented ending proper: pi. gen. *-Ij-(í, *-ov-(ú, *-(ú, dat. *-í-(mu, *-u-(mií, -á:-(mií, instr. *-ī-(mi, *-ú(mi, *-á:-(mi, loc. *-i-(xú, *-ú-(xií, *-á:-(xú, du. dat./instr. *-í-(ma, *-ú-(ma, *-á:-(ma. In Lithuanian, however, the dat. pi. desinence consists of accented stem vowel and ending -ms, which has undergone syncope from OLith. -mus. In addition, the dat. du. ending, when stressed, regularly exhibits acute intonation, in contrast to the circumflex of the instr. du.: cf. dat. langám, arkliám, ¿iemóm, katétn, akim, súnúm vs. instr. langam, arkliam, ¿iemom, katém, akim, sunum to lángas 'window', arklys 'horse', iiema, katē 'cat', akis, sunils. Dybo (1981:26-39, passim) ascribes this difference in intonation to an earlier difference in placement of stress: dat. du. -ám, -óm, -em, -im, -urn should continue *amV, *-á:mV, *-é:mV, *-imV, *-úmV, whereas the circumflex of instr. du. -am, -om, -ém, -im, -um to must result from retraction from a word-final short vowel which has since been apocopated, i.e. from pre-Lith. -amV\ -a:mV\ *-e:mV\ *-imV\ *-umV\25 That the only case ending in the plural of /-, u-, and i-stems to deviate from the pattern observed in Slavic is also the only ending which has undergone phonetic change, i.e. syncope, and so altered in shape within the brief attested history of Lithuanian, immediately raises suspicions as to the antiquity of this discrepancy. Although the syllabic dat. ending proper -mus is preserved in Old Lithuanian textss of the 16th and early 17th centuries, the stress never falls on the vowel u of -mus: cf. visómus, akimus, svnumus, piemenimus to visa 'all' (fern., AP 4), akis 'eye' (AP 4), sunus 'son' (AP 3), piemuð 'shepherd boy' (AP 3), Senn 1966a:95. This regularity would appear to confirm the
25
If the Latvian dat. pi. in -m continues the Proto-(East-)Baltic datVinstr. du. ending, the final vowel lost in both Latv. and Lith. was probably short; hence the notation *V\ On the dat. and instr. du. in Baltic, see further Kazlauskas 1970.
171
reconstruction of stress on the stem vowel in the dat. pi., as opposed to the instr. and loc, where it survives to this day on the endings -mis, (dial.) -su. However, there is another possibility, one which not only allows a reconciliation of the accentuation of the Lith. dat. pi. with that of the other pi. oblique endings and the Slavic evidence, but also sheds interesting and welcome light on stress retraction and loss of word-final jers in Slavic. Like other instances of sound change, the loss of u in the dat. pi. ending -mus did not take place overnight. Based on phonetic analyses of studies of sound change in progress in speech communities over the last several decades, one may reasonable deduce that syncope of u involved variation along age, gender, and perhaps class lines across early modern Lithuania, as well as geographical variation.26 This variation, furthermore, did not involve only "full" short u and fully syncopated 0; instead, one can imagine that the duration of this u, whose variation was subject to the sociolinguistic variables just mentioned, became progressively shorter over time, culminating in its disappearanace by the late 17th century. At some point early in the process of M-1OSS, after the vowel had begun to be phonetically weakened or reduced, I propose that u became incapable of bearing stress. As a result, the endings *-a-(mus, *-ó:-(mus, *-é:-(mus, *-i-(mus, *-u-(mus came to be reanalyzed as -a-mus, -ó-mus, -é:-mus, -i-mus, -ú-mus, with accented stem vowel. This shift of stress has been fully carried out by our earliest Old Lithuanian documents, as the forms above make clear. The phonetic loss of the segment u, on the other hand, was not completed until some time later, in the 17th c; from that point on, and in the modern language, the dat. pi. endings must be analyzed as underlying -(á-ms, -(ó-ms, -(é:-ms, -(ims, -(u-ms. The above hypothesis accounts for the acute intonation of the dat. pi. endings: after u had lost the ability to bear stress and the stress had shifted to the stem vowel in -āmus, -ómus, -émus, -imus, -umus, loss of u allowed the sonorant m to form a diphthong with the preceding short stem vowel in -áms, -ims, -urns. In -óms and -éms, by contrast, there is no need to assume an exceptional "trimoraic" diphthong óm: here m after a long vowel belongs to the syllable coda.27 One may contrast this outcome with the consonant-stem gen. sg. and nom. pi. in -(e)s, e.g. OLith. and dial, vandenés, dukterés vs. modern standard 26
0n sociolinguistic factors in linguistic variation and change, cf. most recently Labov 2001. It goes without saying that these new sequences of long vowel + tautosyllabic sonorant were not subject to OsthofPs Law, which shortened long vowels before tautosyllabic sonorant in prePBS1.: *V:R]0 > *VR]C. 27
172
vandens, duktefs. Here the circumflex in the shorter form points to a different phonetic mechanism: if the stress in vandenés had been retracted to *vandenes (or *vandēnes, with lengthening of e in open syllable) prior to the loss of the e of the ending, one would expect "vandéns " with acute intonation. Note furthermore that neither Old Lithuanian nor the dialects attest an intermediate stage "vandénes" or "vandēnes"; cf. Senn 1966a: 138-41. Most importantly, the retraction of stress from and loss of u in the Lithuanian dat. pi. provides a close parallel for the mechanisms behind stress retraction and "neoacute" intonation in forms containing a stressed word-final jer in (pre-)Proto-Slavic. In Slavic language such as Russian and Serbo-Croatian, the weak word-final or -internal jer remains in the underlying representation, so that Accent Retraction and Jer Deletion must continue to in the synchronic grammar. In contrast, the weakening and loss of u in the dat. pi. is isolated within Lithuanian, so far as I can tell. Rather than an assuming an underlying ending -(mVs with ad hoc vowel deletion and accent retraction, it therefore seems preferable to analyze the dat. pi. ending in modern Lithuanian as -(ms: the accent is retracted to the preceding stem vowel, which surfaces with its underlying intonation after unaccented nominal bases of APs 3 and 4, i.e. -ó-ms, -é-ms in the case of o- and é-stems. The parallel between the Slavic and Lithuanian phenomena is illustrated below: *
Russian
*
*
(* (*)
go
lo
vO
—>
*
* (* (*)
go
lóv
* *
Cakavian
*
* (*
(* (*)
gla a va
—>
gla av = gláv
*
Lithuanian
*
*
(* (*)
ma?
ció
mus
*
—>
mar
* (*
t*\
cióms
As for the intonation of the dat. du., I follow Cowgill (class notes, 1980) in attributing the circumflex intonation of -(i)anī, -om, -ém, -im, -urn to secondary differentiation from dat. -(i)ám, -óm, -em, -im, -úm on the basis of o-stem instr. -aīs vs. dat. -áms. Among other IE languages which have preserved dual number in nominal case inflection, Sanskrit and Avestan have the same ending for dat. and instr. (Skt. -bhyām, o-
173
stem dev-é-bhyām with pronominal -e-, ō-stem senābhyām, etc.; Av. -biia [GAv. -biiā], e.g. duuaēibiia to duua 'two', vqββabiia to v0/fa- 'herd, flock'). In Slavic, both cases likewise share the ending -ma which, judging from the scanty evidence of medieval accented documents and reflexes in modern languages, was accented in PS1. for both dat. and instr. Once again, one cannot exclude the possibility of a precious PIE archaism preserved only in Lithuanian. Yet given the existence of a ready analogical model in the ostem dat. and instr., I consider preferable a secondary origin for the intonational distinction in the dual, and unnecessary Dybo's tentative reconstruction of pre-Lith. du. dat. *-amV, *á:mV, etc. vs. instr. *-amV\ *-a:mV\ etc. If the above is correct, the resulting picture for pre-Lithuanian almost perfectly matches that of Proto-Slavic in the plural. Both /'- and w-stems contrast unaccented endings in the nom. and ace. pi. with accented endings in the oblique cases, namely gen. -(hj, -(y, dat. -i-(ms, -u-(ms, instr. -i-(mis, -u-(mis, and loc. *-i:-(su for *-i-(su, *-u-(su < pre-Lith. gen. *-iy-(ō:n, *-uw-(ō:n,28 dat. *-i-(mus, *-u-(mus, instr. *-i-(mins, *-u-(mins, loc. *-i(su, *-u-(su. The latter, like their PS1. counterparts, may be analyzed as sequences of (unaccented) stem vowel *-i-, *-u- plus ending proper, i.e. gen. *-(ō:n (dat. *-(mus, instr. *-(mins, loc. *-(su. Similarly, the d-stem oblique plural desinences consist of unaccented *-á:- and the same endings: gen. *-(o:n, underlyingly probably -á:-(ō:n with deletion of the first vowel, dat. -á:-(mus, instr. *-á:-(mins, loc. *-á:-(su (—> standard -ose). In the ostems, all of whose endings should originally have been unaccented (cf. §5.1.1 on Slavic), the accentuation of dat. *-a-(mus (> -dms) can easily have been modeled on that of the other stem classes; similarly for gen. *-a-(ō:n (once again with deletion of the first of two consecutive vowels), which, except for the underlying accent, directly continues PIE *-oHom (Ved. -ām, GAv. -qm, often scanned as disyllabic -aām, -aqm). We therefore arrive at the following scheme for pre-Lithuanian: o-stems a- stems nom. sg. -as -(I:s* -(á: -(e~: -(e:s gen. -(ā:s -ā: -e:I -o:I -a:i dat. ace. -an -an -en instr. -ó: -án -én loc. -Gevoc. -e -(i: -a -e
/-stems -(is
-(iēs
«-stems -(us -(aiis
n-, r-stems -(ō:,-(ē: -(es
**
**
**
-in
-un
X
-i-(mi**
-u-(mi
χ**
-(iē
-(au
X
28
For the u-stem gen. pi., one may assume the application of a synchronic morphophonemic rule deleting the first of two vowels in -u-(\j > -y. Gen. pl.-ty in /'-stems is the regular development of *-ijy, cf.y'o-stem brólis 'brother', gen. pi. bróliiļ, arklys 'horse', gen. pi. arklii}.
174
n./a. du. dat./instr.
-ái -ái -ó: -a-(mV -á:-(mV -é:-(mV
nom. pl. gen. dat. acc.
-(āi
instr. loc.
-(ō:n
-a-(mus -ó:s
-(aís
-ā:s
-í: -i-(mV
-ē:s
-(5:n -á:-(mus -é:-(mus -áns -é:s
-Is -iy-(ō:n -i-(mus -í:s
-á:-(mins -é:-(mins -i-(mins -á:-(su -(é:-su
*
-u: -u-(mV
X X
-u-(mus -ú:s
-es -(ð:n X X
-u-(mins
X
-ú:s -(ō:n
So far as I am aware, these accentual values are completely consistent with the Latvian evidence; in any case, the Latvian data are of comparative value only in so far as the intonation of their initial syllable reflects acute accented, acute unaccented, and circumflex initial (root) syllables in Proto-(East-)Baltic. The extremely scanty evidence of the Enchiridion tells us little about intraparadigmatic stress alternations in Old Prussian, but the pattern of nom. sg. fern, antra 'second' vs. acc. āntran, spigsnā 'bath' vs. spīgsnan, and of acc. sg. mērgan 'girl' and dat. pl. mergumans, exactly parallels that of e.g. Lith. iiema, iiēmq, íiemóms. Stang (1957:60-1, 1966:173-4) is surely correct in interpreting these as unaccented stems with mobile stress; based on the alternations, we may tentatively presume that OPr., like Lithuanian, reflects the unstressability of pre-PBSl. * in ō-stem acc. sg. -an (§5.1.2) and the retraction of stress from ending to stem vowel in mergumans — if this is not an archaism, i.e. does not continue original pre-PBSl. *-(a:-mas (§5.1.1). Other discrepancies between (pre-Lithuanian and Slavic are less easily explained than that of the dat. pl. and du. Most notably, the /- and «-stem nom. sgs. in -(is and -(us are underlyingly accented in Lithuanian: cf. paradigm 3 Sirdis, ¿vérls, sunus (acc. širdļ, ¿vérļ, sunu). Although the accented gen. ending of these two stem classes may readily be connected with PIE proterokinetic *-éy-s, *-éw-s, one would expect an unaccented ending and hence unaccented enclimenon forms in the nom. sg. Here I can only appeal to analogy from the n- and r-stems in -uo, -e, which continue a mixture of PIE aerostatic and hysterokinetic paradigms: «-stem aerostatic *-ō, with accentuation from hysterokinetic *-e: (which does not survive in Lithuanian); r-stem aerostatic *-o(r), likewise accented on the analogy of hysterokinetic *-ē.29 Finally, the nom. pl. of o-stems is accented -(ai, in contrast to that of all other stem classes: -os, -és, -ys, -us (<— *-ewes), OLith. and dial. -(e)s. Garde (1976a:32) suggests 29
The correspondence of the accentual alternations of these consonant-stems with their Sanskrit and Greek cognates was noted by de Saussure already in his famous article of 1896.
175
that this accentuation has spread to the originally masc. ending *-ai from neut. nomVacc. pi. *-(a (cf. PS1. *-(a, Russ. polj-á, SC pdlj-a to unaccented pólj-e, polj-e) which, along with neuter inflection in general, was subsequently lost in East Baltic. Such an influence of neuter inflection on a masculine ending might seem unlikely, but a rough parallel is furnished by Russian plurals in -á, which first become common in the 18th c. and increase in productivity throughout the 19th, encompassing many loanwords (professor-á, pasporta). As Stang (1957:16-7) correctly notes, this chronology makes it very unlikely that the ending is in origin that of the o-stem nomVacc. du. -a, which was almost certainly not underlyingly accented in any case;30 he therefore proposes that accented -á has been generalized from the neuter. Whether or not neut. pi. *-(a is responsible for the accentuation of Lith. -(ai, it is worth noting that an accented nom. pi. ending would tend to have been reinforced by the inverse morphophonological relationship between the accentuation of the nom. sg. and that of the nom. pi.: o-stems
nom. sg. unaccented -as
ā-stems /-stems M-stems cons.-stems
accented -(ó: accented -(is accented -(us accented-(uð,-(é:
nom. pi. accented -(ai vs. unaccented -ōs unaccented -ys unaccented -u:s unaccented-es
5.3. Conclusions: PIE survival and Balto-Slavic innovation Based on the results of the previous two sections, the underlying accentuation and intonation of Proto-Balto-Slavic nominal inflection may be reconstructed as follows:
30
Cf. Lith. l-úol, unstressed in AP 3 nouns such as lángu 'two windows', kātilu 'two kettles" (similarly fern, /-id in gálvi 'two heads', pāmoki 'two lessons'). The pattern of unaccented nom.(/acc.) vs. accented oblique in the pi. also suggests that the nom./acc. du. ending, in opposition to the gen./loc. and dat./instr., should be unaccented, cf. Slov. du. moid 'two men' to unacc. mq¿. Stang (1957:17-8, 77-82, esp. 80-2) emphasizes that the stress of Russ. dva Casa 'two hours', dva rjadá 'two rows, lines', must be considered together with other oxytone forms such as nom./acc. pi. rjady, adverbial instr. sg. verxóm 'on horseback' and need not preserve the old dual stress. Contra Stang (18), Bulg. pi. rogá, kraká to rog 'horn', krak' leg' are likely to be old duals: the original stress *róga, *kráka (cf. the brojna forma, i.e. "counting form" after numerals '2' and above, e.g. mnógo róga 'many horns', Cetiri kráka 'four legs') would then have been influenced by neut. pi. -á.
176
(9-stems
nom. sggen. dat. ace. instr. loc. voc.
-as
-an
-ā:
-(ā:s -ā:i
-ān
-ān3i -á:-(mi -a:i -a
-ð:i
-e
d-stems -(á:
-ó: -aí -an
/-stems -(is -(els* -ei? -ín -i-(mi -ei -(eí
«-stems -(us -(eus* -ewei -ún -u-(mi -eu -(eú*
-eyes, -Is -ey-(ō:n -i-(mas -í:s -i-(mins -i-(su
-awes -ew-(ō:n -u-(mas -ú:s -u-(mins -u-(su
cons.-stems -(ō:, -(ē:
-(es -ei -In -(mi -i???
-0 "
nom./acc. du. dat./instr. nom. pl. gen. dat. acc. instr. loc.
-ai
-(á: -ð:n -a-mas -ó:s -aís -aí-su
-(ā:s -(ō:n
-á:-(mas -á:s -á:-(mins -á:-(su
-es -(ō:n
-(mas -í:s -(mins -(su
*or -(als, -(aus, -(aú A glance at this table reveals that PBS1., in addition to having undergone a number of innovations in accentual mobility, has preserved several important features of the PIE system of nominal accentuation presented in §2.2. We have seen in §5.1.1 that once one peels away the influence of other stem classes on the oblique plural of o-stems in both Lithuanian and Slavic (in particular that of «-stems, whose effect on and incipient inflectional merger with o-stems in Slavic may be observed already in the OCS period), unaccented masculine o-stems are unaccented throughout their paradigm. Similarly, neuter o-stems are unaccented in the sg., and probably were also in oblique cases of the pl. The accentuation of the nom.-acc. pl. in *-(á:, which continues accented PIE coll. *-(eh2, has spread to the rest of the pl. in Slavic at least by the time of the accented medieval Russian and Bulgarian documents, and perhaps already by the Proto-Slavic stage; in Lithuanian the ending, along with neuter gender in general, has disappeared, but may have exerted some influence on the masc. nom. pl. ending -(ai for unaccented PBS1. *-ai (> PS1. *-i; cf. the discussion in §5.2 above). Most importantly, the non-o-stem classes exhibit a consistent contrast in the plural between unaccented nom. and acc. endings on the one hand and accented oblique case 3 'The circumflex intonation has either arisen by analogy to o-stem *-an, /"-, cons.-stem *-in. H-stem *-un or has regularly developed from PIE and pre-PBSl. *-ān (cf. §5.1.3); in the latter case, the acute intonation of the acc. pl. must have resulted from *-ans <— *-ās.
177
endings on the other. This distribution, of course, corresponds exactly to that of "strong" and "weak" cases in PIE ablauting nominal stems. As observed in §2.2.1, two types of iand «-stems, proterokinetic and hysterokinetic, may be reconstructed for PIE: the latter alternate between suffixal stress in the strong cases and ending-stress in the weak cases; the former — by far the better and more widely attested in the daughter languages — stress the root and suffix in the strong and weak cases, respectively. In the plural, however, most ancient IE languages have generalized a single form for each case suffix, as the following chart makes clear: nom. p]. gen. dat./abl. ace. instr. loc. nom. pi. gen. datVabl. ace. instr. loc.
PGr. Skt. Av. *-eyes -ayas -aiiō [-lnām] [*-īnām] *-eyo:n -ibhyas -ibiiō (-In, -īs) *-inS *-ins *-ip h i -ibhis -ibiS *-isi -isu -išu
Goth. -eis
Lat.
Lith.
PS1. -*i *-ījī
-ium -ibus
(-*)
-ys -/y
-im
-im(u)s
*-lmú
-Is
-ins
-is
*-i
-imis *-isu
*-ími ♦■3x5
-ēs
*-ewes -US -avas -auuō [-unām] [*-unam]*-ewo:n -uum -ubiiō -ubusl -ubhyas (-u«, -us) *-unš *-uns -us *-ewas -ubhis -ubiS [*-esi] ■usu -uSu
PIE? *-ey-es *-ey-ōm *-i-b h os *-i-ns *-¡-bh¡s *-i-su
-um
*-ov-e *-ov-ú -um(u)s *-úmú
-uns
-us
*-u
*-ew-es *-ew-ōm ♦-ii-t^os *-u-ns
-umis -usu
*-úmi *-úxú
*-u-b h is *-u-su
-jus (-is) (-iw-ē) ■v
It is obvious that the distribution of ablaut variants of the stem vowel *-ey- ~ *-i-, *-ew- ~ *-u does not correspond to the division between strong and weak cases. Instead, the evidence of Indo-Iranian, Greek, Latin, Germanic, and Balto-Slavic points to a pattern of egrade *-ey-, *-ew- before endings beginning with a vowel, i.e. nom. *-es, gen. *-ōn, and 0-grade stem vowel *-i-, *-u- before endings beginning with a consonant: for the nom. and ace, cf. Gr. -e:s < *-eyes vs. -i:s < *-ins (cf. Cretan trees, triins <— *trins '3'), ProtoItalic *-ēs < *-eyes vs. *-Is (e.g. Osc. nom. trís [trēs] < *treyes, Umbr. trif, tre(f), trif. treif[trēf] < *trins, Buck 1904:126-7; preserved in some Lat. /-stems, e.g. nāvēs, nāvīs 'ships'), Goth, -eis < *-eyes vs. -ins (e.g. gasteis, gastins 'guests'). In the gen., the only case for which both ablaut grades are well attested, zero-grade in Lith. -y < *-uwo:m (Lat. -uum may continue *-ewōm, *-owōm, or *-uwōm) can easily be analogical to the other oblique cases and/or to the i-stem gen. pi. in -i'y, cf. §5.2. Such a pattern squares poorly with the pattern of the various ablaut- and accentalternating types described in §2.2: one would have to assume that full-grade suffix *-eywas generalized from hysterokinetic stems, and zero-grade *-i- from proterokinetic stems. 178
precisely before vowels and consonants, respectively.32 Unfortunately, neither Sanskrit nor Greek furnishes any information as to the underlying accentuation of these endings, as i- and «-stem nouns in both languages are regularly columnar in their accentual paradigm. Here, then, the evidence of Baltic and Slavic is especially crucial for the reconstruction of PIE accent. It has been argued in §5.1.1 that the nom. and ace. pi. endings were unstressed in PS1., whereas the oblique case desinences were stressed on the ending proper; since the accentual system of Old and Modern Lithuanian may be derived from the same starting point, and I am not aware of any contradictory evidence from Old Prussian, this reconstruction may be assumed for PBS1. If the accentuation of the oblique endings is old, it follows that at some point in late or post-PIE or in the prehistories of the individual IE languages they were remodeled on the basis of the root-noun and consonant-stem endings, which were likewise stressed in the pi. (cf. §2.2.1). gen. pi.
cons.-stems -(o:n
dat. instr. loc.
-(mas -(mins -(su
—>
/-stems -ey-(ó:n -iy-(ó:n -i-(mas -i-(mins -i-(su
«-stems -ew-(ð:n -uw-(ð:n -u-(mas -u-(mins -u-(su
—>
a:-stems -á:-(o:n -á:-(mas -á:-(mins -á:-(su
Thus, Balto-Slavic appears to have been conservative in preserving the (post-)PIE accentual alternation of unstressed nom./acc. vs. ending-stressed obi. in the plural of /'- and «-stems. Subsequently, stress retractions connected with the loss of word-final jers in Slavic and «-deletion in the Lithuanian dat. pi. obscured this picture somewhat, but the original accentuation has survived in Slavic forms such as Russ. ljud'mi, det'mi (cf. also gen. pi. -cov < *-ovu in Russ. dialects with a contrast between co and o) and in the Lith. instr. and (Zemaitian) loc. This pattern has in turn influenced the d-stems, which must originally have had accented *-(á:- throughout: pi. dat. -(á:-(mas,* -(á:-(mi:s, *-(á:-(su have been replaced by *-á:-(mas, *-á:-(mi:s, *-á:-(su after *-i-(mas, *-i-(mi:s, *-i-(su and the corresponding i<-stem endings.
32It is not impossible that at least the contrast between nom. *-ey-es and ace. *-i-ns reflects a difference in the accentuation of the two case endings: recall that the ace. pi. patterns as a "weak" case in Sanskrit, e.g. nom. pi. pād-as 'feet', ad-ánt-as 'eating', pit-ár-as 'fathers' vs. ace. pad-ás, ad-axas, pit-f-n (and presumably also Avestan, which shares the same ablaut alternations in e.g. dpō 'waters', vācō 'voices', pres. act. ptcp. haņíō 'being' vs. apð, vacō, hatō).
179
Alongside these archaisms, Balto-Slavic has of course introduced a number of innovations which have given rise to some of the more salient patterns of accentual mobility in Lithuanian or Russian. One of the most obvious was the unstressability of the vowel *ā, which arose in a number of eh2-stems endings either already in PIE by Stang's Law (ace. sg. *-ām < **-eh2m, ace. pi. *-ās < *-eh2ns) or as a result of early laryngeal loss and contraction (dat. sg. *-āi < *-eh2-ey), which gave rise to the mobile d-stem paradigm of the type of Lith. iiema, iiēmq, Russ. zimá, zímu, SC zíma, zīmu (ná zīmu), so characteristic of those languages. The innovation which most sets off Balto-Slavic from other accentually archaic IE languages such as Sanskrit or Greek, however, is undoubtedly the spread of accentual mobility from the consonant-, /-, and «-stems to the originally columnar ā- and o-stems. We have noted above that the contrast between unaccented nomVacc. and accented oblique endings in root nouns and consonant-stems, which may be reconstructed with complete confidence for the PIE protolanguage, may have influenced the /- and «-stems in prePBS1., particularly in the accentuation of the endings of the plural oblique cases (§§5.1.1, 5.2). At least in the singular, however, this contrast continues that of PIE proterokinetic iand «-stems, and the remodeling of the plural may quite possibly be a late or post-PIE innovation shared by other IE branches, given the conflicting evidence for the reconstruction of /- and «-stem plural endings. Be that as it may, both Sanskrit and Greek (probably independently) have generalized columnar stress from o- and ō-stems to the paradigms of /-, u-, n-, r-, and sstems — in short to almost all nouns other than monosyllabic root nouns. As a result, the original PIE system of interacting accent and ablaut alternations must be reconstructed on the basis of ablaut grades alone in such paradigms as the following: nom. sg. voc. ace. gen. datVabl. instr. loc.
Sanskrit gur-ús gúr-o gur-úm gur-ós gur-áve gur-úņā gur-āu
NVA du. IDA gen./loc.
gur-ú gurú-bhyām gurv-ós
Greek (Horn.) bar-ús »9
bar-ún bar-é(w)os baré(w)i bar-é(w)e [bar-é(w)oiin] 9»
180
Sanskrit pitā pítar pitár-am pitúr pitr-é pitr-ā pitár-i pitár-ā(u) pitŗ-bhyām pitr-ós
Greek (Horn paté:r páter patér-a patr-ós patr-í (dat.) patér-e [patér-oiin] i*
nom./voc. pi. ace. gen. dat./abl. instr. loc.
gur-ávas gur-ún gur-úņām gur-úbhyas gur-úbhis gur-úsu
bar-é(w)es har-é(w)as bar-é(w)o:n bar-ési
pitár-as pitfn pitŗņám pitŗ-bhyas pitŗ-bhis pitŗ-su
patér-es patér-as patr-5:n patrá-si (dat.)
Here the original alternating stress in e.g. nom. gurus, bams vs. gen. gurós, baré(w)os, inferred on the basis of the suffixal variants *-u- vs. *-ew-, has been eliminated in both languages, as has all trace of root ablaut. Similarly, expected Skt. "pitŗ-bhyām", "pitŗbhyás", "pitŗ-bhís", "pitŗ-sú", Gr. "patra-sr have been replaced by forms with stressed zero-grade suffix. The only paradigms which preserve mobile stress, and therefore the contrast of unaccented and accented endings so characteristic of PEE, are root nouns such as Skt. pāt, acc. pād-am, gen. pad-ás, Gr. pó.s, acc. pód-a, gen. pod-ós, and a handful of exceptional nouns such as Gr. guns: 'woman', gen. gunaik-ós; see §§2.2.1.1, 2.2.1.5 (end) for details. In contrast, Balto-Slavic has extended the accentual alternations in the plural of consonant, /'-, and «-stems to the more numerous d- and o-stems. For d-stems, this innovation is common to Baltic and Slavic and so may safely be reconstructed for ProtoBalto-Slavic: earlier pi. dat. *-(á:-(mas, instr. *-(á:-(mins, loc. *-(á:-(su, du. dat./instr. *(á:-(ma, with accented stem vowel *-(á:-, have become *-á:-(mas, *-á:-(mins, *-á:-(su, *á:-(ma with stressed endings proper *-(mas, *-(mi:s, *-(su, *-(ma, just as in the /- and ustems. Thus d-stems have thus come to acquire an alternation in the plural (and dual) between unaccented nom. *-ā:s, acc. *-á:s and ending-stressed gen. *-(ð:n, synchronically probably *-á:-(ó:n, and dat. -á:-(mas, instr. *-á:-(mins, loc. -á:-(su, fully parallel to that in the /- and «-stem plural. For o-stems, the evidence is less unanimous, due to a greater amount of remodeling of endings in both Baltic and Slavic. As we have seen, pre-Lithuanian attests gen. *-(ij, dat. *-a-(mus, and instr. *-(a!s in contrast to nom. *-(ai and acc. *-ó:s; if the accentuation of the nom. pi. ending is secondary (cf. §5.2), we recover the same division between nom./acc. and oblique found in the other stem classes. In Slavic, however, mobile stress in unaccented o-stems appears to be a parallel yet independent innovation of the separate languages (most of which have in any case generalized the endings of one or another stem classes and/or undergone case syncretism in the plural, e.g. Russian, standard Serbo-Croatian): although Old Russian and Middle Bulgarian (or rather MidBulg. Church
181
Slavonic) accented texts already attest reflexes of ending-stressed plural oblique forms, the archaic eakavian dialects of Novi and Orlec still preserve unaccented variants, cf. §5.1.1. Other facts, such as the predominance of stem-final palatalization among Russian o-stem masculine nouns with alternating plural stress, and the near-total limitation of alternating plural stress in Serbo-Croatian to monosyllabic nouns with the infix -ov- (<— «-stems), strongly suggest that accented oblique plural endings are not original to the o-stems, but have been introduced from the /'- and «-stems. It therefore appears likely that the accentuation of o-stem oblique plural endings is a post-PBSl. innovation, which has arisen separately in Lithuanian and a number of Slavic languages. This contrast in the accentual evolution of Balto-Slavic on the one hand, and Sanskrit and Greek on the other, may be depicted in simplified fashion as follows: o-stems
d-stems
i-, «-stems
cons.-stems, root nouns
PIE
columnar
columnar
mobile
mobile
Skt., Gr.
columnar
columnar
columnar
columnar
PBS1.
columnar mobile (—> mobile)
mobile
mobile
In closing, it is worth noting that once one subtracts the two major innovations just discussed, namely the "deaccentuation" of certain d-stem endings and the spread of accentual alternations from non-thematic stem classes to the plural of o- and d-stems. Balto-Slavic has preserved more of the PIE system of unaccented and accented nominal case endings, and the mobile stress connected with them, than either Sanskrit or Greek. This may seem a trivial observation, given the complexity of accentual alternations of Lithuanian and Slavic languages such as Russian or Serbo-Croatian: unless one wishes to dismiss them being of somehow "secondary" or unrecoverable origin, these must in some way continue the alternations reconstructed for PIE on the basis of accent and ablaut alternations in the two classical languages. Yet the latter have almost entirely generalized columnar stress outside of root nouns and certain specific classes of suffixed nouns, so that the evidence for the elaborate PEE accent-ablaut system advanced by Eichner, Narten, and the like is founded upon four principal sources (cf. §2.2.1):
182
1. stress alternations in Sanskrit and Greek root nouns, and in certain Sanskrit suffixed classes, e.g. nt-stevn pres. act. ptcp. masc. sg. ace. ad-ánt-am vs. gen. ad-at-ás (plus a few longer nouns in Greek, e.g. gen. gunaik-ós, ace. gunaik-a): 2. variation in ablaut grades in Sanskrit and Greek; and 3. relics of archaic accent and ablaut in Anatolian (e.g. Hitt. nomVacc. tēkan 'earth', dat. taknī, loc. tagān (tāgan), nomVacc. wātar 'water', gen. witen-aS, pl./coll. widār); and 4. relics of ablaut alternations in Germanic, Italic, Celtic, Tocharian, Armenian, etc., particularly in r- and «-stems. With the exception of Sanskrit and Greek root nouns and a handful of archaic forms in Hittite, then, it is Balto-Slavic which furnishes the direct evidence for the placement of stress, and hence for the underlying accentual specification of individual roots, stems, and endings, in PIE. This is a crucially important and widely overlooked fact: despite their late date of attestation, the Baltic and Slavic languages must assume a primary role in any future treatments of PIE accent. That this has not so far been the case is to be ascribed not only to the undeniable complexity of the surface facts, but also to the failure of previous studies to separate archaism from innovation and integrate the facts of Balto-Slavic accent into IE historical linguistics. Perhaps the most striking and pleasantly surprising result of the investigation in §§5.1, 5.2 is that the alternations arrived at for Proto-Balto-Slavic on the basis of internal reconstruction from the Baltic and Slavic data alone almost exactly match those posited for PIE from the four sources listed above. Moreover, virtually every innovation to have affected the Baltic and Slavic languages may be explained through eminently plausible analogies.
Once these are set aside, the forbidding maze of
complicated and often conflicting Balto-Slavic facts becomes reduced to an eminently transparent system, whose evolution, and value, may be appreciated and grasped by all Indo-Europeanists.
183
Chapter 6. More surprises in store: The evidence of Balto-Slavic accentuation for the Proto-Indo-European verb 6.1. Introduction: simple thematic presents in (post-)PIE Until a generation or so ago, simple thematic presents figured among the most secure reconstructions of IE historical linguistics. These presents, characterized by full grade of the root followed by the thematic vowel *-e/0-, are represented in every major branch of IE which was known to the Neogrammarians and figured in Brugmann's famous Grundrifi, as illustrated by the following reflexes of the 3sg. and 3pl. of *bher-e/0'bear, carry', *h2ed-e/0- 'drive': Vedic Avestan Greek Latin
ocs Gothic Old Irish Tocharian B Armenian
*bher-e-ti, *bher-o-nti bhárati, bháranti baraiti, baraņti phérei, phéro:si feruferunt bereií, bergtī (-/u) baíriþ, baírand •beir, -berat párā-m, párem beré, (beren)
*h2eá-e-ti, *h2eg-o-nti ájati, ájanti azaiti, azaņti ágei, ágo:si agit, agunt vezetī, vezgtī (-tu) ON ekr, aka ■aig, -agat āšā-tņ, ākem
Alongside this present type, PIE also possessed various derived thematic presents, formed by means of a suffix containing the thematic vowel. By far the most common and widely attested suffixes are *-yé/<5- and *-s¿é/ó-: both are underlyingly accented, and the preceding root appears with "zero-grade ablaut", e.g. Gr. βccívco, Lat. veniō < *gwm-yé/<3-, Ved. gácchati, Av.jasaiti (for *gasaiti) 'go, come', Gr. βáaKco 'come' < *gwm-slcé/ó-, forming presents to the root aor. *gwem- 'come'.1 This consistent picture was disturbed by the discovery of two new IE branches at the beginning of the 20th century, namely Anatolian and Tocharian. Whereas simple thematic presents do survive as a present type in Tocharian (class II in the scheme of ]
Cf. Rix 1976:213-4. The root aorist *gwém - *gwm- is attested by 3sg. Ved. ágan, gan [for *ájan, *ján], GAv. Jan, Arm. e-kn < *(é-)gwém-t, 3pl. Ved. ágman, gmán, GAv. g3man [metrically monosyllabic] < *(é-)gwm-ént, iptv. 2sg. Ved. gadhi, gahi, GAv. gaidī, 3sg. Ved. gántu [for *jántu]. GAv.Jantu < *gwiņ-dhí, *gwém-tu. TB šem probably continues a thematized *gwem-e-s, *-e-d <— *gwēn < PIE *gwém-s, *gwém-d; cf. R. Kim 2001b.
184
Krause and Thomas 1960), only three Tocharian class II presents, including the two listed above, actually have simple thematic cognates in other IE languages (cf. Ringe 2000:12131).2 Worse yet, the Anatolian languages, although attesting *-slcé/ó- (> Hitt. -ske/a-, Luv. -z(z)a-) as a productive iterative-imperfective suffix and preserving numerous examples of old derived presents in *-yé/<5-, lack any certain reflexes of simple thematic presents. Recent scholarship (e.g. Jasanoff 1997b, Ringe 2000) has attempted to reconcile the absence of simple thematic presents in Anatolian, and relative rarity in Tocharian, with their abundance in the "classical" branches. Following the emerging consensus that Anatolian and Tocharian were the first two branches to "split off from the protolanguage, these studies propose that true PIE, i.e. the common ancestor of all the attested IE languages, did not in fact have simple thematic presents such as *bher-e/0-, *wegh-e/0-: their creation is a common innovation of non-Anatolian IE (or "IE Proper", in the terminology of Cowgill (1979) and other adherents of the old "Indo-Hittite" hypothesis), following the separation of Anatolian. This process was still in its early stages when Tocharian separated from the other IE branches, thus accounting for the scarcity of the type in Tocharian. Whatever its ultimate origins or early prehistory, the simple thematic present had become well established as a major inflectional type by the stage of Proto-"NucIear IE" or "Brugmannian IE", i.e. the immediate ancestor of the "classical" IE languages known to Brugmann's generation of Neogrammarians, including Greek, Italic, Germanic, BaltoSlavic, and Indo-Iranian. The present chapter considers the accentuation of these presents in Slavic and Baltic. The results of this investigation are not only of interest for the reconstruction of the accent of the (post-)PIE verb, but shed light on the historical development of accentual alternations in general. It will be seen that the brackets-and-edges model of stress provides a plausible synchronic and diachronic interpretation of the BS1. facts and their origin and development, along with various distributional facts of Slavic verbal morphology which have until now remained opaque.
2
The third, TB 3sg. Saim, 3pl. šāwem (and šāyem) < PT *šawya - - *sawē- < PIE *gwih;(w-e/0-, is cognate with Ski. jívati, Av.juuainti, Gr.zo.o:, Lat. vivo, OPr. giwa (§6.5), OCS ¿ivetu (cf. fn. 27). This verb is clearly not in the same class as 'bear' and 'drive'; it must be some kind of early denominative to *gwih3w-ó- 'living, alive' (Ved. jīvá-, Lith. gfvas; §§4.2.2, 4.3.2).
185
6.2. Extra-Balto-SIavic evidence for the accentuation of the simple and characterized thematic present Since Brugmann, every IE handbook and general reference has assumed root stress for the simple thematic present, i.e. columnar stress on the syllable before the thematic vowel *-e/0-: hence *bhér-e/0-, *wégh-e/0-. Yet how well supported is such a reconstruction? To be sure, Vedic simple thematic presents, i.e. presents of class I, attest columnar root stress, e.g. 1 bhárāmi, bhárasi, bhárati, bhárāvasi, bhárathas, bháratas, bhárāmasi, bháratha, bháranti, and it is this pattern that has traditionally — and tacitly — been projected backwards to PIE. Other IE languages, however, offer distressingly little support for the Vedic pattern, for instance, Greek presents such as phéra:, phéreis, phérei superficially share the columnar root stress of Vedic, yet it has long been recognized that finite verb forms in Greek have generalized the enclitic variants originally proper only to main clauses (Wackernagel 1877:457-8; cf. §3.3.1.). The "recessive" accent oiphéro: and other finite verbal pardigms is the result of the Three-Syllable Rule, which assigns surface stress to the mora farthest to the left permitted by the final syllable; in Hoenigswald's words, "recessivity is the guise taken perforce by their former enclisis, which can be reconstructed by comparative evidence" (1998:271). Only in Germanic is there indirect evidence for the placement of stress in simple and suffixed thematic presents. Whereas the 3sg. ending in West Germanic languages continues PGmc. *-p (cf. OE bir-p 'bears, carries', hilp-p 'helps'), the Gothic 3sg. ending is underlyingly 1-61; the latter, like 2sg. l-zl and 2pl. /-ð/, is regularly devoiced in word-final position, but surfaces when followed by an enclitic such as -u (interrogative particle), -uh 'and, so, then', e.g. 2pl. qipip 'ye said', qipid-uh 'and/so ye said'.3 Thus both *-p and *-ð must be reconstructed for PGmc; the contrast in voicing suggests that these are in origin alternants arising from the operation of Verner's Law, with *-p continuing pre-PGmc. *-éti and *-ð the Verner's Law outcome of pre-PGmc. *'-eti (Ringe 2001:126fn.4).4 As the former may immediately be identified with the various derived presents with stress on the thematic vowel — in particular *-yé/ó-, the suffix underlying the "weak" verb formations ^Unfortunately, there appear to be no examples with 3sg. -d-* for -þ. Cf. Campbell 1959:297fn. 1 on the necessity of deriving 2sg. *-isi, 3sg. *-iþi < *-ési, *-éti. — By itself, of course, voiced PGmc. *-ð could also continue pre-PGmc. *-eti, with *t followed by stressed hic-et-minc particle *-i (see below), but this is ruled out by Verner's Law alternations in rootfinal obstruents; cf. below. The reconstruction of PGmc. *-éti, *'-eti is motivated by the syncope of / in the OE endings 2sg. -(i)s, -(i)þ (e.g. wierþ 'becomes' in the table below), which requires final *-i to have survived into the prehistory of OE; for full discussion see Ringe 2002. 4
186
of Germanic — the only logical origin for *-ð must lie in the simple thematic presents, the source of the PGmc. "strong" verb classes, and indeed a PLE *bhéreti with the stress of Ved. bhárati will produce PGmc. *biriði > Goth, baíriþ, baírid-u*. This inference is corroborated by the various "Verner's Law" alternations in Germanic strong verbs, in which a PIE root-final stop develops to a voiceless fricative in pres. and past sg. but to the corresponding voiced fricative in the past pi. and past ptcp.: in thefirsttwo categories, the stress must have preceded the stop, i.e. fallen on the root vowel, whereas the stress in the latter categories was on the syllable following the stop. pres. 3sg. pres. inf. pf. 3sg.5 pf. 3pl. past ptcp.
Sanskrit várt-a-ti
post-PIE *wért-e-ti
—
—
va-várt-a va-vŗt-úr (yŗt-tá-
*we-wórt-e *we-wŗt-ér
*wŗt-tó-)
PGmc. *wirþ-iði *werþ-ana. *wárþ *wurð-un *wurð-an-az
OE wierþ weorð-an wearþ wurd-on (se)word-en
OHG wirdit wérdan ward wurtum (gi)wori
Examples of this pattern are numerous, particularly in OHG and OE: cf. OE cēosan, teas, curon, (ze)coren, OHG kiosan, kōs, kurun, gi-koran < PGmc. *kéusan$, *káus, *kúzun, *kúzanaz, ; for further examples and discussion, cf. Streitberg 1896:126-30, Campbell 1959:163, 303ff., Braune 1975:98-100, 272-3. (Pre-)PGmc, then, must have contrasted *-ði in strong verbs with *-pi in weak verbs: Gothic and OHG have generalized the former, and OE the latter, with no trace (as far as I am aware") of the original distribution of variants. The traditional reconstruction of columnar root stress in the (post-)PIE simple thematic present thus finds confirmation in only two branches: Indo-Aryan, where Sanskrit directly attests the stress, and Germanic, where the effects of Verner's Law (and the voiced 3sg. variant ending *-ði < *-'eti) point umambiguously to root stress. Here, then, more than in many other grammatical categories, the evidence of Balto-Slavic takes on central importance for the proper reconstruction of PIE accent. Although it has been known for decades that the reconstructible Proto-Slavic accentual pattern of the simple 5
The past forms of "strong" verbs in Germanic continue the PIE reduplicated perfect, with dereuplication of the sg. in all classes and of the pi. in classes I-HI (PIE roots of the shape *CeyC*CewC-, *CeC- > PGmc. *Ci:C-, *CeuC-, *CeC-/*CiC-). In classes IV and V, PGmc. *m < *e in the pi. results from deletion of the First consonant of the root in a reduplicated perfect with lengthening of the reduplication vowel *e, e.g. 0-grade 3pl. *kwe-kwm-un —> *kwēmun 'came' (vs. o-grade 1. 3sg. *kwe-kwam —> *kwam by dereduplication); cf. Cowgill 1957:109-21, 1980:71 (§56) and McCone 1986:236-7, who offers a parallel account of the Olr. ó-pret. in the context of a possible "Western IE" innovation shared by Celtic, Germanic, and Italic.
187
thematic present is decidedly non-columnar (§6.3), this idiosyncrasy has remained generally unknown to the majority of Indo-Europeanists, with the result that its wider historical significance for PIE has never been fully appreciated. 6.3. The accentuation of the simple thematic present in Proto-Slavic 6.3.1. The evidence In contemporary standard Russian, simple thematic presents exhibit columnar stress on the thematic vowel: e.g. to peC 'cook', we have pekú, peðeš, peðēt, peðém, peðete, pekút. A synchronic analysis of Russian would therefore require an underlyingly unaccented root and accented thematic vowel -(e/0-; the metrical grids for lsg. pekú and 3sg. peCet would then be as follows: * *
pe
*
(* (*)
*
(* (*)
hi
pe
Set
Data from elsewhere in Slavic, however, make it clear that this pattern is not the original accentual distribution reconstructible for Proto-Slavic, but is rather the result of at least two innovations. In the first place, Old Russian manuscripts with accentual markings preserve an earlier system in which the lsg. of simple thematic presents was unaccented, i.e. took default initial stress in the phonological word, e.g. íívu 'I live, am living', *né Zivu 4I do not live, am not living'. In the earliest accented Russian text, the Cudovskij Novyj Zavet, one finds lsg. enclinomena such as ¿ívu, réku,pó-trjasu,prí-vleku,pó-loiju,pó-gublju. pó-ienu, né pošðafju (Stang 1957:109-13, 118-9, 1966:450, Dybo et al. 1990:61ff, Lehfeldt 1993:26, 55, 84-5 with refs.). Such forms clearly reveal that the lsg. of unsuffixed thematic presents in ORuss. had no underlying accentual specification, neither on the root nor the theme vowel/ending -u. Of the other person-number forms, the 2 and 3sg. and 1 and 3pl. stressed the thematic vowel, as in present-day Russian, but the 2pl. originally stressed the ending -té and only gradually shifted its stress to the thematic vowel (see below). The same contrast between enclinomenon lsg. and ending-stress in the rest of the paradigm also survives in Middle Bulgarian manuscripts from Veliko Tarnovo, e.g. rékg, rekg ¡ī, prinesg, prinesqti (< *prines9 ti with retraction from final syllable, frequent
188
in eastern MidBulg.), bláiq, úblaiq, blaiq ¿e vs. (iz)reðétv, (na)reðémú, prinesémú, prines&, blaiqt m$ (Hinrichs 1985:86ff., Dybo et al. 1990:231-6). An alternation between lsg. and non-lsg. forms in the simple thematic present survives in at least three (or more precisely, four) dialect areas in the contemporary Slavicspeaking world. In most modern Bulgarian dialects and in the standard language, the inherited alternation (still attested in Middle Bulgarian, cf. above) has been eliminated in favor of a uniform stress on the thematic vowel, as in Russian. Compare the following Bulgarian paradigms with their Russian cognates:
lsg. 2 3 lpl. 2 3
Bulgarian6 'bring' do-nes-á do-nes-é-š do-nes-é do-nes-é-m do-nes-é-te do-nes-á-t
Bulgarian 'be gathering' ber-á ber-é-š ber-é ber-é-m ber-é-te ber-á-t
Russian 'be carrying' nes-ú nes-é-š' nes-ē-t nes-é-m nes-é-te nes-ú-t
Russian 'be taking' ber-ú ber-é-š' ber-ē-t ber-é-m ber-é-te ber-ú-t
In contrast, the dialects of the central Rhodope mountain range in south-central Bulgaria, maintain an accentual alternation in the present. In these dialects, which have preserved numerous archaisms in nominal inflection (cf. Stojkov 1968:87-91, Scatton 1993:245), one finds paradigms such as plét-q {-am), plet-éš, plet-é, plet-ém, plet-éte, plet-át, -it 'weave' (MiletiC 1903:262, cf. Stang 1957:111, 1966:450-1). Farther west in the Balkan peninsula, certain Torlak dialects of southeastern Yugoslavia, along with adjoining South Slavic dialects in northeastern F. Y.R. of Macedonia and western Bulgaria, attest the same contrast between lsg. and other forms. Consider the following data from Alexander (1975:307-23): southeastern Serbia
lsg. iz-metu
non-lsg. 3sg. iz-meté lpl. iz-métemo
ná-beru
northwestern Bulgaria
6
'sweep out, shit
drl-u zá-boravu
lpl. z-béremo 2pl. dri-íte 2sg. za-boráviš
'wrinkle, pleat' 'gather, meet' 'hold' 'forget'
zá-pleta prí-stana pó-seda
2sg. za-pletéš 3pl. pri-stánat lpl. po-sedímo
'entangle' 'agree, join' 'sit a while'
The lsg. and 3pl. endings are phonetically [-a] and [-at].
189
central western Bulgaria
dó-nesa péra zá-kolja
2sg. do-neséš 2sg. peréš lpl. da za-kóljem
'bring' 'wash' 'slaughter'
northeastern Macedonia
kóva zá-kova
3sg. kové 3sg. za-kóve
'forge' 'fasten, nail'
Finally, the Slovincian forms cited in Lorentz (1903:210) prove that the alternation of the lsg. with other forms in the present also survived in West Slavic. Here, as also to a large extent in the noun, Slovincian has generalized the stress alternation originally characteristic of unaccented stems, i.e. of accentual paradigm c. Consequently, all verbs contrast default initial stress in the lsg. with penultimate stress in the rest of the present; the latter results from the regular retraction of stress from final syllables (cf. Stang 1957:111, Garde 1976a: 113-4, 291-2, Stankiewicz 1993:313, 324). The following forms are taken from Garde's simplified version of Lorentz's phonetic transcription: lsg. 2 3 lpl. 2 3
/do-nes-/ 'bring' duonosq domeseš domesd domesemd domeseca doáúosoiļa
/ne klopoc-/ 'not to trouble' meklopocq ne klopuoceš ne klopúocā ne klopuocema ne klopuoceca ne klopiiocoy
As noted above, the Old Russian and Middle Bulgarian data demonstrate that the lsg. was underlyingly unaccented, so that the stress could surface on the first syllable of a larger accentual unit, in particular ne 'not'. This default initial stress has been reinterpreted as word-initial stress in those modern dialects which preserve the alternation in the present; furthermore, the alternation has been extended to all verbs regardless of their original accentual properties, so that the synchronic rule in these dialects is now "initial stress on the lsg. regardless of the position of stress in non-lsg. forms." As a result, one encounters not only the inherited alternation of initial stress in lsg. vs. word-final stress in non-lsg., but also a secondary, "non-peripheral" alternation between initial stress in 1 sg. and stem stress in the other forms of originally accented present stems, e.g. Torlak zá-boravu, zaboráviš 'forget' ("non-peripheral" alternation, cf. Alexander 1975:X, X). Notwithstanding this post-PSl. innovation, the fact remains that reflexes of unaccented lsg. vs. word-final stress in non-lsg. are found in all three branches of Slavic: East (Old Russian), West
190
(Slovincian), and South (Middle Bulgarian, the Rhodope dialects of Bulgarian, Torlak and adjoining dialects of Serbia, Macedonia, and Bulgaria), so that the alternation may safely be reconstructed for Proto-Slavic. How exactly should the alternation in question to be reconstructed? At first it might seem that the non-lsg. forms were stressed on the thematic vowel (3pl. *--, elsewhere *e-; see §4.2.2, fn. 14), as in the contemporary Rhodope, Torlak, or Slovene dialects described above, or in Slovincian. In the 2pl., however, Old Russian has forms with stressed ending -té (see below). Although instances of -éte, -íte are found at an early date, these may easily be explained as analogical to the theme stress of the 2sg., 3sg., lpl., and 3pl., which were analyzed — or rather reanalyzed, cf. immediately below — as containing underlyingly accented theme vowel + ending. This shift of stress from ending to thematic vowel was a gradual change over several centuries, and did not reach all areas of East Slavic: ending-stressed forms occur in accented manuscripts through the 17th century, e.g. deríité 'you are holding' (Korm. 1650), prineseté 'you will bring forth' (Triod' 1621), and survive to the present day in peripheral northern dialects, e.g. Tot'ma paset'ó (Stang 1957:109, 184fn.90, 1966:451, cf. Lehfeldt 1993:85-6; on Ukr., Bel. -té see below). Once the underlying accentuation of the thematic vowel was extended to the lsg., the result was the columnar thematic stress of modern Russian. Given that the oldest accentuation of the ORuss. 2pl. is ending-stressed -té, one may speculate whether the other four trisyllabic person-number forms, i.e. the 2 and 3sg. and 1 and 3pl., were originally likewise stressed on the ending. In the case of the 3sg. and 3pl., retraction from word-final *i (or *u) and the subsequent loss of weak jers would have produced the attested forms, with stress on the thematic vowel {-ét, -út, modern -ēt, -út). The same would have occurred in the 2sg., which like the thematic 2sg. in all other Slavic languages continues a preform *-eíí rather than OCS -eSi, and in the lpl., for which (Old) Russian -m reflects an ending *-mu (like OCS -mu; cf. Bulg. -m in first- and secondconjugation verbs, i.e. verbs with stem vowel -e- and -/-, e.g. do-nesém 'we bring'. stignem 'we reach', uCim 'we are learning', dariim 'we are holding'). Thus the earliest accentuation of the simple thematic present recoverable by internal reconstruction for pre-ORuss. or Proto-East Slavic appears to have been as follows 1 2 3
sg. péku < *pekQ (unacc.) peðéš < *peCeš1 peðét < *peðeti
pi. peðém < *peCemu peíéte <— peðeté pekút < *peknti
191
In fact, since word-final jers were lost at early date, this system should in fact be projected back to the Proto-Slavic stage. Both direct and indirect evidence from the other Slavic languages confirm the reconstruction of word-final stress outside the lsg. First of all, Ukrainian (and Belarusian) and Serbo-Croatian actually preserve stressed endings in the 1 and 2pl.: cf. Ukr. uōimó, uðité 'we, you (pi.) teach', eakavian pe6e-mo, -re, ¿elī-mo, -re 'we, you (pi.) wish', whence standard SC peðémo, peCete with the usual štokavski shift, i.e. high-tone retraction, resulting in rising pitch on the preceding theme vowel (§4.2.1; on the long thematic vowel cf. fn. 7). In these languages, the lpl. continues a variant ending *-mo which, unlike *-mú, was not affected by retraction and jer loss, so that stress could be — and was in fact — maintained on the ending, as in 2pl. *-te. Indirect support for original end-stress in the non-lsg. forms comes from the ðakavian dialects of Serbo-Croatian, as well as from Slovak. Like the rest of Western South Slavic, Cakavian has generalized originally athematic *-m in the lsg.; this then shifted to -n, like other paradigmatically isolated examples of word-final *-m, e.g. o-stem instr. sg. -on (whence a-stem -un for *-ú < PS1.
*-OJQ),
dat. pi. -ōn, a-stem dat. pl. -ān.
Consider the paradigm of 'cook' in the dialect of Nov! Vinodolski, described by Belie (1909:239-41): lsg. 2
3 lpl. 2 3
peCen peðéš peíé peðemó peCete pekú
< *peðém
< < < < < <
*peðēmí *peðeSí *pecetí *peðemo *peðeté *pekņtí
As Tedesco (1951:77) has pointed out, the rising accent on the lsg. can only reflect an original preform *peCēmi, parallel to 2sg. *peðešl and 3sg. *peðetí (cf. athematic lsg. dán 'I give' < *dami, §6.5)7 This hypothesis will of course also account for the štokavian 7
Tedesco's statement of the changes involved is crucial for the vexed question of "neoacute" on short vowels: unless one wishes to assume that stressed word-final jer in the 2. 3sg. (and 3pl.) survived until the remodeling of the lsg., one must operate with a short rising (i.e. "neo-acute") thematic *-ē- in 2sg. *peCēS, 3sg. *peðē(t); this vowel was introduced into lsg. *pec"ém (> *-én) and lengthened before word-final sonorant to produce long rising *é. Subsequently short rising vowels fell together with short falling in pedes, peCe. while the lengthened é in lsg. peCen retained its rising intonation in Cakavian. For other examples of "neo-acute" on PS1. short vowels, cf. §4.2.2 and see below on Slovak.
192
paradigm, e.g. péðēm,péðēš,péðē,peðémo,peðéte,péku,
i.e. /peðēm/, /peð'ēš/, /peðē/,
/peeēmo/, /peðēfe/, /peku/ (with lengthened thematic vowel extended throughout, cf. Resetar 1937:52-3). In a number of stokavian dialects and the standard language, the (underlying) stress in the 1 and 2pl. is retracted to the thematic vowel if the root contains a long vowel, producing columnar stress in e.g. trésēm, trésēš, trésē, trésēmo, trésēte, írésv 'I am shaking, you are shaking, etc.', but this is clearly a secondary development; for a survey of dialectal variation in length of the thematic vowel -e- and the stress of the 1 and 2pl., and the relationship between the two variables, cf. Stankiewicz 1979b. Finally, the Slovak paradigm nesiem, nesieS, nesie, in which ie continues *e with "neo-acute" intonation, must reflect an earlier *neses1, *nesetu (*-ti), etc. Stang (1952a:274) observes that "das Slowakische dort -em, -eš aufweist, wo das Prasens im Urslavischen entweder konstante Wurzelbetonung oder Wurzelbetonung mit Neuakut von der 2. P. Sg. an hatte. Wo man im Urslavischen konstante Endbetonung hatte, finden wir im Slowakischen -iem, -ieš...". In the terminology of Stang (1957:107ff.), e-presents with -ie- for the PS1. thematic vowel *e in Slovak correspond to verbs of AP c in other Slavic languages, e.g. 2sg. vedieš, pasieš (cf. Russ. vedú, vedēš', pasú, pasēš'), whereas epresents with short thematic -e- correspond to verbs of AP a or b, e.g. budeš, pláðeš (cf. Russ. búdu, búdeš', pláðu, places', AP a), moieš, káieš (Russ. mogú, moíeš', kaiú, kázeš', AP b)β Although the distribution uncovered by Stang is clear, the historical interpretation of length or shortness in the thematic vowel *-e- requires some revision. Since verbs of AP a and b were stressed on the root and the thematic vowel, respectively (with the non-lsg. forms of AP b undergoing retraction by Stang's Law, producing neoacute intonation on a long root vowel), thematic *-e- would have developed regularly to -e-. Simple thematic presents of AP c, in contrast, must have stressed the ending, which was a jer in the 2, 3sg. and 3pl. (and, after the replacement of PS1. *-q by *-e-mI, the lsg. as well): the stress was Note that invoking analogy to presents with the long stem vowels *-á- (< PS1. *-á-je-) and *-ito avoid the need for "'neo-acute" on short *-e- is unlikely: if the thematic vowel *-e- in 2. 3sg. and 1. 2pl. (and lsg. *-em > ðakavian -én) was lengthened, why would it have been subsequently shortened? Indeed, although -e- has been lengthened in most modern Serbo-Croatian dialects and the standard language (undoubtedly by analogy to a- and /'-presents and to 3pl. -u), this appears to be a relatively late innovation, as short -e- survives in a range of peripheral Ca- and Stokavian dialects and is attested in Vuk Karadzic's grammar (Stankiewicz 1979b, ReSetar 1937:52-3). 8 This distribution, as we shall see, almost exactly coincides with the reconstructed accentual paradigms of *-e/0-, *-ye/0-, and *-ne/0- presents in Proto-Slavic (cf. §6.6): with several exceptions (e.g. budeS, mðieš), *-e/0- presents are of AP c, whereas the two suffixed thematic presents belong without exception to AP a and b.
193
retracted to the thematic vowel, and the resulting "neo-acute" *-e- in Slovak yielded long *-ē- and eventually modern -ie- (cf. fn. 7 on neo-acute short *-e- in the prehistory of Cakavian). The combined evidence of Slavic thus leads to the conclusion that the Proto-Slavic simple thematic present was marked by an alternation between an underlyingly unaccented lsg. (with default initial stress) and stress on the ending, i.e. the syllable after the thematic vowel, in all eight other person-number forms: lsg. *pékq (*né pekņ) vs. *pe£eši (?), *peðetí, *peeevé, *peCetá, *peCeté, peðemu (*-mo), *peCeté, *pekQti. 6.3.2. A postaccenting thematic vowel in Slavic? Fortunately, the unaccented lsg. and end-stressed non-lsg. forms of the simple thematic present may be reconciled as synchronically varying surface realizations of a single stem. In order to understand this, however, we must introduce the concept of stray erasure. Halle (1997:285-6) has proposed that the irregular initial stress in /-stem cardinal numerals such as Russ. désjat' '10' may be explained as the result of a lexically specified rule by which any underlying accentual specification, i.e. left or right bracket on line 0 of the metrical grid, which encloses no syllabic heads is automatically erased. Whereas a left bracket before a word-final jer is retracted (and the jer deleted) in normal postaccenting stems, e.g. masc. karandáš, gen. karandaš-á 'pencil', fern, ljubóv', gen./dat./loc. ljubv-i Move', or boród, gen. pi. of borod-á 'beard', stems subject to Bracket Erasure are underlyingly unaccented and hence surface with default initial stress: *
*
*
*
*
(* (*)
ka ran da šá
*
*
*
*
(*)
ka ran da SE
—>
ka ran dáš
*
*
*(
bO Iju > ljubví
but
(* /*\
*
*
(* *)
*
*(
*)
vi
lju
bO
vE
—>
*
*
(* (*)
lju
bóv'
*
*
*(
(*) *)
*
*(
*)
de
sja
ti
de
sja
tE
194
—>
(* *
*)
dé
sjat'
However, it is less than clear that this handful of nominal stems, comprising the /'-stem numerals 5-10, need be explained by recourse to such a lexically restricted synchronic rule. As argued in §5.1.4, fn. 22, it is more likely either these are historical relics of accented /'stem sg. gen. -(i, dat. -(i, or else that originally unaccented '5', '6', '9', and '10' and postaccenting '7' and '8' have mutually influenced each other, resulting in an accentual pattern peculiar to those numerals. Fortunately, the accentual alternation of the PS1. simple thematic present provides a clear and striking confirmation of Halle's stray bracket erasure rule. Note that the lsg. differs in length from the other person-number forms: the former is disyllabic, whereas the latter consist of three syllables. In other words, the combination of thematic vowel and ending proper, e.g. *-e-vē, *-e-te, *-q-ti, consists of two syllables for all person-number forms except lsg. *-Q, owing to a pre-PSl. sound change (cf. §6.3.3). As the non-lsg. forms are end-stressed in PS1., it is reasonable to assume that the thematic vowel is underlyingly postaccenting, i.e. *-e(- ~ *-o(-. The metrical grids for 2pl. *peCete and 3pl. *pekQti are hence
*
*r
* r* *\
*pe
Ce
te
*
*r
* r* *\
*pe
ko.
ti
What of lsg. *pekQ? If Halle's rule holds, the left bracket after the thematic vowel, i.e. the second element on line 0, should be erased, as there are no heads to its right for it to enclose. As the only remaining bracket is the default line 0 RRR edge marker, the lsg. is now unaccented and surfaces with default initial stress in the phonological word, as provided by the head-marking rules of the grammar. *
*
*
*r\
*
*
*/\
*pe
kQ
*ne
pe
ko.
Thus once stray erasure is taken into consideration, the normal rules of PS1. prosodic computation yield precisely the reconstructible contrast between default initial stress in the unaccented lsg. and postthematic stress in the rest of the paradigm. This analysis will be modified and expanded somewhat as a result of the historical considerations in §6.5, but the main point holds: the lsg., being shorter by one syllable, lacks a postthematic syllable 195
head and so undergoes Bracket Erasure, resulting in an "enclinomenon" which contrasts with the uniform ending-stress of the remainder of the simple thematic present. 6.3.3. The evolution of the alternation in Proto-Slavic and the Slavic languages: evidence for the brackets-and-edges model We now turn to the question of how such an alternation arose within the prehistory of Slavic. The PS1. lsg. ending *-Q clearly cannot continue (Brugmannian) PIE *-o-h2 > *-ō, reconstructive on the basis of GAv. -ā, Gr. -o:, Lat. -ō, PGmc. bimoraic *-ō (> Goth. -a, OHG, Anglian OE -u), Olr. «-umlaut (e.g. biur< *beru < *berō), and of course Lith. -u < *-úo < *-ó: (cf. reflexive -úo-si). Instead, it appears to have been remodeled through addition of the lsg. athematic lsg. primary ending *-mi, with apocope of *-ōmi to *-ōm and thence to *-Q; cf. Vaillant 1966:8. Such an extended thematic primary ending finds exact parallels in a number of other IE languages: cf. Sanskrit -ami, Av. -āimi (replacing vestigial GAv. -a), OP -ami, Olr. marbaimm < *-āmi, OHG -ēm, etc. Assuming the remaining person-number endings to have been sg. 2 *-eš-aí,9 3 *eti, pi. 1 *-e-mos (<— PIE *-o-mos), 2 *-e-te, 3 *-o-nti, as generally accepted (and leaving aside the dual endings, whose PIE reconstruction is problematic), it follows that as long as the lsg. remained disyllabic *-ō-mi in pre-PSl., the paradigm of the simple thematic present would have surfaced with columnar stress: if the thematic vowel was 9
I consider this the most likely preform of the attested OCS 2sg. ending -eSi, with its hitherto mysterious final -i for the -i7 (ORuss. -J) of the 3sg. and 3pl. which one might expect as the regular outcome of PIE thematic 2sg. *-e-si. Judging from Lith. -/, refl. -te-s(i), Latv. -i, -ii-s, the BS1. thematic 2sg. ending was acute *-ai. The source which immediately comes to mind is the PIE them, opt. *-o-yhļ-s, with acute intonation resulting from the laryngeal *hļ; so far as I know, this is the only direct evidence for *-oyh|- (> *-oy- with loss of laryngeal, Gr. -oi-, Skt. -e-, Av. -a€-, -ōi-), generally assumed on the basis of internal reconstruction. For a close parallel, cf. PWGmc. pret. 2sg. *-ai from the optative vs. PGmc. *-t < PIE *-th2e, e.g. OHG wurti, nāmi, OE wurde, name vs. Goth, warst, namt, ON varzt,namt (cf. Streitberg 1896:324-6, with unlikely alternative). The replacement of PIE thematic 2sg. *-e-si with opt. *-ais, and then of athematic *-si with *sai, was an important shared innovation of Baltic and Slavic. In Lith., athem. lsg. *-mi has also been replaced by *-mai, cf. OLith. demi, refl. demies (see below, §6.5). Slavic has remodeled PBS1. them. 2sg. *-i to *-e-si by analogy to the athematic conjugation (athem. 3sg. *-tī : 2sg. *-si :: them. 3sg. *-etl : 2sg. X, X = *-e-si), as has Old Prussian (glwasi 'you live', cf. §6.5); this then became *-e-Si, with *-S- generalized from ¿-presents (e.g. nosisi, letiSi, assuming that the ruki-rulc was still active when the suffix *-i- came into being, whatever its prehistory). The -S of medieval and modern Slavic languages results from a post-OCS but apparently common Slavic replacement of *-eSi with *-eš(ī), which brought the 2sg. back into line with the 3sg. and 3pl. (and athem. lsg. -m) but left unaffected athem. *-si (> Ukr. es?, dasy). So far as I can see the only obstacle to the hypothesis proposed here — and it may be insurmoutable — is the absence of -s in Lith. -ie- < *-o-yh|-s. In any case, I see no need to postulate aberrant PIE preforms for the thematic 2sg. ending on the basis of BS1., contra Vaillant 1966:9-10. Watkins 1969:210-4, Beekes 1995:233-4.
196
postaccenting at this stage, as reconstructed above for PS1., all person-number forms would have exhibited word-final stress. Subsequently, lsg. *-ōmi underwent apocope, in line with an early pre-PSl. Auslautgesetz by which word-final *-V:Ni > *-V:N, where N stands for a nasal,10 and the resulting ending *-ōm underwent nasalization and loss of the coda nasal consonant to yield PS1. *-q (OCS -g). The effect of apocope in the lsg. was to eliminate *i as a syllable head or potential "stress-bearing unit" and reduce the 1 sg. to two syllables. As a result, the postaccenting thematic vowel would have undergone stray erasure in the computation of surface stress, yielding the unaccented lsg. and default initial surface stress reconstructed for PS1.:
*pe
kō
*
*
mi
*pe
kōm
To be sure, the apocope of *-i did not happen immediately, but was a gradual process; as the study of sound change in progress has revealed (references), there must have been a period of variation during which both *-ōmi and *-ōm could be heard. What is significant here is that the apocope of *-i must have been accompanied by the elimination of the postaccentuation of the thematic vowel, so that the competing forms were *pekōmi, *ne pekōmi and unaccented *pekōm, *né pekōm; the apocopated variants eventually prevailed, yielding PS1. *pékq, *né pekQ. Interestingly, the final *-m of *-ōmi > *-ōm which was lost in PS1. *-Q was later restored in many Slavic languages, although in different ways, at different times, and with various results.1 • In Serbo-Croatian, which of course has maintained lexical stress to the present-day, athematic lsg. *-mI has spread to the lsg. of all verbs except for mógu 'I can' and hócu 'I want'. What is more, the spread of *-m has been accompanied by the 10
The only other clear example of this sound change is the instr. sg. of d-stems (PS1. *-OJQ < pron. *-ay-āmi <— *-āmi < post-PIE *-eh2-mi(n); instances of -o in OCS are secondary, due to contraction, van Wijk 1931:179-80, Tedesco 1951:173) and fem. /-stems (PSI. *-ījo. < *-iy-āmi, i.e. *i(y)- + original d-stem ending?). The same apocope has occurred (independently?) in Lithuanian dstem instr. sg. -a, -d- (e.g. gálva to galvā 'head', gēra, ger ORuss. -eti) - *-et (> OCS -etv, modern SC, Bulg., Pol., Cz. -e, Slk. -e, -ie), *-qti (> ORuss. -KΓÍ) - *-Qt (> OCS -
197
restoration of postthematic stress, as the following ðakavian and Stokavian paradigms illustrate (cf. §6.3.1 above on Cakavian): lsg. 2 3
Cakavian (Novi) peCen peðéš peðé
stokavian (standard) péeēm péðēš péðē
lpl. 2 3
peðemo peðeté pekú
peðémo peðéte pékv
Clearly, the 1 sg. has been formed on the model of the other person-number forms, as thematic vowel + lsg. ending -m (underlyingly /-ma/). For Stokavian, one could assume that the 2 and 3sg. were analyzed as containing accented theme vowel -e- + ending -S, -0 already in early SC, following retraction from the weak word-final jer and loss of the latter; similar 3pl. -u- + -0. This accented theme vowel was then extended to the 1 sg., resulting in the attested accentuation of péðēm, etc. As Tedesco has correctly observed, however, the evidence of Cakavian indicates that, at least in this archaic dialect, 2 and 3sg. -eš and -e were still underlyingly /-e-(Sa/, /-e-(0/ at the time when the lsg. was remade to *-em (underlyingly */-ema/). The latter became -ēn by lengthening of a vowel before a sonorant and the change of *-m > *-n (cf. above); the rising accent on -én reveals that this ending historically continues *-em(a — or *-em(i, if the front and back jers had not yet merged — and synchronically may continue to be analyzed as -e-(na, as 2sg. -éš, 3sg. -e, 3pl. -ú may also be analyzed as -e-(Sa, -e-(a, -tā-(a.12 The various remodelings of the lsg. ending in pre-Proto-Slavic and in individual Slavic languages demonstrate the linguistic reality of the "brackets" in the brackets-andedges framework, in this case the postaccentuation of the thematic vowel. If the derivation of PS1. present lsg. *-q proposed here is correct, the accentual alternation between the lsg. and other person-number forms must have been an innovation of Slavic, a direct result of early pre-PSl. apocope of *-i in lsg. *-ō-mi. More importantly, the underlyingly unaccented nature of PS1. pres. lsg. forms in *-q provides an important confirmation of the linguistic reality of stray bracket erasure as a productive rule, and of the brackets-andcourse, since lpl. -ēmo and 2pl. -ete survives in Stokavian, it is possible to treat -ém, -eš, -ē, -ua as synchronically underlying -ē-(ma, -ē-(5s, -ē-(a, -u-(a as well. The ie of Slovak nes-iem and other continuations of PSI. presents of AP c, on the other hand, need not be due to restoration of postthematic stress: even if initial stress had already been generalized, lsg. -iem could have been formed on the analogy of 2sg. -ieS, 3sg. -ie, etc.
198
edges framework of stress computation in general. Subsequently, when *-q or its reflexes is replaced by the theme vowel + athematic ending *-mi, e.g. in SC -ēm (underlyingly /-ēma/), the postaccentuation of the theme vowel appears once again in the lsg., e.g. Cakavian -én, underlying -e-(ns. Far from being merely a notational abstraction, then, the left bracket after *-e/0- in Slavic is very much a part of the underlying representation of this morpheme, and its effects surface clearly in the results of both sound change and analogy. 6.4. Baltic evidence for the alternation? As observed at the beginning of §6.3, once pre-PSl. had remade the lsg. ending to *-o:mi, the paradigm of the simple thematic present would have had columnar stress on the final syllable. This innovation, however, is specific to Slavic and cannot be of ProtoBalto-Slavic date, for Baltic has preserved the inherited PIE thematic ending *-oh2 > PBS1. *-ó: (> *-úo > Lith. -u, -úo-s(i), Latv. -u, -uds). But what of the postaccenting thematic vowel reconstructible for the other person and number forms in PS1.? Do the Baltic languages offer any support for a similar accentual pattern, allowing one to project the alternation described in §6.2 back to Proto-Balto-Slavic? At least in East Baltic, the answer is no, if one considers the finite forms of the thematic present. In Lithuanian, where stress alternations have been preserved, the root may be either accented or unaccented. In the former case, the stress is columnar on the root vowel if it is acute; a circumflex or short root vowel loses stress to the underlyingly acute 1 and 2sg. endings (i.e. lsg. -u l-nol, 2sg. -i < l-itf) by de Saussure's Law. These two cases are thus parallel to APs 1 and 2 in nominal inflection (examples from Garde 1976:106-9): AP2: acc. non-acute (rékpa-rēk-ti rēk-ti 'scream once' 'scream'
AP 1: ace. acute (sésé-ti 'semer' ap-sé-ti 'sow over' 'sow' séj-u séj-i séj-a
ap-séj-u ap-séj-i ap-séj-a
rēki-a
pa-réki-ú pa-rék-l pa-réki-a
séj-ame séj-ate
ap-séj-ame ap-séj-ate
réki-ame rēki-ate
pa-réki-ame pa-rēki-ate
réki-u rék-l
In the presents of unaccented roots, stress falls on the last preverb (including the negation marker ne), from which it advances to an acute root syllable by de Saussure's Law, 199
resulting in columnar root stress. As a result, the verbal equivalents of APs 1 and 3 have fallen together in Lithuanian, but the distinction between accented and unaccented acute roots is preserved in the intonations of Latvian, e.g. sēt 'semer', with rising or sustained pitch, vs. graúzt, with "broken" intonation. Stress on a preverb survives in the presents of unaccented non-acute roots, such as at-néšti. AP 3: unacc. acute gráufgráui-ti iš-gráui-ti 'gnaw' 'gnaw out/away'
AP4: unacc. non-acute neš néš-ti at-néš-ti 'bring' 'carry'
lsg. 2 3sg./pl.
gráuí-u gráui-i gráui-a
iš-gráu¿-u ií-gráut-i iS-gráu¿-a
neš-i
nēš-a
át-neš-u át-neš-i át-neš-a
lpl. 2
gráui-ame gráui-ate
iš-gráui-ame iš-gráui-ate
nēš-ame nēš-ate
át-neš-ame át-neš-ate
neš-u
The contrast between accented and unaccented roots also surfaces in the present participle, where the conservative dialects exhibit the following mobile paradigm for unaccented nonacute roots such as néš-ti 'carry', prefixed nu-néš-ti 'carry away' : 13 nom. sg. gen. dat. ace. instr. loc.
masc. neš-q~s neS-anCi-o neS-an Si-am nēš-ant-ļ nēS-an£i-u neš-anði-amé
fem. neS-ant-i neš-anei-ōs nēS-an£i-ai nēš-anei-ņ nēš-anði-a neS-an£i-ojé
fem. masc. nu-neš-ant-1 nu-neš-$s nú-neš-an£i-o nu-neš-an£i-ōs nu-neš-an£i-ám nu-neS-an£i-ai nu-neš-an£i-g nú-neš-ant-ļ nu-neš-an£i-u nu-neS-an£i-a nu-neš-anði-amé nu-neS-anCi-oje
nom. pi. gen. dat. ace. instr. loc.
neš-q~ neš-an£i-y neš-ant-íems nēS-an£i-us neš-anði-aīs neS-an£i-uosé
nēS-anði-os neš-anCi-iļ neš-an£i-óms nē$-an£i-as neS-an£i-om¡s neS-an£i-osé
nu-neš-an£i-os nu-neš-4 nu-neš-an£i-iļ nu-neš-an£i-Q nu-neš-ant-íems nu-neS-an£i-óms nu-neS-an£i-us nú-neš-an£i-as nu-neš-an£i-aīs nu-neS-an£i-omís nu-neS-an£i-uosé nu-neš-an£i-osé
For further discussion of the accentuation of the present, and of the Lithuanian verb in general, cf. Senn 1966a:223-4,246-51, Garde 1976a:106-9,118-48, Lehfeldt 1993:115-23. As far as the accentual specification of endings, however, Lithuanian offers no equivalent to the contrast of unaccented lsg. and end-stressed non-lsg. reconstructed above 13
The standard language has generalized recessive stress in all forms except for masc. nom. sg. neš-<ļs, nom. pi. neS-q, the two forms in which the suffix -ant- and the case-number ending have fused together; cf. Senn 1966a: 172ff. See Cowgill 1970 for the possibility that the nom. pi. is derived from the PBS1. 3pl. ending *-ant < PIE *-onti.
200
for Slavic. Indeed, all the person-number endings in the paradigms above are underlyingly unaccented, the stress of vedu < *veduo, vedi < *vedie vs. vēda, vēdame, vēdate being of course due to de Saussure's and Leskien's Laws. It appears that in pre-Lithuanian, and probably in Proto-East-Baltic, the thematic vowel was unaccented, so that the presents of unaccented roots contained only "enclinomena" with recessive, default initial stress.14 Old Prussian, on the other hand, appears to have preserved a different and more archaic state of affairs in the Baltic verbal endings. In order to appreciate the signifcance of the OPr. data, let us consider the possible origins of the postaccenting thematic vowel reconstructed in §6.3 for the simple thematic present in Proto-Slavic. 6.5. The PIE origins of the Balto-Slavic postaccenting thematic vowel: athematic verbs and primary endings It was observed in §6.1 that Indo-Aryan and pre-Proto-Germanic agree in having columnar root stress in the simple thematic present: this is directly attested for Sanskrit, and may confidently be inferred for Germanic by the effects of Verner's Law. On the basis of those two branches, it has been universally assumed that simple thematic presents in (post-)PIE had columnar stress on the root. Yet Slavic and probably also Old Prussian provide evidence for an entirely different pattern of alternation, one which, as shown in §6.3.2-3, follows from the assumption of a postaccenting stem vowel *-e/0(-- Is the Sanskrit-Germanic columnarity original, and the BS1. alternation an innovation? Or might Balto-Slavic here preserve an archaism of (post-)PIE date, with Indo-Aryan on the one hand, and Germanic on the other, eliminating the alternation in favor of regularized, columnar stress? Rather than adopt an a priori stance on this question, let us instead turn to internal reconstruction of the present verbal endings of PIE. I shall focus on the sg. and 3pl., as the relationship between primary and secondary endings in the 1,2pl. remains unclear, and the reconstruction of the dual endings is generally uncertain. In §6.3.2, it was assumed that the BS1. accentual pattern in simple thematic presents was to be analyzed as the realization of an underlyingly postaccenting thematic vowel. This 14
I assume, following references, that the -a- of 3sg. -a and 1 -ante, 2pl. -ate, as well as the archaic and dialectal 1, 2du., continues *-o-, leveled from *-e/0-. I cannot accept the more radical arguments of Watkins (1969:2lOff. and passim, 1970), which presuppose a wholesale revision of the traditionally reconstructed person-number endings of the thematic conjugation in PIE: other than lsg. *-ō < *-oh2 < *-o-h2e, i.e. thematic *-o- + "protomiddle" *-h2e (> perfect *-h2e, mediopassive *-h2e-r; cf. Jasanoff 1997b), the thematic endings were simply composed of the thematic vowel and the athematic endings, cf. Cowgill 1985.
201
is not the only possible interpretation, however; a linguist, or a speaker of PBS1. or prePS1., could easily deduce that the primary endings of the 2, 3sg. and 3pl. are underlyingly accented. *pekw-ō
*pekw-e(-si *pekw-e(-ti *pekw-o(-nti
*pekw-e-(si *pekw-e-(ti *pekw-o-(nti
OR
Such an analysis is especially appealing when one takes into consideration those (post-)PIE categories in which thematic stems combined with secondary endings, e.g. the imperfect or thematized aorist.15 In these paradigms, the thematic vowel was not followed by another syllable in the sg. and 3pl. Thus there would have been no basis in these forms for speakers to abstract a postaccenting thematic vowel, as Stray Bracket Erasure (§§6.3.2-3) would have produced an underlyingly unaccented form. *pekw-ō
*pekw-e(-si *pekw-e(-ti *pekw-o(-nti
*pekw-e-(si *pekw-e-(ti *pekw-o-(nti
OR
To be sure, a postaccenting thematic vowel would be "recoverable" for speakers in the ending-stressed forms of the 1 and 2pl. (cf. above). More likely, however, speakers of (post-)PIE would conclude from the above paradigms that the primary endings are underlyingly accented, but not the secondary — in other words, that the "hic-et-nunc" particle *-i, which of course distinguishes the former from the latter, was accented.16 The sg. and pi. forms of the simple thematic present and imperfect would then be accented as follows: *pekw-ō
*pekw-e(-si *pekw-e(-ti *pekw-o(-nti
*pekw-e-(si *pekw-e-(ti *pekw-o-(nti
OR
15
The thematic aorist appears to be an entirely post-PIE creation of the individual IE languages: cf. Cardona 1960 and §2.3.1.1. 16 If this is the same particle which was added to originally endingless PIE locatives, it follows that archaic locatives in *-i with full-grade of the suffix such as Ved. murdháni 'on the head' (<—murdhán) were underlyingly *murdh-(an-(i, i.e. *R-(S'-(i; this was later replaced by murdhni. i.e. underlying múrdh-an-(i, on the model of genVabl. murdh-an-(as, dat. múrdh-an-(e, etc. See §2.2.1.3 for further discussion.
202
Unfortunately, the reconstruction of underlying accentuation in *-s-i, *-t-i, and *-nti, and presumably also in the other person-number primary endings, encounters difficulties elsewhere in the system of verbal accent reconstructed for PIE. If the primary endings were uniformly accented, (post-)PIE would have had columnar ending-stress not only in thematic presents, but also in athematic presents, e.g. root presents such as *hiey- ~ *hii'go' and *g wh en- ~ *g wh n- 'strike'. This is decidedly not the ablaut-accent pattern reconstructed for PIE (cf. §2.3.1.1), which has been continued so well in Hittite and Old Indo-Iranian and, as far as ablaut alternations are concerned, survives in Greek, with relics (mostly of *hļes- ~ *hļS- 'be') attested elsewhere. In PIE root presents, the root-stress and full-grade root ablaut of the sg. contrasts with ending-stress and zero-grade root ablaut in the du./pl.:17 1 2 3
sg. *hjey-mi *hjéy-si *hjéy-ti
du. *hļi-wós *h]i-tés(?) *hii-tés (?)
pi. *hi¡-mós *hļi-té ^y-énti
1 2 3
*gwhén-mi *gwhén-si *gwhén-ti
*gwhņ-wós *gwhņ-tés(?) *gwhņ-tés(?)
*g wh ņ-mós *gwhņ-té *gwhn-énti
Since there is no a priori reason to suppose that the accentual properties of the primary endings (or any other morphemes!) originally varied depending on the morphology of the stem to which they were attached, let us take the well-established accentual alternation of the PIE athematic presents as a starting point for the reconstruction of the accentuation of the simple thematic present. Such a procedure is also historically justified, in that the athematic presents, being well attested in Hittite (and Tocharian), may unquestionably be reconstructed for "real" PIE, whereas simple thematic presents are largely an innovation of the "Brugmannian" IE branches; cf. §6.1. From the paradigms given above, it is obvious that the sg. endings, both primary *mi, *-si, *-ti and secondary *-m, *-s, *-t, are unaccented, and that the (du./)pl. endings, primary *-mos, *-te (?), *-enti as well as secondary *-me, *-te, *-ent, are accented. Assuming the same accentual specifications for the thematic present, the post-PIE paradigm for e.g. *pekw- 'cook' would likewise have contrasted unaccented sg. with ending-stressed du./pl.: 7
On Narten presents to underlyingly accented roots, cf. §2.3.1.2. 203
unaccented
lsg. 2 3
*pékw-oh2 *pékw-e-si *pékw-e-ti
vs. ending-stressed
lpl. 2 3
*pekw-o-(mós *pekw-e-(té *pekw-o-(ntí
I propose that both the Sanskrit-Germanic pattern of columnar root stress and the BaltoSlavic alternation reconstructed in §§6.3-4 may be derived from an ancestral paradigm of this type. In the case of Indo-Aryan and Germanic, which have generalized the default initial, i.e. root-stress of the sg. to the rest of the paradigm, the key stage was the reanalysis of sg. * * * as (* * *. Once again, we are dealing with the question of reanalysis of underlying forms by language learners and/or speakers: after pre-Indo-Aryan and preGermanic had (mostly) lost or eliminated the possibility for stress to appear on a proclitic (or enclitic) attached to an underlyingly unaccented form, speakers, and learners, would no longer have been able to deduce that a surface form such as *pékweti, with default initial stress, was in fact underlyingly unaccented (in Dybo's words, an "enclimenon") rather than had an accented root (*. Once the root is accented, of course, the du./pl. forms will receive root stress as well, according to the rules for stress computation on the underlying forms (* * (*; the result is columnar root stress throughout the paradigm. initial-stressed
lsg. 2 3
*(pékw-oh2 *(pékw-e-si *(pékw-e-ti
lpl. 2 3
*(pékw-o-(mos *(pékw-e-(te *(pékw-o-(nti
Proto-Balto-Slavic, on the other hand, appears to have retained the proposed alternation of default initial stress in the sg. and ending-stress in the du. and pi., given the evidence of Old Prussian. Stang (1966:452-3) has suggested that pres. lpl. giwammai (2x), giwemmai (lx), i.e. [glvámai], has retracted the stress from earlier ending-stressed *gīvamái, much as in Russian, Bulgarian, Serbo-Croatian (after long-vowel roots; cf. §6.3.1), and Slovincian. This stress on the thematic vowel was generalized to singular forms such as 2sg. giwassi [gīvásei] (2x), 3sg. giwa [glvá]. The variant 2sg. glwasi (lx) may however preserve the original unaccented preform *glwasei reconstructed above for 204
PBS1. (so Stang, p. 452), as may 3sg. gīwu, if Stang is correct in interpreting it as a misprint for gīwa*. This deduction is borne out by the handful of athematic forms found in the Enchiridion: in 2sg. ēisei 'go', dāse 'give', root stress is expected with the unaccented ending -sei (se)J* Contra Stang (1966:451-2), the stress of lpl. perēimai 'we come' need not imply an unaccented athematic lpl. ending -mai, but may simply reflect the same retraction as in thematic giwammai (giwemmai), i.e. *[per-éimai] < *per-eimái. Based on this handful of forms, it therefore appears that sg. and pi. endings remained underlyingly unaccented and accented, respectively, in both athematic and thematic paradigms in Old Prussian! In contrast, East Baltic and Slavic have generalized the underlying accentuation of the non-sg. primary endings to 2sg. *-si, 3 *-ti.19 Direct support for this hypothesis is furnished by the accentuation of athematic presents in medieval and modern Slavic. Already in PS1., athematic inflection in verbs had been reduced to the five presents *jes-, *dad-, *jéd-, *véd- (3sg. *jes-tí, *das-fí, *jés-tí, *vés-fí),20 and *jíma- (of which the last is clearly a Slavic innovation), and scattered relics such as pres. 2sg. *moši, 3sg. *mot'I 'you, s/he can' < *mog-sei, *mog-ti, OCs sqtl 's/he says' < *sengwh-ti (van Wijk 1926:288-9), pres. act. ptcp. OCS gor-gšt- (¿-stem 3sg. gor-i-tī<— 3pl. gorrtK *-ņti); cf. in detail Vaillant 1934. Where reflexes of these forms survive, in Old Russian manuscripts or in modern languages such as Serbo-Croatian or Ukrainian, they unanimously exhibit stress on the ending, in the sg. as well as the pi.; cf. Stang 1966:45118
In 3sg.perēit 'comes', the vowel of the ending *-ti has been syncopated (as optionally in Old Lithuanian), so that its accentuation tells us only that the default initial stress of ei- 'go' must have been generalized to compounds, hence per-ēit for *pér-ei-ti. 19 But probably not the corresponding secondary endings *-s and *-t, which have remained unaccented in the aorist. In Serbo-Croatian, the aorist (i.e. the productive oxu-aorist of late OCS) exhibits a well-known alternation between unaccented 2, 3sg. and (underlying) stress on the thematic vowel in the lsg. and the pi., e.g. do-vede 'you, s/he led, brought' vs. lsg. do-véd-o-h, lpl. do-ved-osmo, 2 do-ved-o-ste, 3 do-véd-o-še. The stress of the latter forms has been taken from the thematized aorist, but the 2, 3sg., already attested in OCS, can be in origin nothing other than an old pre-PSl. (< post-PIE) thematic imperfect. Pre-PSl. *wed-e-s, *-t, *rek-e-s, *-t > PS1. *ved-e, *reð-e, etc. must therefore have remained unaccented; their incorporation into the paradigm of the sigmatic aorist (OCS rēxú, vé's-ú) and subsequently the oArti-aorist created the distinctive alternation which has been eliminated in standard Bulgarian (cf. Garde 1976a: 115), but has preserved to this day in the Rhodope dialects and in the transitional Torlak/Macedonian/western Bulgarian dialect zone, as well as standard SC (Alexander 1975:324-67). See R. Kim, in progress b. 20 Archaic 1 sg. védē to vēsti, which famous continues *woyday < PIE *woyd-h2e + primary *i, is attested in OCS, Old Russian, Old Czech (vide, until the 16th c ! mod. viem), and the Old Slovene Freising Leaves (is-po-vede); elsewhere and in later texts, it has been replaced by vein] after the pattern of the other athematic presents (Vaillant 1966:76, 448-50).
205
2, Garde 1976a: 117, 162-3. Cf. the following paradigms of 'be' in Cakavian and standard štokavian Serbo-Croatian ("long", i.e. orthotonic forms):
Isg. 2 3 lpl. 2 3
SC (Cak., negated) riīsán nisi ní
SC (standard)21 jésam jési jést(e)
nīsmo
jésmo jéste jésu
riīsté
riīsú
Similarly, the reflexes of 'give', 'eat', and 'know' follow the same pattern: in Ukrainian, the disyllabic 2sg. and 1, 2pl. are stressed on the ending; in Cakavian, the stress of the 1, 2pl. is confirmed by the rising intonation on the other forms (cf. Belie 1909:252-3). Ukrainian lsg. 2 3
dáty 'give' dam dasy dast'
jísty 'eat' jim
βsy βst'
opo-vísty 'tell' opo-vím opo-visy opo-víst'
lpl. 2 3
damó dasté dadút'
jimó jisté jidjáf
opo-vimó opo-visté opo-vidját'
Cakavian lsg. 2 3
dán dáš dá
jín
βš
β
po-vín po-víš po-ví
dāmo
jīmo jltéjīsté jldú
po-vīmé po-vīté, po-vīsté po-vīdú
lpl. 2 3
dāté, dāstš [dajú]
Ending-stress probably also survives in the plural forms of 'give' and 'eat' in Russian, which have been remodeled with the endings of the i-conjugation on the basis of 3pl. PS1. *-e.ti < PIE *-énti: dad-ím, -íte, -út (<— *dadját; cf. Ukr. dadút'); ed-ím, -he, -ját. All the evidence, then, points to columnar stress on the ending in the four archaic athematic presents of PS1.22 21
The negative forms in standard SC are usually ni-sam, -si, nije, m'-smo, -ste. -sir, uncontracted ni-jesam, etc. are also found in some regional varieties, e.g. in Montenegro. 22 SC (Dubrovnik) lsg. imam, cited by Derksen (1991:75), is surely an error for imam < *inrām < *jíma-nrī, which matches lpl. imámo < *imāmo < *jīma-m'o, *-miá.
206
That this columnarity is of PBS1. date emerges clearly from the Lithuanian data, which is much more probative here than for the original accentuation of thematic presents (cf. §6.4). Just as Lithuanian has preserved consonant-stem nominal declensions to a far greater degree than Proto-Slavic, so the Old Lithuanian manuscripts, together with data from the dialects, attest many more "/m'-verbs" than the five of Slavic — eighteen, according to Senn (1966a:286-97). In accented OLith. texts such as that of Dauksa, athematic presents are for the most part accented on the ending, just as in Slavic forms cited above; cf. the following examples: brfti 'be': sg. 1 esmi (neg. nesmi), 2 est (esie-gu tu? 'are you?'), pi. 1 esmé, 2 esté; 3 ēst(i) (cf. modern ēsti 'there is' vs. copula yrá) déti May': derrii (refl. demies), 2 desi* (refl. desies), pi. 1 demé; 3 dēst(i) eīti 'go': sg. 1 eimi, 2 eisi, pi. 1 eimé, 2 eité, eisté (the latter by analogy to esté, desté*, dúoste); 3 eīti (cf. Zemaitian eitu, -i, eīt, eītame, -te) gélbéti 'help, save': 3 gélbt(i) (dial, gélbti, gélbstí) likti 'leave, remain': sg. 1 liekmi, 2 lieksi, pi. 1 liekmé; 3 liēkti (also dial.) mégti 'like': 3 mégti (alongside méga, mégia, mégsta; latter now standard) miegóti 'sleep': sg. 1 miegmi, 2 miegsi, pi. 1 miegmé, 2 miegté, miegsté; 3 miēgti (also dial.) skaudéti 'hurt': 3 skausti (modern skauda) snigti 'snow': 3 sniēgti (also widespread in dialects alongside sniēga; standard sninga) vs. dúoti 'give': sg. 1 dúomi, 2 dúosi, pl. 1 dúome, 2 dúoste; 3 dúost(i) ésti 'gobble', OLith. and dial, 'eat': sg. 1 émi, 2 ési, pl. 1 éme, 2 éste; 3 ést(i) giedóti 'sing, crow': sg. 1 gíemi; 3 gíest(i) (also dial, giesti; standard gíeda) kliedéti 'misspeak': sg. 1 klíemi, 3 klíesti raudóti 'cry loudly, sob': sg. 1 ráumi, 2 ráusi; 3 ráusti sérgéti 'hiiten': sg. 1 sérgmi, pl. 1 refl. sérgamés; 3 sérgti (also dial., along with other variants; standard sérgi) In diiomi, ami, and the other verbs in the latter list (to the extent that these reflect old athematic presents), the stress may be the result of Hirt's Law: PBS1. *dó:d-'mí, *do:d- si. etc. have retracted stress to the preceding acute syllable, giving *dó:d-mi, *-si.23
23
The columnar ending-stress of PS1. *dami, etc. would then be analogical to that of e.g. *jesmi and *v
207
The complete agreement of Slavic and East Baltic in having columnar ending-stress in athematic conjugation strongly implies that the sg. primary endings *-mi, *-si, *-ti were accented in thematic presents as well. We would therefore expect columnar ending-stress in the PBS1. thematic presents *ber-e/a-, *pek-e/a-, etc., with one prominent exception: the lsg., where the PBS1. thematic ending *-ó: < (post-)PIE *-ō differs markedly from athematic *-mi, and lacks a postthematic syllable. lsg. 2 3
*ber-o: *ber-e-(si *ber-e-(ti
*pek-o: *pek-e-(si *pek-e-(ti
lpl. 2 3
*ber-a-(más *ber-e-(té *ber-a-(nti
*pek-a-(mas *pek-e-(te *pek-a-(ntl
In the light of historical reconstruction, it is possible to reformulate the evolution of simple thematic present accent in BS1. as follows. Old Prussian preserves the PBS1. situation, in which the sg. endings are underlyingly unaccented and the du./pl. endings accented, as in PIE (see above). In East Baltic and Slavic, the athematic primary sg. endings adopted the underlying accent of the du./pl.: as a result, the thematic present contrasted non-lsg. *ber-e-sei (?), *ber-e-tī, *ber-a-más, etc., with stress on the accented ending, vs. unaccented lsg. *ber-o:, containing the peculiar monosyllable thematic ending *-o: in which the original PIE theme vowel *-o- and ending *-h2 have fused. Speakers of pre-EBalt. and pre-PSl. could then have reanalyzed the non-lsg. forms with a postaccenting theme vowel *-e/a-; this analysis would have had the beneficial effect of unifying the accentual properties of the present, as unaccented lsg. *ber-o: could now be interpreted as underlying *ber-ó:(, with Stray Bracket Erasure (§§6.3.2-3). In pre-PSI., the PBS1. lsg. ending *-ó: was replaced by *-ó:-mi, thus eliminating the accentual alternation between the lsg. and other forms. Since lsg. *-mi had survived in athematic paradigms — which were doubtless far more numerous in pre-PSl. than in OCS! — one might suppose that the accentuation of *-o:(rni merely reflects the addition of accented *-(mi. This assumption is unnecessary, however: if the thematic vowel in the non-lsg. forms (and in unaccented lsg. *berņ, etc.) was postaccenting, it would have surfaced in *ber-o:-mi, *pek-o:-mí, etc. as stress on the new postthematic syllable head *i. Be that as it may, the subsequent apocope of word-final *-i after sonorants led once again to unaccented ("enclinomenon") lsg. *berq, *pek, restoring the PBS1. alternation:
208
this state of affairs is attested in Old Russian and Middle Bulgarian, and survives in altered form as a morphophonological rule in Bulgaro-Macedonian-Torlak dialects and in Slovincian (§6.3.1). Subsequently, certain Slavic languages, e.g. Serbo-Croatian have remodeled the lsg. to vowel + athematic *-m (e.g. ðakavian -ēn, underlying /-ena/, stokavian -ēm, underlying /-ēma/) and thereby eliminated the PS1. alternation. The SC thematic present lsg. thus results from three (!) shifts of stress since the PBSi. period, all resulting from the vagaries of the fickle ending *-ō —> *-ōmi > *-Q —> *-emī: (post-)PIE PBSI. pre-PSl w w *pek -o-h2 > *pek -ō —> *pek-ō( —> *pek-ō(-mi > PS1. *pekq( —>
pre-SC peðe(mi >
ðak. peðén, štok. péðēm
In East Baltic, however, all person-number endings in the thematic present became unaccented (cf. Stang 1966:451). The reasons for this are not clear, but undoubtedly involve the prehistory of the 2 and 3sg. endings: if 2sg. *-ái (> Lith. -i, -ie-s(i)) is an innovation of PBSI. date (cf. fn. 9), the only sg. form containing a syllable head after the thematic vowel would have been 3sg. *-e-ti. Once the final *-ti of the latter ending had been lost (probably via apocope followed by loss of newly word-final *-t; cf. PSI. *-eti > SC, WS1. -e), all three sg. forms would have been unaccented by the argument in §6.3.3: the default initial stress of the sg. was then extended to the (du. and) pi. 6.6. The distribution of presents and aorists in Proto-SIavic: an unexpected IndoEuropean inheritance Having established that simple thematic presents have underlyingly unaccented roots and postaccenting thematic vowels, we may now consider the relationship between accentual properties and the morphology of the present and aorist in PSI. and OCS. In 1926, van Wijk published a short article in which he noted the correlation between a morphological peculiarity of the aorist in OCS and the accentual properties of verbal stems in the modern Slavic languages, above all Serbo-Croatian and Russian. In PSI. and OCS, verb stems ending in a vowel, i.e. those of the shape *Ci, *Cu, *C^, *CQ (or *CVN), and *CVR, may form their infinitive stem in one of two ways: either with the suffix *-a-, e.g. *bir-a-ti 'gather, take', pl'Iv-ati 'spit' to pres. *ber-e/0-, *pl'u-je/0-; or
209
without a suffix, e.g. *j$-ti 'die', *bor-ti (OCS brati) to pres. *jim-e/0-,24 *bor-je/0-. To the former group, like other verbs with vowel-final infinitive stems, one finds endingless aor. 2, 3sg, past pass. ptcp. in *-nu, and verbal noun (derived from the past pass. ptcp. by means of the suffix *-ije) in *-n!je, e.g. OCS na-pisa 'wrote', na-pisanu 'written', pīsanie (for *-Ije) 'writing'. For the latter group, however, there are two possible formations of the abovementioned categories in OCS. Van Wijk demonstrated that those vowel- or diphthong-final stems whose reflexes in Serbo-Croatian and Russian are columnar barytone, and therefore had acute intonation in PS1., exhibit endingless aor. 2, 3sgs., past pass, ptcps. in *-nu, and verbal nouns in *-mje. By contrast, those stems whose reflexes exhibit mobile stress in SC and Russian take the ending *-tu in the aor. 2, 3sg., and form the past pass. ptcp. and verbal noun in *-tii and *-tije, respectively.25 OCS inf. bi-ti 'beat' li-ti 'pour' kry-ti 'hide' my-ti 'wash' Si-ti 'sew' sé-ti 'sow' otú-vē-ti 'answer' kla-ti 'stab' trv-ti 'rub' Zru-ti 'sacrifice'
aor. 2, 3sg. bi vu-li kry my vú-sé otu-vé -kla -tru, -trī (po-¿rú)
past pass. ptcp. bījenú -kruvenú -múvenú íívenú vú-sē-no klanv (koljenv) (sú-tīrení) po-2rútú
pi-ta pi-tú pi-ti 'drink' vi-ti 'wrap (up), turn' po-vi-tú po-vitú u-mrē-ti 'die' u-mré-tú pē-ti 'sing' pē-tú pētú pro-strš-ti 'stretch out' pro-strē-tú po-2rē-ú 'swallow' po-irētú -óq-ti 'begin' -Cq-tú jŗ-ti 'take' j§-rt klŗ-ti 'swear' klŗ-tú
24
SC aor. b¡ (fl) krí ml (mi) (Si)
Russian pret. fem. sg. bí-la (li-lá) kry-la my-la ší-la
tr, o-tr A,
PI
pi-lá vi-lá u-mer-lá (pé-lá) (pro-stēr-lá)
po-ðē
na-ða-lá vz-ja-lá (pro-)klja-lá
vi ú-mr(ij)e
klé
E.g. in *vúz-^ti 'take', pres. *vúz-ímetí > Russ. vzjat', voz'met, SC úzēti, úzmē. Meillet (1902-5:136) had already noted the "coincidence...remarquable" of endingless 2, 3sg. aor. and past pass. ptcp. in -enu to roots in -y- (kryti, myti, Siti) and aor. and ptcp. in -tu to verbs such as j'f//', (po)viti, (pro)stréti, but apparently did not realize the full extent of the correlation. Cf. also van Wijk 1931:222-3, 231-2. (The statement in R. Kim 2001a:154fn.7 is inaccurate and should be ignored.) 25
210
Subsequent research (e.g. Dybo 1961) has confirmed van Wijk's hypothesis, and accounted for a number of exceptions which he pointed out in his original formulation. For instance, van Wijk (1931:241) and Stang (1942:70fn.2) had noted that one would expect PSI. verbs in inf. *-ov-ati, pres. *-u-je/0- to belong to the second group, with aor. 2, 3sg. in *-tu and past pass. ptcp. in *-tu, given the accentual evidence of e.g. SC ; yet the only aor. forms attested in the OCS corpus are endingless na-tru 'fed', plu 'sailed, swam', vvzdru 'bellowed'. On the basis of an exhaustive study of Church Slavonic manuscripts, Koch (1985) has provided the missing evidence in the verbal nouns OCS slutije 'fame, reputation', CS1. slutije,plutīje, rutīje (r'utlje), with the endings predicted by van Wijk's generalization. The OCS forms have merely replaced aor. slutií*, plutv, rutú*, a process which has already neared completion in the late OCS Suprasliensis manuscript, where van Wijk demonstrated that the few aor. 2, 3sg. forms in -tv are mostly to be found in otherwise archaic passages closely resembling the Glagolitic text.26 Intriguing as this pattern was, it remained isolated within historical treatments of Slavic and Indo-European, not only because the accentual and intonational system of Proto-Slavic was not well understood at this time, but also because the origin of the various endings was less than fully clear, and van Wijk's generalization was confined to voweland sonorant-final roots only. In 1961, Dybo took a major step forward in the study of the PSI. verb by extending van Wijk's observation to obstruent-final stems. On the basis of a comparison of simple thematic presents of obstruent-final stems and their corresponding aorists, Dybo noted the following pattern. PSI. verbs for which the comparative evidence of the modern Slavic languages points to an originally unaccented present paradigm — i.e. of the sort discussed above in §6.3 — formed sigmatic aorists in OCS. On the other hand, for verbs with root-accented or postaccenting presents, i.e. presents belonging to Stang's accentual paradigms a or b, the earliest OCS sources attest no sigmatic aorists, but only thematic aorists. The following forms make this clear:27 26
Koch (1985:129) has identified two possible CS attestations of the expected OCS aor. in -fu, namely vvzdrjutl (in bykií uspe i vúzdrjutī velīmi 'the bull fell asleep and bellowed greatly'. Gr. eu^'tis ho taítros mukε:thmon apotelésas mégan) in a 15th-c. Russian CS manuscript, and proslutī se selo 'the village was famed' in a 16th-c. Serbian CS text. 27 I omit the aorist of athematic *jéTs-ti 'eats', jés-ú, -omu, -ŗ (also jaxújéstú/jastú, jéxomú. y'ēíe. under the influence of byxJ, daxú, vēxu*), which has clearly been remodeled from pre-PSl. 2, 3sg. root aorist *jér (-é" in compounds is probably secondary, cf. van Wijk 1931:223-4). The root *hļedoriginally had no aorist in PIE; the root aorist of 'eat', as well as that of 'know' (<— post-PIE perf. *woyd- ~ *wid-), are analogical on those of *da-, *by-. Although its root ends in a glide rather than an obstruent, note that PSI. *Ziv-e-fi (AP c) 'lives' forms a thematic aorist (o-, po-)Ziv-e; the late OCS forms ¿i, Ziti, first attested in
211
*virg-u (OCS vurg-g) pad-Q u-krad-g lēz-g jéd-g sēd-ú, -omií, -g *by (OCS by-xú, -stú, -xomú, -šq, etc.) -rētú, -omú, -g, -ete leg-g
*virf-e-fí *pád-e-tí *krád-e-tí *léz-e-tí *jéd-e-tí *s^d-e-ti *b<5d-e-tí *nft-(j)e-tl *l$Z-e-fí
'throws' 'falls (pf)' 'steals' 'climbs' 'rides' 'is sitting' 'is' 'meets' 'lies down'
*moz"-é-tí *jíd-é-tí
moiete, mogomii, -g can 'goes, walks' *jidu (OCS idu, -omv, -ete, -g)
*bljud-e-ti *bod-e-ti *cvít-e-á *clt-e-ti *greb-e-ti *klad-e-tí *l?k-e-tí *me.t-e-íi *nes-e-tí *rek-e-tí *sék-e-ti *súp-e-ti *tek-e-á *tr^s-e-ti *ved-e-tí *velk-e-ti *vīrz-e-ti *vez-e-tí
'observes' 'pierces' 'blooms' 'observes' 'digs, rows' 'puts' 'bends' 'stirs' 'carries' 'says (pf)' 'cuts' 'spreads' 'runs' 'shakes' 'leads' 'pulls' 'shuts, opens' 'transports'
*2eg-e-ti
'burns'
bljus-u, -q bas-q CViS-q
Sis-q grēs-q klasq (Bosnian CS, 1404) lqx-ú, -omv mqs-v, -omv, -q (also mqSq) nēs-ú, -ta, -q rēxv, rēsta, rēxomv, rēsta, rēšq sēšq ra-su-ša (-šg, -sé) sq2% tēxv, tēsta, tēšq trqs-ú, -ta vēs-v, -ta, -te, -q vlēxv, vlēxomv, vlēste, vlēšq vrēs-v, -ta, -q -vésta (Bulg. CS, 1 2 t h c ; Bosn. CS, 1404) ¿aS-q
As Reinhart (1992:369-70) points out, all deviations from this regularity are late and sporadic, and involve the substitution of a thematic aorist for the expected sigmatic form, e.g. 3pl. trqsg 'they shook' with thematic -g.29 The latter was already a historical relic by the time of earliest OCS, a fact discernable from the replacement of the expected 2, 3sg. Suprasliensis, have clearly been made to the vocalic stem 2i- of inf. ii-ti (Koch 1976:123-35). This verb, however, is anything but an ordinary e-grade simple thematic present; cf. §6.1, fn. 2. 28 For expected *-šu-še. < *seup-. On the original paradigm of this verb, in particular aor. *suxú, *su-še., etc. ~ *súp-e, see Koch 1986. 29 For examples of this phenomenon in Croatian Church Slavonic manuscripts in Glagolitic script, cf. Reinhart 1988.
212
with suppletive thematic forms, e.g. rei-e 'you, s/he said', nes-e 'you, s/he was carrying'. The PS1. thematic aorist, on the other hand, survived longer in Church Slavonic and other medieval Slavic written documents and appears to have enjoyed at least a small measure of productivity alongside the innovative and eventually completely dominant "erweiterter Aorist". As with the pattern determined by van Wijk for stems ending in a vowel or diphthong, the historical justification for the synchronic division of obstruent-final stems into two groups, each corresponding to a particular accentual paradigm in the simple thematic present and a particular aorist formation, appeared to be lacking. Two of Dybo's students, S. L. Nikolaev and S. A. Starostin, have attempted to explain away the Slavic facts by projecting the twofold division of verbal stems back into PIE (1981). This study, which betrays a number of misconceptions about the morphology of the Vedic and Classical Sanskrit verb and Greek historical phonology, has been rightly, if roundly and perhaps savagely criticized by Reinhart (1992:371-5). Yet Reinhart's entirely negative conclusion, that "einerseits wird man die synchrone Bindung von Akzentparadigma und aksl. Aoristbildung nicht leugnen konnen, andererseits spricht nichts dafiir, dieser Verteilung grundsprachliches Alter zuzuerkennen" (375), is difficult to accept. Why would Slavic have created a arbitrary division of verbal stems into two groups with distinct phonological and morphological properties — a division, furthermore, without apparent semantic or functional connection? It would be far preferable to assume once again that, just as Slavic has preserved features of the accent of the PIE noun (cf. ch. 4) and, if indirectly, of the post-PIE simple thematic present (§6.5), here too Slavic has inherited a distinction between underlyingly accented and unaccented morphemes, one, moreover, with some relevance to the morphology to the late or post-PIE verb. Before investigating this problem, let us summarize the correspondences discussed above in tabular form: sonorant-final roots PS1. pres. accented acute (AP a) unaccented (AP c)
aor. 2,3sg. *-0 *-ru
obstruent-final roots PS1. pres. accented (AP a, b) unaccented (AP c)
aor. thematic sigmatic
213
PPP *-nú *-nl
v.n. *-nije *-tije
Vaillant (1966:55-6) has conjectured that the contrast between *-0 and *-tú in the aor. 2, 3sg. of sonorant roots must reflect a stress-conditioned sound change involving word-final *-t. For now I leave this problem aside, focusing below on the accentually distributed aorist formations of obstruent-final roots. After analyzing the comparative and philological evidence, Garde (1976a: 149-54) and Dybo (1981:208-9 with refs.) have concluded that the majority of simple thematic presents were originally unaccented in PS1. We have seen above in §6.3.1 that these have developed into paradigms with columnar stress on the thematic vowel in standard Russian and Bulgarian, and into paradigms with stressed 1 and 2pl. in Cakavian (-mo, -té; retraction in standard štokavian SC) and Ukrainian (-mó, -te). Interestingly, the majority of PS1. simple thematic presents for which an unaccented, i.e. mobile paradigm may confidently be reconstructed share the root vowel *e. For a number of these verbs, this is only to be expected: cognates in other IE languages reveal that the following continue post-PIE simple thematic presents of the *bhér-e/0- type (cf. LIV s.w.): *bljud-e/0- 'observe' < *bhewdh-e/0-, cf. Ved. bódhati 'notices, pays attention to', Gr. Horn. peúthomai 'come to know, perceive', Goth, ana-biudan 'order, command'; [*l?k-e/0- 'bend' < *lenk-e/0-, no secure cognates, but cf. Lith. lenkiu 'bend, bow';] *me>e/0- 'stir' < *ment-e/0-? cf. Ved. mánthati 'stir (up)'; *pekw-e/0- 'cook' < *pekw-e/0-, cf. Ved. pácati, Gr. pésso:lpétto:, aor. épepsa, Lat. coquō (< *kwekw-); *rek-e/0- 'say (perfective)' < *rek(H)-e/0- 'order, determine' (not securely attested elsewhere, but cf. Skt. racayati 'produce, effect', deverbal OCS rokv 'date', Lith. rākas 'date, border', TB reki, TA rake 'word'); *ved-e/0- 'lead' < *wedh-e/0-, cf. Goth, ga-widan 'bind', Olr. fe(i)did 'leads', Lith. vedii 'lead, marry (of a man)' (Hitt. uwate- 'carry' < *au- + *wodh-éye/0'lead', cf. Lehrman, cited in Melchert 1994a:66, 134) *vez-e/0- 'travel' < *weg'h-e/0-, cf. Ved. váhati, Av. vazaiti, Gr. Pamphyl. iptv. 3sg. wekheto: (Horn. ókhos 'chariot'), Lat. vehō, Lith. vetti, OE wegan 'carry', Alb. vjedh 'steals'; *2eg-e/0- 'burn (intr., tr.)' <— *d h eg wh - e / 0 - (with assimilation of the first consonant), cf. Ved. dáhati, Av. daiaiti, Alb. djeg 'burn (tr.)', Lith. degii 'burn (intr., tr.)', TB pketār (tr.), pakstār (intr.), root /pak-/ with delabialization (R. Kim 1999:159-61). As we have seen above, most of these verbs form sigmatic aorists in OCS: grēs-q 'they dug', lŗx-ú 'I bent', mŗs-v 'I stirred', nēs-ú 'I carried', réxú 'I said', trŗs-ú 'I shook', vés-
214
u 'I led', vlēxv 'I pulled', otī-vēsta sq 'the two of them sailed off, 2aš-ŗ 'they burned'.30 We have here the "classical" IE system, according to which simple, i.e. uncharacterized and unsuffixed presents form characterized aorist stems. In Slavic, as in a number of other IE languages (particularly Indo-Iranian, Greek, Italic,31 and Celtic), many simple thematic presents form a sigmatic aorist, containing lengthened-grade *ē in the root and a suffix *-s, to which are added the secondary person-number endings. This formation, like the simple thematic present, appears to be an innovation of "Brugmannian" PIE, with Hittite and Tocharian preserving an earlier formation in which the sigmatic suffix is confined to the 3sg.32 Already by the time of Vedic Sanskrit and Gatha Avestan, however, sigmatic aorists are well established, although with certain synchronic peculiarities which betray their relatively "late" prehistoric origin.33 In the classical languages, they have become a productive formation to thematic presents, including derived verbs which had only a present stem in PIE: cf. Gr. phor-éo: (< post-PEE iter. *bhor-éye/0-, cf. §2.3.2.2c), denom. paid-eúo: 'instruct' (<—pats,paid-ós 'child'), aor. ephórε:sa, epaídeusa; Lat.jungo 'join' (thematized from nasal-infix present, see below), maneō 'stay, remain' (< stative *mn-éhi- to *men-), perfect jvnxī, mānsī. With regard to Slavic, the observation that simple thematic presents with root vowel *e, as reflexes of post-PIE simple thematic presents or thematized root presents (or aorists-turned- presents), unanimously take the expected characterized sigmatic aorist is hardly new. Already at the beginning of the 20th century, Leskien noted that "Verba mit e 30
Excluded from this statement are verbs such as OCS ber-e-tu itself, which have acquired a suffixed infinitive in -a-, e.g. blr-a-ti: the corresponding aorists, e.g. bīr-a (cf. pīs-a at the beginning of this section), have replaced whatever formation previously existed. 3 'More precisely, Latin, as the Sabellic languages preserve no sigmatic perfects continuing the sigmatic aorist indicative, but only s-futures from the (post-)PIE sigmatic aorist subjunctive (§2.3.1.3). 32 Despite the productivity of sigmatic aorists in many IE languages (cf. the examples below in the main text), it appears that the formation was just coming into being as an inflectional category at the time of the breakup of the last common ancestor of the classical ("Brugmannian") languages: this is suggested by the peculiar fact that only two sigmatic aorists have cognates in three different IE branches, namely Av. dais, Gr. édeikse (<— *deiks) 'showed', Lat. dixit 'said' to *deyk'- (pres. Av. 3pl.daēsaiieiņti < *doy¿-éy e / 0 -, Lat. dīcō, Gr. deík-nu-mi; Ved. pres. dišá- from thematized aor.? cf. LIV:92-3, who however take the j-aor. to be an innovation), and the aor. to *weg'h- 'convey, ride' (see below). 33 In particular, the absence of a corresponding (active) optative: in Vedic and Avestan. sigmatic aorists are synchronically associated with root aorist optatives, e.g. Ved. opt. 2, 3sg. sah-yās 'you, s/he might conquer, seize' to sāks- < *sēg h -s- GAv. opt. lpl. zaētnā (< *za-ī-me for PIE *g lļ h r ih r me) 'we might win' (< 'leave behind') to inj. zāh-* < *g' h ēh r s- (Ved. has-; cf. innovative Av. opt. 3sg.fra-za-h-lt), opt. lpl. varazimā-ðā 'we might perform' to inj. vāraš*, subj. varašaiti < *wérg'-se-ti (Hoffmann 1967:31-2, 1968:4-5fn.4). On the prehistory of the sigmatic aorist, see Jasanoff 1988.
215
in der Wurzelsilbe des Prasens bilden diesen Aorist iiberhaupt nicht, also kein *nesú *vedú *tekuzu nesq vedq tekq, abgesehen von der 2. 3. sg. nese vede teðe" (1919:198; cf. Leskien 1962:132). Yet although Dybo has established that the majority of simple thematic presents were unaccented in PS1., the three-way connection between present inflection, accentual class, and aorist formation for PS1. obstruent-final roots has never been explicitly described. In short, simple thematic presents — with certain exceptions, to be discussed — are underlyingly unaccented (cf. the references above). As uncharacterized presents in post-PIE, they formed characterized aorist stems, in this case sigmatic aorists of the type ot(l)-vēsta sŗ 'the two of them sailed off (translating Greek apépleusan), corresponding to Skt. 2, 3sg. ávāt 'conveyed', Av. subj. uz-uuaiat 'he may convey out', Gr. Cypr. éwekse 'brought', Lat. vēxit 'rode'. There is thus no direct historical relationship between accentual type and aorist formation: both are tied to, and in a sense "determined" by, the simple thematic inflection of the present. Synchronically, of course, speakers of (pre-)PSl. must have associated the accentual properties of the present with choice of aorist; additional evidence for this hypothesis will be provided below. To this inherited core were added several other verbs, for which one may reconstruct an athematic root or nasal-infix present for (post-)PIE:34 *bod-e/0- 'pierce, stick' < *bhodh-e/0-: cf. Lith. bedu 'stick, dig', Lat. fodiō 'dig', mf.fodere < *b hodh-ye/0-, Hitt. paddai 'digs' (originally "^¿-conjugation" pres. with *o ~ *e ablaut? Jasanoff 1979:84-5, 1994:158-9); *cvit-e/0- 'bloom', only cognate Latv. kvitu 'glitter, sparkle' (thematized root aorist like the following?); *clt-e/0- 'count, read, honor' <— thematized root aor. *kwit-é/<5-? cf. GAv. (a)cistā 'recognized, realized', subj. cōiβat, Ved. cít-āna- 'coming into appearance', sigm. aor. acait <— root aor. acet* (Narten 1964:114); *greb- e / 0 - 'dig, row' < *g h reb h - e / 0 -, cf. Latv. grebju 'scrape, hollow out', Goth. graban 'dig' (also originally a "^e-conjugation" pres.? cf. 'pierce' above); *nes- e / 0 - 'carry' < thematized root aor. *hinelc-e/0-, cf. Lith. nešú; redupl. aor. Horn, énegkon 'brought, carried' (to phéro:), GAv. nqsat 'disappeared' (< *nansat <— *na'asat, cf. Ved. nešaf, Harðarson 1993:118-9fn.91) < *h]nehļņk'-e/0-; sigm. aor. GAv. nīšnāšāmā 'we will carry out' (LIV:222-3 s.v. *h¡nelc, but cf. *hine£) *sék-c/0- 'cut' < *sēkH-c/0-, thematized from Narten pres. *sēkH- ~ *sékH-? cf. Hitt. pres. 3sg. šākki, 3pl. šekkanzi 'know', Lat. secō, -are 'cut', sciō, scire 'know' (Vaillant 1942-5, Rix 1999:524-7);
34
0ther examples, such as *2en-e/0- 'chase' <— PIE athematic *gwhén- - *gwhn-' (Hitt. 3sg. kuēn-zi, 3pl. kun-anzi, Skt. hán-ti, 3pl. ghn-ánti), have acquired infinitive stems in *-a-, in this case *gun-a-ti.
216
*tek-e/0- 'run, flow' < *tekw-e/0-, cf. Ved. takti 'shoots into', Av. -taCinti, Olr. ■teich, Lith. tekii, Alb. n-djek 'pursue'; *tnjs-e/0- 'shake' < *tre-n-s-e/0-, thematized from nasal-infix pres.? cf. Lith. trļsu, inf. tristi 'become restless', Av. tarasaiti 'is afraid' < *tŗs-sJcé/ó-, but Ved. trásati 'tremble, is anxious, Gr. tréo: 'be afraid, flee from fear' < *trés-e/0-; *velk-e/0- 'pull, drag' < *welk-e/0-, thematized from *wélk- ~ *wļk-', cf. Av. inj. 3pl. -varacinta 'carried away', Lith. velkw, *virz-e/0- 'shut, open' < *wŗgh-e/0-, thematized from pres. *wérgn- ~ *wŗgn-', cf. OE wyrgan 'strangle', Lith. veriiii 'tie up, confine', Alb. z-vjerdh 'disaccustom' — or aor. stem? (cf. LIV:629, s.v. *vergh). These presents were apparently thematized in pre-P(B)Sl. at an early enough date to become fully asismilated in accentuation and verbal morphology: the only trace of an athematic origin is the root vocalism of *cvit-e/0-, *clt-e/0- and *v!rz-e/0-, which have generalized the zero-grade of a former ablauting root present (or aorist). As a result, they belong to AP c (unaccented root, enclimenon lsg. vs. ending-stressed non-lsg.) and, as we have seen, take sigmatic aorists in OCS. What of the exceptions alluded to above, those thematic presents in *-e/0- which nevertheless are accented or postaccenting in PS1.? Dybo (1981:208) reconstructs eleven such stems, listed below in their 3sg. form: APa *vir2-e-tl *léz-e-tí *pád-e-tí *krád-e-tí *jēd-e-tí *s?d-e-tí *b<>d-e-tí *r$t-(j)e-tí *l^Z-e-tí
'is throwing' 'is climbing' 'is falling' 'is stealing' 'travels' 'is sitting' 'is'35 'is meeting' 'lies down'
AP6 *moZ-e-ti *jíd-é-fí
'can' 'is walking'
As Dybo correctly notes, it is immediately obvious from the shape of these presents that they, unlike the unaccented presents previously discussed, do not continue post-PIE thematic or thematized presents: the nasal vowel of *s^d-e-ti, *bQd-e-ti, *r^t-(j)e-tl, *1^2350riginally perfective, regularly developing to 's/he will be' in Russian, Polish, etc., where it serves as the auxiliary in the imperfective future (with inf. in Russ., inf. or impfv. /-ptcp. in Pol.): confined to dependent clauses in Serbo-Croatian and Bulgarian, where it functions as the subjunctive of 'be', e.g. SC neka bude dobro 'let/may it be good', ako bude ¡(ij)epo, motemo da idemo {motemo ici) naplain *if it is warm, we can go to the beach', Bulg. závtra Ste bade (or Ste e) slanCevo "tomorrow it'll be sunny', ako bade (or ako e) studeno, da ostanem vkaSte 'if it's cold, let's stay home'.
217
e-tí reflects an nasal infix originally confined to the present, whereas the vocalism of the remaining presents contrasts with the normal root *e of *peð-e-ti, *nes-e-ti, etc. In the case of the nasal-infix presents, the columnar stress on the thematic vowel suggests that the thematization of the originally ablauting athematic paradigm has proceeded from the 3pl. (and possibly also the pres. act. ptcp.), just as in later Sanskrit, Italic, and Celtic, among other IE languages:36 PIE *s(e)-né-d-ti —> *send-é-ti *s(e)-n-d-énti —> *send-ó-nti > *send-ó-nti cf.
> PS1. *sefd-e-fí > *s^d-Q-fí
PIE *yu-né-g-ti 'joins, yokes' > Vedic yu-ná-k-ti —> later Skt. yunj-a-ti *yu-n-g-énti > yu-n-j-ánti > yunj-a-nti PIE *yu-né-g-ti —> *yung-e-ti *yu-n-g-énti —> *yung-ó-nti > *yung-o-nti
> Lat. jungit > jungunt
PIE *dhi-né-gh-ti 'moulds' —> *dhingh-e-ti *dhi-n-gh-énti —> *dhingh-o-nti > *dhingh-o-nti
> Olr. -ding 'presses' > -dengat
Unlike presents such as *bod-e/0-, *c!t-e/0-, or *nes-e/0-, which were thematized early and became fully assimilated to the accentuation of *bher-e/0- type, this small class of verbs retains underlying root accent and so must have been thematized at a later date. To the four verbs *s£d-e/0-, *br¿d-e/0-, *r§t-(j)e/0-, *Hg-e/0- with synchronic nasalinfix, i.e. characterized presents, one would expect the reflex of a PIE root aorist, according to the pattern — the reverse of that described above for uncharacterized presents — by which uncharacterized (root) aorists correspond to characterized presents, either suffixed with *-yé/<5- or *-slcé/ó- or containing a nasal infix. Examples of the latter, nasal-infix presents to verbs with root aorists, are numerous in Indo-Iranian, Tocharian, and Italic; the following are typical:37 36
In that case, the root vowels in PS1. could reflect either the expected zero-grades to the roots *sed-, *ret-, *leg-, e.g. *s-íj-d-e/0-, *l-ņ-g-e/0-, or "morphological" zero-grades *sed- (= *sed-), etc., e.g. *se-n-d-t70-, *le-n-g-e/0-. The other examples of thematized nasal-infix presents below have regularly generalized zero-grade of the root from the 3pl. I have omitted other sound changes in the development of the Lat. and Olr. forms, e.g. loss of lexical accent, apocope of *-i in primary endings, or vowel weakening in non-initial syllables, as the relative chronology of these and thematization of nasal-infix presents is indeterminate. On thematized nasal-infix presents in Olr., cf. McCone 1986:227-8 on verb classes B III and IV (in the notation of Thurneysen 1946:355-6) 37 Nasal-infix presents and root aorists to jeŗ-roots, i.e. PIE roots ending in a laryngeal, have been the subject of a number of studies by Strunk, e.g. Strunk 1967. Don Ringe (p.c.) has reminded me that among the nasal-infixed (class VII) presents in TB, four exhibit palatalization of the initial consonant in the preterite <— PIE root aorist, which can only have been caused by a following *e: cf.
218
PIE *kwreyh2-: aor. Horn. mp. 3sg. priato (<— 3pl. príanto < *kwrih2-énto), TB karya* < PT *kwarvya < *kwrih2-t, lpl. karyām; pres. Skt. krlņāti, krlņánti < *kwri-né-h2-ti, 3pl. *kwri-n-h2-énti (-ī- from v. adj. krī-tá-, etc.; cf. Pali kināti), Olr. crenaid < *kwrina- < *kwri-nā- (on ORuss. pres. lsg. krinju see fn. 45); PIE *leykw-: aor. Ved. mp. 2sg. prá rikthās 'you towered out over' < *likw-' (sigm. āraik <— athem. ārek*, Narten 1964:223-4), Arm. elik', Gr. ¿Upon, Lat. līquī 'left (behind)'; pres. Skt. riņákti, Av.irinaxti < *li-né-kw-ti, Lat. linquō, thematized < *Ii-n-kw-, cf. Arm. Ik'anem (Olr. léicid < *ley-n-kw-? cf. McCone 1991:47-52); PIE *pelh2-: aor. Gr. Horn. plε:to 'approached, drew near' < *pļh2-tó (Att. epélassa), pres. Av. mp. lsg. parane 'approach (s.o.)', Gr. Horn, pilnamai (for *pálnamai, cf. pítnε:mi 'fly' < *pt-né-h2-mi) < *pļ-n-h2-, Lat. pellō, inf. pellere 'strike, push, move, drive' (backformed from e.g. lpl. pellimus < *pelna- < *pļ-n-li2-C-; cf. compound ap-pellō, inf. -are 'address' backformed from e.g. 3sg. *ad-pelnā- < *ad-pļ-neh2-, Rix 1995:401-6), Olr. ad-ella 'visits' (fut. eblaid < *pi-pļh2-se-); PIE *peyg/g- (*peylc-): aor. ptcp. Ved. pišāná- 'adorning'; pres. Lat. pingit, pingunt 'paint', thematized < *pi-n-¿-, TB 3sg., pi. pinkām 'paint, write' (athematic! cf. Schmidt 1985:426-9).38 As noted in §2.3.2.1, PIE root aorists have been thematized in most IE branches: this process may be observed in the history of Sanskrit from the Vedas to the classical period, and has already affected almost all non-vocalic roots in Homeric Greek. Beginning with the 3pl. in *-ént —> *-ónt —> *-ó-nt and ptcp. in *-ónt- ~ *-ņt-' —> *-ónt > *-ó-nt-, e.g. *likw-ént 'they left (behind)', *likw-ónt- 'leaving (behind)' —> *likw-ó-nt, *likw-ó-nt-, the thematic vowel *-él(,- spreads to the other person-number forms, producing a paradigm with zero-grade of the root and columnar stress on the thematic vowel. This process has been carried to completion in Slavic, where the only forms which continue PIE root aorist preforms are vowel-final 2, 3sg. aorists such as zna, which have in turn been reinterpreted as endingless sigmatic forms (e.g. PIE *gneh3-t > PS1. *zna, reinterpreted as /zna-s-s/, TB pres. 3sg. prantsām < *pr-ņ-s- (also cl. VI parsnām, as if < *prs-n-h2-), pret. pirsa < PT *p*á rsa < *pers- (vs. mp. 3sg. parsāte < PT *pa rsa-tē < *pŗs-). Thus some verbs have generalized sg. full-grade, others du./pl. zero-grade of the root aorist, resulting in root-initial palatalization or non-palatalization, respectively. 3°Contra LIV (418-20), these forms may all be reconciled under a single root, probably *peyg-, with Indo-Aryan having generalized the variant with voiceless *-k- before an ending beginning with voiceless stop, i.e. 2sg. *-s(i), 3sg. *-t(i). Fluctuation in voicing or aspiration of a root-final consonant may be a diagnostic of a former athematic paradigm, e.g. Gr. Att. dékhetai vs. Ion. déketai 'receives' <— *délc- - % £ - ' (Horn. pres. 3pl. dékhatai 'look out, wait', impf. 3sg. dékto; LIV:93-4), Gr. eikei 'yields' vs. OE wīcan 'yield', Skt. impf. avije 'escaped' <— root aor. *(h|)wéyg- - *(h|)wig-' (Skt. aor. inj. mp. 2sg. vikthas, 3 vikta 'take fright, be taken aback'; TB inf. wišsi 'avoid', subj. 3sg. wišām could be from *weyg- or *weyk-); cf. Ringe 2000:132-3.
219
/zna-s-t/, whence lsg. zna-xú, 3pl. zna-šŗ with the endings of rē-xú, rē-šq or nos-i-xú, nos-i-šŗ). Of the four nasal-infix presents surviving in PS1., the vowel-final root *by- 'be' attests a root aorists *by-: OCS lsg. by-xv, 2, 3sg. by, by-stu, lpl. by-xomú, 3pl. by-šŗ, etc. The other three, with roots ending in a consonant, should take thematic, i.e. thematized root aorists (without the present-forming nasal), and that is exactly what we find: present *s
aorist sed-e/0- (sēd-ú, -omú, -g) rét-e/0- (-rētv, -omv, -g, -ete) leg-e/0- ilegg)
For the remaining accented simple thematic presents, however, it is less than obvious why they should form thematic aorists. PS1. *mog- 'be able', for instance, exhibits the same anomalous root vocalism as the Germanic preterite-present *may(Goth. 1, 3sg. mag, lpl. magum, OE mce3, magon) and Gr. mákhomai 'fight, do battle', Lith. magu, inf. magéti 'please, find approval' (cf. LIV:379); whatever its prehistory, it is clear that this verb does not belong to any of the well-established classes of simple or derived thematic presents. Similarly, PS1. *pad- 'fall' finds cognates only in Indo-Iranian (Ved. pád-ya-te 'falls, comes (in)to', Av. subj. paiSiiaite, perf. Ved. papāda, caus. pādáyati 'brings down, makes fall', stative aor. 3pl. apadran 'are fallen') and Germanic (ON feta, OE ge-fetan, OHG fezzan); whether or not its *a continues PIE *o lengthened before nonaspirated *d by "Winter's Law" (Winter 1978), it clearly is not of the ordinary *bher-e/0- type with *e-grade of the root.39 Whatever their origins, why should simple thematic presents form thematic aorists — or, if the latter go back far enough, root aorists, subsequently thematized? To understand the source of these exceptionally accented presents and their aorist formations, let us consider the other major classes of presents, namely derived presents with the suffixes * - j e / 0 - and *-ne/0-.4° Dybo (1981:209-12; cf. Dybo et. al 1990:65-6, 39
Jasanoff (p.c.) plausibly conjectures that this verb, with its aberrant root vocalism, may continue a PIE '7i2e-conjugation present" with original *o ~ *e ablaut: *pod— *ped-, thematized with *-e/0- and generalized o-grade in Slavic, but with *-ye/0- and generalized e-grade in Sanskrit. On PS1. *bod-e/0- and *greb-e/0-, two other likely candidates, see below. 4( -*I write the PS1. suffix as *-j e / 0 - in order to confirm to standard Slavist practice and to avoid confusion between the glide *j (= *[y]) and the high central/back vowel *y (< *u). — As the charts in Dybo 1981:203-5, 245-51 and Dybo et al. 1990:62-3, 75-8 make clear, presents to roots ending in a nasal or liquid are either unaccented or postaccenting (i.e. belong to AP b or c); since all such roots contain a short vowel, they do not undergo Hirt's Law. The verbs listed by Dybo as being presents to roots ending in *-j- are in reality *-j e / 0 - presents to long-vowel roots: if the suffix is accented, stress is
220
Garde 1976a: 154) observes that presents formed with either of these suffixes for which sufficient evidence exists to reconstruct their PS1. accentuation belong to accentual paradigm a or b, i.e. were either accented or postaccenting. In his view, all of these originally had columnar accent on the root; if the root contained a short vowel or circumflex long vowel or diphthong, the accent was advanced one syllable by Dybo's Law (cf. §4.3). Cf. the following examples: AP a *pláð-je-tí 'cries, weeps' *réz-je-tí 'cuts, chops, saws'
APb *pís-jé-tí 'writes' *£es-jé-fí 'plucks, combs, gathers' *lōZ-jé-ti 'lies'
Here, just as in the case of PS1. nominal accentuation, the distribution discovered by Dybo may be fitted more elegantly into our understanding of PIE accent if one substitutes his proposed forward shift of accent from a non-acute vocalic to the thematic vowel with a retraction from the thematic vowel syllable to a previous (i.e. root-final) syllable with acute intonation — in order words, Hirt's Law. In PIE, suffixed presents in *-yé/<5- were marked by zero-grade of the root and stressed thematic vowel: for examples and discussion, see §2.3.2.2. It is thus far more likely that these presents, like the handful of Slavic oxytone nouns of PIE origin (e.g. *div-u 'demon'), have preserved the accentual specification of their PIE preforms into pre-PSl., i.e. columnar stress on the suffix *-jé/ó-As a result of Hirt's Law, the inherited stress was retracted to an acute vowel in the preceding root syllable, producing e.g. PS1. *pl'áð-je-tí, *réz-je-fí < *plá£-j'e-íí, *rēz-je-tí, in AP a. The columnarity of AP b was then disturbed by the late pre-PSl. change known as Stang's Law, by which stress was retracted from a non-acute vowel in a non-final syllable, producing a new stress alternation in these presents, e.g. Russ. pisát', pres. pišú vs. píšeš',píšet,píšut < PS1. *píš-q, *-e-ši, *-'e-tí, *-'Q-tī, SC písati vs. (pišēm,) pišēš, pišē, pišu.4* retracted to the root in accordance with Hirt's Law, e.g. *bi-je-tl, *zna-je-fi > *bí-je-tí, *zná-je-fí. Hence this latter set of presents are either unaccented or postaccenting, i.e. belong to AP a or c. 41 The assumption of Hirt's Law also accounts for the fact that iteratives in PS1. *-iti with ograde of the root fall into either AP a or b (Garde 1976a: 158). This class of verbs continues the PIE *wos-éye/0- type with accented suffix *-éye/0- (§2.3.2.2c), which developed to PS1. *-i-. If the root vowel had acute intonation, the stress was retracted from the suffix, e.g. SC pres. lsg. lázīin, 3 lázl, inf. láziti 'climb' (as in ii-laziti 'enter, go in', iter, to u-l(j)esti), if the root vowel was circumflex or short, the stress was uniformly on the suffix *-i- in pre-PSl. before undergoing Stang's Law in the non-lsg. forms, e.g. Russ. pres. lsg. voZti, 3 vódit, inf. vodit' 'lead, drive' (nondetermined counterpart of vesti), SC pres. (lsg. proslm,) 3 pros!, inf. prdsiti 'ask'. Herein lies the accentual difference between iteratives and denominative i-stem verbs of AP c (derived from nouns of AP c) which was pointed out
221
So far nothing has been said of the prehistory of verbs in *-ne/0-. This important class, made up principally of intransitives, became moderately productive and has been well preserved in all the modern Slavic languages: cf. PS1. *dvig-, *dvi-ne/0- 'move' > 3sg. Russ. dvinet 's/he will move', Cz. z-dvihne 's/he will left', SC dlgnē 'moves' (with g restored from the thematic aorist, /-ptcp., and past pass, ptcp., see below). In a brilliant article published in 1948, Tedesco argued on the basis of a thorough philological investigation of OCS and Russian CS manuscript evidence that imperfective presents in *ne/0- have replaced earlier PS1. presents in *-j e / 0 -, relics of which have been preserved in a number of OCS and Russian CS texts. The *-ne/0- presents which Tedesco demonstrated to be of secondary origin are collected and listed below, together with the attested forms of the *-j e / Q - presents which preceded them (in OCS unless othewise indicated): u-gas-netu 'will be extinguished' u-gli(b)-ne/0- 'sink' po-grŗz-netú 'will be submerged' (po-, iz-)gyb-ne/0- 'perish' *vis-kris-ne/o- 'will rise' *mirz-ne/0- 'freeze' pri-lī-ne/0- 'will cleave to' vuz-nik-ng 'shoot up, emerge' i-sŗk-ne/0- 'will dry up' u-sv-netī 'falls asleep' i-svx-netú 'dry' pro-zŗb-netú 'will put forth' u-vr-nQ-ti 'wither' vyk-ne/0- 'get used to'
(ne) gašgšt< *gas-je/0Russ. CS u-gaSeti u-glībļje/0< *gfib-je/0ORuss. po-grqZetii < *gre-z-je/0(po-, iz-)gyblje/0< *gyb-je/0vvs-krúši < *krIs-je/0ORuss. za-merietxi < *mirz-je/0(Novgorod Zalovannaja Gramota, 1477)42 < *ííp-j<70iptv. 3sg. pri-llpli < *nik-je/0niðetī e Russ. CS i-sjaðjutí < *s(?k-j /0e Russ. CS u-svplete < *sup-j /0< *sux-je/0i-svš-etú, -Qtv Russ. CS pro-zŗbletī < *ze,b-je/0'will spring up' ORuss. u-vjaZd-eti, < *ved-je/0OSerb. u-veZd-eti vy&/0< *vyk-je/0-
As a result of Tedesco's study, which has unfortunately not received the proper attention in subsequent treatments of the PS1. verb or handbooks of PS1. and OCS grammar (particularly among Slavic accentologists), it is probable that all OCS and later Slavic by Brugmann and Delbriick (1916:244) and Stang (1942:25): cf. Russ. pres. lsg. gošðú, 3 gostít, inf. gostít' 'be visiting, staying' to unaccented gost' 'guest'. Stang's claim (1952b, 1957:112-3) that iteratives originally belonged to AP b and causatives (factitives) to AP c is probably inaccurate; cf. the discussion in Garde 1976a: 158-62. 42 Not only are *-j e / 0 - presents especially common in Russian Church Slavonic and Old Russian sources (all of which may contain OCS influence), but this form in a secular, vernacular document indicates that such forms survived into the 15th century in "genuine Old Russian" (Tedesco 1948:381-2).
222
imperfective presents in *-ne/0- are replacements of PSI. verbs in *-j c / 0 -. The two classes of suffixed verbs may therefore be considered together in historical investigations of PSI. verbal morphology. In PSI., as exemplified by OCS, verbs in *-j e / 0 - fall into two subclasses with regard to their infinitive stem, just like simple thematic presents. The majority, including all *j e / 0 - presents to obstruent-final roots in "classical" OCS (but see below), take the suffix *a-, e.g. pis-a-ti 'write', stīl-a-ti 'spread' (with zero-grade of the root), kolēb-a-ti 'toss, agitate' to pres. pišetú, stel'etv, kolšbl'etú < *pis-j e / 0 -, *stel-j e / 0 -, *kol
As for imperfective *-ne/0- presents, OCS attests almost exclusively thematic aorists: u-gleb-v (< *u-glīb-ú), u-glīb-g 'sank'; gyb-e (po-, iz-gyb-e), po-gyb-g 'perished'; vus-kris-e 'rose'; pri-līp-e 'stuck'; i-sjqð-e 'dried up'; i-sox-u, /'-, u-svš-e, i-súx-g (so-súxg sjŗ) 'dried (off, out)' (forms in -ng-, e.g. i-sŗk-ng, Supr. u-yúx-ng, are rare and late).43 Tedesco (ibid., 362-5) notes that these forms cannot be proper to the *-ne/0- presents only, but must also be the aorists to the earlier *-j e / 0 - presents underlying them (cf. above); in fact, the innovative presents in *-ne/0- were most likely built to the thematic aorist, along with the /-ptcp. and pres. act. ptcp., after the model of perfective *-ne/0- presents such as dvig-ne-tu, aor. dvii-e (po-dvig-lií, po-dvig-úš-). The traditional assumption of infinitives in *-a- to these archaic *-j e / 0 - presents, e.g. "gyb-a-ti", "svx-a-ti", is thus mistaken: we are dealing here with a class of verbs which formed *-ye/0- presents and (with the exception of velar-final roots in SC, unattested) unsuffixed infinitives, a class limited to sonorant-final roots in traditional OCS grammar. OCS aor. vvs-kris-e therefore corresponds to inf. *vus-kris-ti (—> vús-krīs-ng-ti), pres. iptv. 2sg. vús-krúši, and SC aor. nice 'sprouted' (whence lsg. nlkoh, etc.), /-ptcp. nikao, f. nikla, Russ. voz-nik, f. voz-nikla 'arose' to inf. *nit'i < *nik-ti (SC nici, —> Russ. voz-ník-nut'), pres. OCS niðetī (—> vúz-nik-ng, SC nlk-nē, Russ. voz-ník-nut. The parallelism between PS1. pres. *súS-e/0- (3pl. -svšgtu), aor. *sux-e/0- (OCS -soxú, -súše, súxg) and Skt. šúsyati, aor. a-šusat44 is complete: both continue the classical PIE pattern by which aoristic roots form characterized, i.e. derived suffixed presents, here with the most common present suffix *-yé/ó- (ibid., 350-1, 364). Tedesco (ibid., 383) has speculated that that PS1. perfective *-ne/0- presents are also reformations of older *-j e / 0 43
In Russian, perfective *-ne/0- verbs exhibit -nu- throughout the paradigm, e.g. derz-nú-t' 'dare', pres. lsg. derz-n-ú, 3sg. derz-n-é-t, pret. derz-nú-l, past gerund derz-mi-v(Si), but imperfective *nc/0- verbs limit it to the present, e.g. mérz-nu-t' 'freeze', pres. mérz-n-u, -et vs. pret. merz (fern, mérz-la), past gerund za-mérz-ši (also mérz-nu-l, mérz-nu-v(ši)), sóx-nu-t' 'dry (intr.)', sóx-n-u, -et, sóx (sóx-¡a), u-sóx-ši. Tedesco (1948:383-4) points out that this morphological discrepancy reflects the different age of the two classes of verbs: the nasal formant of the older perfective *-ne/0- verbs has spread from the pres. and inf. to all infinitive-derived forms, including the /-ptcp. and past gerund (< past act. ptcp.); in the newer, post-PSI. imperfective *-ne/0- verbs, by contrast, the nasal suffix remains confined to the present, where it has replaced * - j e / 0 - , and the inf. in -nu-, which has replaced an unsuffixed (root) infinitive. (This difference in the extent of -nu- also accounts indirectly for the differing accentuation of the two classes, e.g. suffix-stressed perf. derznút' vs. root-stressed sóxnut', as noted already by Meillet 1902-5:146.) Other Slavic languages are more archaic in this regard: in SerboCroatian, for example, all unsuffixed obstruent-final roots with *-ne/o- presents form unsuffixed oxUaorists and past participles, e.g. aor. lsg. digoh, 2, 3sg. dite, /-ptcp. digao, fern, digla, past pass. ptcp. diien to perf. didi (dlgnuti), pres. dignēm, like glboh, -e, gibao, glbla to imperf. ginuti, ginēm (Leskien 1914:469-70). 44 To which one may add OCS perf. act. ptcp. (i-)súx-uš- and Skt. pf. šu-šus-, ptcp. šu-šus-us-.
224
presents, in which case all presents in *-ne/0- to consonant-final roots in Slavic would be secondary for *-j e / 0 - presents.45 Thus both of the principal characterized present classes in Slavic, formed with the suffixes *-j e / 0 - and *-ne/0-, for the most part continue pre-PSl. presents in *-j e / 0 - < PE *yé/ó-. Depending on whether the root vowel was acute or non-acute (circumflex or short), these presents belonged to accentual paradigm a or b, respectively, i.e. according to whether the stress was retracted from the suffix by Hirt's Law or not (see above). Those presents whose infinitive stem was not suffixed with *-a- or *-nņ- formed root aorists in pre-PSl.: if the root ended in a vowel or diphthong, the aorist was regularized or "sigmatized" with the sigmatic aorist endings, e.g. zna-xv, zna; if the root ended in an obstruent, the aorist was thematized, e.g. i-súš-e, i-svx-g, dvii-e. The relationship among present-stem formation, accentual paradigm, and aorist formation is represented in the table below: latePSl., OCS
pre-P(B)Sl.
AP a/b c aA)
present suffixed *-j e / 0 *-ne/0simple *-e/0"
aorist thematic/"sigmatized" sigmatic thematic
postacc. unacc. postacc.
suffixed *-yé/(>simple *-e/0"
root sigmatic root
A glance at this table immediately reveals that as far as the formation of the aorist is concerned, the minority of unsuffixed, i.e. simple thematic presents which are underlyingly accented — i.e., belong to AP a or b and, according to the analysis proposed here, continue pre-PSl. postaccenting paradigms with columnar stress on the thematic vowel — behave exactly like suffixed presents in *-j e / 0 - and *-ne/0-, which are also accented: both form thematic aorists. For some of these eleven accented simple thematic presents, a thematic
45
Nasal presents to vowel-final roots in Slavic, on the other hand, appear to continue PIE nasal-infix presents. ORuss. lsg. krinju 'buy', *krin-je/0- <— *kri-n-e/0- (< PIE *kwrih2-; later replaced by kupiti <— OHG koufen <— Lat. caupō 'merchant') was probably thematized at an early date from nasal-infix pres. 3pl. *krin-anti < *kwri-n-h2-ónti <— *kwri-n-h2-énti (or directly from *krinanti < *kwrin-h2-énti after the merger of *a and *o?). Similarly, Russ. rinut' 'stream (intr., pf.)', rt'nut'sja 'rush, plunge', Cz. finouti se 'stream' < PS1. *ri-ne/0- has been thematized from *h3ri-n-H-énti (PS1. *i < *ī from pass. *h3riH-yé/ó-), cf. Ved. riņāti 'make stream', mp. riņīté, ríyate 'flow' (Tedesco 1948:348; LIV:271-2).
225
aorist is expected on morphological grounds: cf. the discussion above on the originally nasal-infix presents *s*d-«/0-, *b$d-e/0-, *nft-(j)c/0-, *l$2-c/0-I propose that based on the pattern of all suffixed thematic presents and the overwhelming majority of simple thematic presents to consonant-final roots, speakers of pre-PSl. abstracted a morphophonological pattern (or rule) according to which all postaccenting presents (> AP a/b), regardless of stem formation, took thematic aorists, and conversely that all unaccented presents (> AP c), which happened to be restricted to the simple thematic type, formed sigmatic aorists. This analogical process is represented below: AP
present
aorist
c
simple *-e/0-
sigmatic
but
a/b
simple *-e/0-
thematic
like
a/b
suffixed *-j e / 0 -
thematic
Thus, simple thematic presents such as *pád-e-tí, *s^d-e-tí, *moZ-é-tí, *jíd-é-tí which have a variety of origins, came to acquire a thematic aorist on the model of other presents which shared the same accentual specification, namely postaccenting or accented on an acute root-final vowel (with retraction by Hirt's Law). Seen in this light, the aorists *pade /0-, *moZ-e- ~ *mog-o-, etc. may instantly be recognized as accentually conditioned Slavic — or Balto-Slavic? — innovations. The preceding analysis of present formation, present accentuation, and aorist formation dramatically illustrates just how well (Balto-)Slavic has preserved the system of verbal morphology inherited from late (post-)PIE. Just as in classical IE languages such as Sanskrit, Avestan, or ancient Greek, so also in Slavic one finds the reflection of a fundamental contrast between "present" and "aorist" stems. Whether, as seems likely, these originally corresponded to "atelic" and "telic" aspectual meaning, the reconstructible pattern is clear: "present" roots formed uncharacterized simple thematic presents and characterized, here sigmatic, aorists; conversely, "aorist" roots formed uncharacterized root aorists, which have also been thematized in Slavic, and characterized presents, formed with the accented suffix *-yé/<5-- Comparative reconstruction leads to the conclusion that unsuffixed presents with root vowel *e, which continue (post-)PIE simple thematic presents, have unaccented stems in Proto-Slavic (and, given the evidence of Old Prussian,
226
Proto-Balto-Slavic), whereas suffixed presents, which in Slavic continue the accented suffix *-yé/ó-, are accented on the stem: either the thematic vowel, as in PIE, or a preceding long acute root vowel by Hirt's Law. Those few PS1. synchronic simple thematic presents which have other origins, and are underlyingly accented, follow the model of the likewise accented derived presents in taking thematic aorists. For those presents which historically contain (and synchronically may continue to be analyzed as containing) a nasal infix, a thematized root aorist is expected; for others, such as *mog-, *pad-, or *jld-, only the accentually conditioned morphological patterning argued above can account for the formation of the aorist. Here, as in the noun, we find that despite having undergone certain innovations, BaJto-Slavic, and especially Slavic, has faithfully retained many features of the PIE system of accent and morphology to a degree hitherto unrecognized by either Indo-Europeanists or Balto-Slavists.
227
Conclusion. The importance of theory for historical linguistics The previous pages have demonstrated some of the wide range of insights afforded by a (relatively) new theory of stress into the accentual system of an unattested protolanguage and its development in the ancient and modern descendants of that protolanguage. We have seen that not all the data are equally compatible with metrical bracket theory: the PIE proterokinetic and hysterokinetic nominal inflections in particular (§§2.2.2.4-5) pose some difficulty for analysis in terms of underlying accentuation of morphemes, and it is less than clear that Greek, with its extensive generalization of recessive stress (§§3.3.1,3.6), need be analyzed in such a fashion. Nevertheless, the brackets-and-edges framework of stress computation does offer a welcome and fresh perspective on often longstanding historical questions, even if it does not always provide definite answers. This is especially true for the Balto-Slavic accentual alternations discussed in chapters 5 and 6; all the relevant facts have been known for some time now, yet their proper evaluation has been hampered by the lack of a sufficiently abstract and powerful model of stress computation. Not all theoretical advances will find immediate applications in historical linguistics, to be sure; but accentology seems to provide a promising arena in which historical linguistics and phonologists can work together to mutual benefit.
228
Bibliography Adams, Douglas Q. 1988. Tocharian Historical Phonology and Morphology. (American Oriental Series, Vol. 71.) New Haven: American Oriental Society. —. 1999. A Dictionary of Tocharian B. (Leiden Studies in Indo-European 10.) Amsterdam: Rodopi. Alexander, Ronelle. 1975. Torlak Accentuation. (Slavistische Beiträge, Band 94.) Munich: Sagner. —. 1993. Remarks on the evolution of South Slavic prosodic systems. American Contributions to the Eleventh International Congress of Slavists, Bratislava 1993: Literature, Linguistics, Poetics, ed. by Robert A. Maguire and Alan Timberlake, 181201. Columbus: Slavica. Bach, Emmon. 1975. Long vowels and stress in Kwakiutl. Texas Linguistic Forum 2,919. Bammesberger, Alfred. 1985. Thematische und athematische Wurzelaoriste im Indogermanischen. Grammatische Kategorien: Funktion und Geschichte. Akten der VII. Fachtagung der Indogermanischen Gesellschaft, Berlin, 20.-25. Februar 1983, ed. by Bernfried Schlerath and Veronica Rittner, 71-8. Wiesbaden: Reichert. —. 1993. Zur Vorgeschichte von litauisch zinöti. Linguistica Baltica 2,83-90. Barton, Charles R. 1993. Hittite zinni-, tuwarni-, zig and related matters. Journal of the American Oriental Society 113:4, 551-61. Beekes, Robert S. P. 1972. The nominative of the hysterodynamic noun-inflection. ZVS 86:1,30-63. —. 1995. Comparative Indo-European Linguistics: An Introduction. Amsterdam/Philadelphia: Benjamins. Belić, Aleksandar (BeliC, Aleksandr"). 1909. Zame'tki po čakavskim" govoram". Otdelenije Russkago Jazyka i Slovesnosti Imperatorskoj Akademii Nauk". Sankt"Peterburg". Benveniste, Emile. 1935. La question du locatif singulier. Origines de la formation des noms en indo-europeen, Vol. I, 87-99. Paris: Maisonneuve. Bethin, Christina Y. 1994. On the phonology of the Neoštokavian accent retraction in Serbian and Croatian. Die Welt der Slaven 39:2,277-96. —. 1998. Slavic Prosody: Language Change and Phonological Theory. (Cambridge Studies in Linguistics, 86.) Cambridge/New York: Cambridge U. Press. Braune, Theodor Wilhelm. 1975. Althochdeutsche Grammatik. 13. Auflage, bearbeitet von Hans Eggers. Tübingen: Niemeyer. Browne, E. Wayles. 1993. Serbo-Croat. Comrie and Corbett (eds.) 1993, 306-87. — and James D. McCawley. 1965. Srpskohrvatski akcenat. Zbornik Matice Srpske za Filologiju i Lingvistiku 8, 147-51. Brugmann, Karl. 1911. Grundriß der vergleichenden Grammatik der indogermanischen Sprachen. Zweiter Band: Lehre von den Wortformen und ihrem Gebrauch. Zweiter Teil. Straßburg: Trübner. — and Berthold Delbrück. 1916. Grundriss der vergleichenden Grammatik der indogermanischen Sprachen. Zweiter Band: Lehre von den Wortformen und ihrem Gebrauch. Dritter Teil. Strassburg: Trübner.
229
Buck, Carl Darling. 1904. A Grammar of Oscan and Umbrian. Boston: Ginn. (Reprinted 1979, Hildesheim/New York: Olms.) Buckley, Eugene. 2000. Final extrametricality in Latin and Manam. University of Pennsylvania Working Papers in Linguistics 6.3, 269-77. Campbell, Alistair. 1959. Old English Grammar. Oxford: Clarendon. Cardona, George. 1960. The Indo-European Thematic Aorists. Ph.D. dissertation, Yale University. Chantraine, Pierre. 1942. Grammaire homérique (phonétique et morphologie). (Collection de Philologie Classique I.) Paris: Klincksieck. Chomsky, Noam and Morris Halle. 1968. The Sound Pattern of English. New York: Harper and Row. Collinge, N. E. 1985. The Laws of Indo-European. (Amsterdam Studies in the Theory and History of Linguistic Science, Vol. 35.) Amsterdam and Philadelphia: Benjamins. Comrie, Bernard and Greville G. Corbett, eds. 1993. The Slavonic Languages. London/New York: Routledge. Cowgill, Warren. 1957. The Indo-European Long-Vowel Preterits. Ph.D. dissertation, Yale University. —. 1960. Greek ou and Armenian of. Language 36:3, 347-50. Cowgill, Warren. 1968. The first person singular medio-passive of Indo-Iranian. Heesterman et al. (eds.) 1968,24-31. —. 1970. The nominative plural and preterit singular of the active participles in Baltic. Magner and Schmalstieg (eds.) 1970, 23-37. —. 1979. Anatolian /»'-conjugation and Indo-European perfect: instalment II. Hethitisch und Indogermanisch, ed. by Erich Neu and Wolfgang Meid, 25-39. Innsbruck: Innsbrucker Beiträge zur Sprachwissenschaft (Band 25). —. 1985. The personal endings of thematic verbs in Indo-European. Grammatische Kategorien: Funktion und Geschichte. Akten der VII. Fachtagung der Indogermanischen Gesellschaft, Berlin, 20.-25. Februar 1983, ed. by Bernfried Schlerath and Veronica Rittner, 99-108. Wiesbaden: Reichert. —. 1986. Einleitung. Indogermanische Grammatik: Band I, 1. Halbband. Ed. by Manfred Mayrhofer; trans, and bibl. by Alfred Bammesberger and Martin Peters. Heidelberg: Winter. Crist, Sean. 2001. Conspiracy in Historical Phonology. Ph.D. dissertation, University of Pennsylvania. Debrunner, Albert and Jacob Wackernagel. 1930. Altindische Grammatik von Jacob Wackernagel. Band III: Nominalflexion — Zahlwort — Pronomen. Göttingen: Vandenhoeck und Ruprecht. Delbrück, Berthold. 1897. Grundriss der vergleichenden Grammatik der indogermanischen Sprachen. Vierter Band: Vergleichende Syntax der indogermanischen Sprachen. Zweiter Teil. Strassburg: Trübner. Derksen, Rick. 1991. An introduction to the history of Lithuanian accentuation. Studies in Slavic and General Linguistics 16 (Studies in West Slavic and Baltic Linguistics), 4584. Dolobko, Milij. 1927. Nod' - nodes ', ôsen ' - osenés ', zimâ - zimiis ', léto - létos '. Slavia 5:4,678-717. Dybo, Vladmir A. 1961. Udarenie slavjanskogo glagola i formy praslavjanskogo aorista. Kratkie soobSäenija Instituta Slavjanovedenija 30, 33-8. 230
—. 1962. O rekonstrukcii udarenija v praslavjanskom glagole. Voprosy Slavjanskogo Jazykoznanija 6, 3-27. —. 1981. Slavjanskaja akcentologija: opyt rekonstrukciji sistemy akcentnyx paradigm v praslavjanskom. (Akademija Nauk SSSR, Institut Slavjanovedenija i Balkanistiki.) Moscow: Izdatel'stvo «Nauka». —, G. I. Zamjatina, S. L. Nikolaev. 1990. Osnovy slavjanskoj akcentologii. Moscow: Izdatel'stvo Nauka. Ebeling, C. L. 1967. Historical laws of Slavic accentuation. To Honor Roman Jakobson: Essays on the Occasion of his Seventieth Birthday, 11 October 1966, Vol. I, 577-93. (Janua Linguarum, Series Major 31.) The Hague/Paris: Mouton. Eichner, Heiner. 1972. Die Etymologie von heth. mehur. Münchener Studien zur Sprachwissenschaft 31, 53-107. —. 1985. Das Problem des Ansatzes eines urindogermanischen Numerus 'Kollektiv' ('Komprehensiv'). Grammatische Kategorien: Funktion und Geschichte. Akten der VII. Fachtagung der Indogermanischen Gesellschaft, Berlin, 20.-25. Februar 1983, ed. by Bernfriend Schlerath and Veronica Rittner, 134-69. Wiesbaden: Reichert. Endzelins, Janis. 1899. Über den lettischen Silbenakzent. Beiträge zur Kunde der indogermanischen Sprachen 25, 259-74. —. 1913. Weiteres zu den lettischen Intonationen. IF 23, 104-118. —. 1922. Des intonations lettes. Revue des Etudes Slaves 2,56-68. Eska, Joseph F. 1994. Rethinking the evolution of Celtic constituent configuration. MSS 55,7-39. Euler, Wolfram. 1979. Indoiranisch-griechische Gemeinsamkeiten der Nominalbildung und deren indogermanische Grundlagen. (Innsbrucker Beiträge zur Sprachwissenschaft, Band 30.) Innsbruck: Institut für Sprachwissenschaft. Garde, Paul. 1961. Reflexions sur les differences phonětiques entre les langues slaves. Word 17, 34-62. —. 1966. Fonctions des oppositions tonales dans les langues slaves du sud. BSL 61:1, 42-56. —. 1973. Le paradigme accentuel oxyton est-il slave commun? Revue des Etudes Slaves 49, 159-71. —. 1976a. Histoire de Vaccentuation slave. (Collection de manuels de l'Institut d'Ĕtudes Slaves, 7.) Paris: Institut d'Ětudes Slaves. —. 1976b. Neutralization of tone in Common Slavic. Slavic Linguistics and Language Teaching, ed. by Thomas F. Magner, 1-19. Cambridge, MA: Slavica. Garrett, Andrew. 1990. The origin of NP split ergativity. Language 66:2, 261-96. Goldsmith, John. 1995. The Handbook of Phonological Theory. Oxford and Cambridge, MA: Blackwell. Hackstein, Olav. 1993. Osttocharische Reflexe grundsprachlicher Präsensbildungen von idg. *gnehj- '(er)kennen'. Indogermanica et Italica: Festschrift für Helmut Rix zum 65. Geburtstag, ed. by Gerhard Meiser, 148-58. (Innsbrucker Beiträge zur Sprachwissenschaft, Band 72.) Innsbruck: Institut für Sprachwissenschaft. —. 1995. Untersuchungen zu den sigmatischen Präsensstammbildungen des Tocharischen. (HS, Ergänzungsheft 38.) Göttingen: Vandenhoeck und Ruprecht. Hajnal, Ivo. 1994a. Das Brugmansche Gesetz in diachroner Sicht und seine Gültigkeit innerhalb der arischen a-Stämme. HS 107, 194-221.
231
—.
1994b. Die lykischen a-Stämme: zum Werdegang einer Nominalklasse. In Honorem Holger Pedersen: Kolloquium der Indogermanischen Gesellschaft vom 25. bis 28. März 1993 in Kopenhagen, ed. by Jens Elmegard Rasmussen, 135-71. Wiesbaden: Reichert. Halle, Morris. 1959. The Sound Pattern of Russian. The Hague: Mouton. —. 1962. Phonology in generative grammar. Word 18, 54-72. —. 1973. The accentuation of Russian words. Language 49, 3\2-4S. —. 1997. On stress and accent in Indo-European. Language 73:2,275-313. —. 2001. On accent, stress and quantity in West Slavic. Lingua 111, 791-810. — and William Idsardi. 1995. General properties of stress and metrical structure. Goldsmith (ed.) 1995,403-43. — and Jean-Roger Vergnaud. 1987. An Essay on Stress. Cambridge, MA: MIT Press. Hammond, Michael. 1999. The Phonology of English: A Prosodie Optimality-Theoretic Approach. Oxford/New York: Oxford U. Press. Hamp, Eric. 1982/3. Indo-European substantives in *-mo- and *-mä. ZVS 96:2, 171-7. HarSarson, Jon Axel. 1993. Studien zum urindogermanischen Wurzelaorist und dessen Vertretung im Indoiranischen und Griechischen. Innsbrucker Beiträge zur Sprachwissenschaft, Band 74.) Innsbruck: Institut für Sprachwissenschaft. Hayes, Bruce. 1981. A Metrical Theory of Stress Rules. Ph.D. dissertation, MIT, 1980. (Revised version distributed by Indiana University Linguistics Club, Bloomington, IN, and published by Garland Press, New York, 1985.) —. 1987. A revised parametric metrical theory. Proceedings of the North East Linguistic Society 17, ed. by J. McDonough and B. Plunkett, 274-89. Amherst, MA: Graduate Linguistics Student Association, U. of Massachusetts, Amherst. —. 1995. Metrical Stress Theory: Principles and Case Studies. Chicago: U. of Chicago Press. Heesterman, J. C , G. H. Schokker, and V. I. Subramoniam. 1968. Pratidâ^nam: Indian, Iranian and Indo-European Studies Presented to F.B.J. Kuiper on His Sixtieth Birthday. The Hague: Mouton. Hinrichs, Jan Paul. 1985. Zum Akzent im Mittelbulgarischen. (Studies in Slavic and General Linguistics, Vol. 6.) Amsterdam: Rodopi. Hirt, Herman. 1895. Der indogermanische Akzent. Ein Handbuch. Straßburg: Trübner. Hoenigswald, Henry M. 1960. Language Change and Linguistic Reconstruction. Chicago: U. of Chicago Press. —. 1987. Aiei and the prehistory of Greek noun accentuation. Studies in memory of Warren Cowgill (1929-1985): Papers from the Fourth East Coast Indo-European Conference, Cornell University, June 6-9, 1985, ed. by Calvert Watkins, 51-3. Berlin and New York: Walter de Gruyter. —. 1997. Some archaisms in the Iliad. Historical, Indo-European, and Lexicographical Studies: A Festschrift for Ladislav Zgusta on the Occasion of his 70th Birthday, ed. by Hans Henrich Hock, 79-85. (Trends in Linguistics, Studies and Monographs 90.) Berlin/New York: Mouton de Gruyter. —. 1998. Eu, èû and the accent of non-thematic neuters. Mir Curad: Studies in Honor of Calvert Watkins, ed. by Jay Jasanoff, H. Craig Melchert, and Lisi Olivier. 271-4. (Innsbrucker Beiträge zur Sprachwissenschaft, Band 92.) Innsbruck: Institut für Sprachwissenschaft. Hoffmann, Karl. 1955. Vedisch "gâmati". MSS1, 89-92. 232
—. 1965. Idg. *sneigyh. MSS 18, 13-28. (Reprinted in Hoffmann 1976,442-54.) . 1967. Der vedische Prekativtyp yes.am, jes.ma. MSS 20, 25-37. (Reprinted in Hoffmann 1976,465-74.) —. 1968. Zum Optativ des indogermanischen Wurzelaorists. Heesterman et al. (eds.) 1968,3-8. —. 1976. Avest. vatjhuuqm. Aufsätze zur Indoiranistik, ed. by Johanna Narten, Band 2, 593-604. —. 1986. Avestisch S. Studia Grammatica Iranica: Festschrift für Helmut Humbach, hrsg. von Rüdiger Schmitt und Prods Oktor Skjaerv0, 163-83. München: Kitzinger. — and Bernhard Forssman. 1996. Avestische Laut- und Flexionslehre. (Innsbrucker Beiträge zur Sprachwissenschaft, Band 84.) Innsbruck: Institut für Sprachwissenschaft. Houtzagers, H. P. 1982. Accentuation in a few dialects of the island of Cres. South Slavic and Balkan Linguistics, ed. by A. A. Barentsen, R. Sprenger, and M. G. M. Tielemans, 117-29. (Studies in Slavic and General Linguistics, Vol. 2.) Amsterdam: Rodopi. —. 1985. The Cakavian Dialect ofOrlec on the Island of Cres. (Studies in Slavic and General Linguistics, Vol. 5.) Amsterdam: Rodopi. Hrozny\ Bedf ich. 1917. Die Sprache der Hethiter: ihr Bau und ihre Zugehörigkeit zum indogermanischen Sprachstamm. Ein Entzifferungsversuch. (Boghazköi-Studien 1.) Leipzig: Hinrichs. Idsardi, William. 1992. The Computation of Prosody. Ph.D. dissertation, MIT. IlliC-SvityC, Vladislav M. 1963. Imennaja akcentuacija v baltijskom i slavjanskom. Moscow: Institut Slavjanovedenija, Akademija Nauk SSSR. (English edition: Nominal Accentuation in Baltic and Slavic. Translated by Richard L. Leed and Ronald F.Feldstein. Cambridge, MA and London: MIT Press, 1979.) Inkelas, Sharon and Draga Zee. 1988. Serbo-Croatian pitch accent: the interaction of tone, stress, and intonation. Language 64:2, 227-48. Ivic, Pavle. 1958. Die serbokroatischen Dialekte: ihre Struktur und Entwicklung. The Hague: Mouton. —. 1976. Serbocroatian accentuation: facts and interpretation. Slavic Linguistics and Language Teaching, ed. by Thomas F. Magner, 34-43. Cambridge, MA: Slavica. Jakobson, Roman. 1963. Opyt fonologiCeskogo podxoda k istoriCeskim vopr.osam slavjanskoj akcentologii. American Contributions to the 5th International Congress of Slavists, Sofia 1963, 153-78. The Hague: Mouton. Jamison, Stephanie W. 1983a. Function and Form in the -âya-Formation of the Rig Veda and Atharva Veda. (ZVS, Ergänzungsheft 31.) Göttingen: Vandenhoeck und Ruprecht. —. 1991. A cart, an ox, and the perfect participle in Vedic. MSS 52,77-100. Janda, Laura A. 1994. The spread of athematic lsg -m in the major West Slavic languages. Slavic and East European Journal 38:1,90-119. —. 1996. Back from the Brink: A Study of How Relic Forms in Languages Serve as Source Material for Analogical Extension. (LINCOM Studies in Slavic Linguistics, 1.) Munich: LINCOM Europa. Jasanoff, Jay H. 1976. Gr. än
233
—. 1979. The position of the ^/-conjugation. Hethitisch und Indogermanisch, ed. by Wolfgang Meid and Erich Neu, 79-90. (Innsbrucker Beiträge zur Sprachwissenschaft, Band 25.) Innsbruck: Institut für Sprachwissenschaft. —. 1983. A rule of final syllables in Slavic. Journal of Indo-European Studies 11, 13949. —. 1988a. PEE *gnë- 'recognize, know'. Die Laryngaltheorie und die Rekonstruktion des indogermanischen Laut- und Formensystems, ed. by Alfred Bammesberger, 22739. Heidelberg: Winter. —. 1988b. The sigmatic aorist in Tocharian and Indo-European. TIES 2, 52-76. —. 1989. Old Irish &?'woman'. Ériu 40, 135-41. —. 1994. Aspects of the internal history of the PIE verbal system. Früh-, Mittel-, Spätindogermanisch: Akten der IX. Fachtagung der Indogermanischen Gesellschaft vom 5. bis 9. Oktober 1992 in Zürich, ed. by George E. Dunkel, Gisela Meyer, Salvatore Scarlata, and Christian Seidl, 149-68. Wiesbaden: Reichert. —. 1997a. Gathic Avestan ciköitaraS. Sound Law and Analogy: Papers in Honor of Robert S.P. Beekes on the Occasion of His 60th Birthday, ed. by Alexander Lubotsky, 119-30. (Leiden Studies in Indo-European, Vol. 9.) Amsterdam and Atlanta: Rodopi. —. 1997b. The thematic conjugation revisited. MirCurad: Studies in Honor of Calvert Watkins, ed. by Jay Jasanoff, H. Craig Melchert, and Lisi Olivier, 301-16. (Innsbrucker Beiträge zur Sprachwissenschaft, Band 92.) Innsbruck: Institut für Sprachwissenschaft. Kager, René. 1995. The metrical theory of word stress. Goldsmith (ed.) 1995, 367-402. Kazlauskas, J. 1970. On the Balto-SIavic dative plural and dual. Magner and Schmalsties (eds.) 1970,87-91. Kim, Ronald I. 1999. The development of labiovelars in Tocharian: a closer look. TIES 8, 139-87. —. 2000a. Reexamining the prehistory of Tocharian B'ewe'. TIES 9, 37-43. —. 2000b. 'To drink' in Anatolian, Tocharian, and Proto-Indo-European. HS 113, 15170. —. 2001a. The distribution of the Old Irish infixed pronouns, Cowgill's particle, and the syntactic evolution of Insular Celtic. Indo-European Perspectives: Papers from the 18th East Coast Indo-European Conference, University of Texas at Austin, May 1999, ed. by Mark Southern and Sara Kimball, 151-74. (Journal of Indo-European Studies Monograph Series, Vol. 41.) Washington, DC: Institute for the Study of Man. —. 2001b. Tocharian B sem = Latin vënitl Szemerényi's Law and *e in PIE root aorists. MSS 61, 119-47. —. Forthcoming. Uncovering the prehistory of the Tocharian class II preterite. To appear in HS. —. In progress a. Now it's there, now it's not: secondary and non-underlying jers in Russian and Serbo-Croatian accentual alternations. —. In progress b. Akcenat srpskohrvatskog aorista i poreklo tematiCkog i novoradenog aorista na slovenskom. [The accent of the Serbo-Croatian aorist, and the origin of the thematic and innovative aorist in Slavic] Kimball. Sara. 1987. *H3 in Anatolian. Festschrift for Henry Hoenigswald, ed. by George Cardona and Norman H. Zide, 185-92. Tübingen: Narr. —. 1999. Hittite Historical Phonology. (Innsbrucker Beiträge zur Sprachwissenschaft. Band 95.) Innsbruck: Institut für Sprachwissenschaft. 234
Kiparsky, Paul. 1967. À propos de l'histoire de l'accentuation grecque. Langages 8,7393. —. 1973. The inflectional accent in Indo-European. Language 49:4,794-849. —. 1979. Metrical structure assignment is cyclic. Linguistic Inquiry 10,421-41.
Klemensiewicz, Z., T. Lehr-Spt-awiński, and S. Urbańczyk. 1965. Gramatyka historyczna języka polskiego. Warsaw: Państwowe Wydawnictwo Naukowe. Koch, Christoph. 1976. Slavische v-Präsentien? Die Welt der Slaven 21:1, 121-42. —. 1984. Aksl. perçprëti. Ein bislang übersehener Konjugationstyp des Slavischen. MSS 43, 137-74. —. 1985. Zur Flexion der aksl. Paradigmen mit Verbalstamm auf -ov u-. MSS 45, 127-33. —. 1986. Der Vokalwechsel im Paradigma ksi. szpç suti. MSS 47, 109-19. Kortlandt, F. H. H. 1974a. Old Prussian accentuation. ZVS 88, 299-306. —. 1974b. On the history of Baltic accentuation. Historical Linguistics: Proceedings of the First International Conference on Historical Linguistics, Edinburg 2nd - 7th September 1973. Vol. I: Theory and Description in Phonology, ed. by J. M. Anderson and C. Jones, 295-309. (North-Holland Linguistic Series, 12b.) Amsterdam: North Holland. —. 1975. Slavic Accentuation: A Study in Relative Chronology. Amsterdam: Lisse. —. 1978. On the history of Slavic accentuation. ZVS 92,269-81. —. 1988. The laryngeal theory and Slavic accentuation. Die Laryngaltheorie und die Rekonstruktion des indogermanischen Laut- und Formensystems, ed. by Alfred Bammesberger, 299-311. Heidelberg: Winter. Krause, Wolfgang and Werner Thomas. 1960. Tocharisches Elementarbuch. Band I. Grammatik. Heidelberg: Winter. Kroch, Anthony. Reflexes of grammar in patterns of language change. Language Variation and Change 1:3, 199-244. Kuiper, F. B. J. 1942. Notes on Vedic Noun-Inflexion. (Mededeelingen der Nederlandsche Akademie van Wetenschappen, Afdeeling Letterkunde, Nieuwe Reeks, Deel 5, No. 4.) Amsterdam: Maatschappij. Kurytowicz, Jerzy. 1958. L'accentuation des langues indo-européennes. 2nd ed. Wroclaw: Polska Akademia Nauk. —. 1962. O nêkotoryx fikcijax sravnitel'nogo jazykoznanija. Voprosy Jazvkoznanija 1962:1,31-6. —. 1968. Indogermanische Grammatik. Band II: Akzent/Ablaut. Heidelberg: Winter. Labov, William. 1994. Principles of Linguistic Change. Vol.1: Internal Factors. Oxford and Cambridge, MA: Blackwell. —. 2001. Principles of Linguistic Change. Vol. II: Social Factors. Oxford and Cambridge, MA: Blackwell. Langston, Keith. 1997. Pitch accent in Croatian and Serbian: towards an autosegmental analysis. Journal of Slavic Linguistics 5:1, 80-116. Lehfeldt, Werner. 1993. Einführung in die morphologische Konzeption der slavischen Akzentologie. (Vorträge und Abhandlungen zur Slavistik, Band 23.) München: Sagner. Lejeune, Michel. 1972. Phonétique historique du mycénien et du grec ancien. Paris: Klincksieck.
235
Leskien, August. 1881. Die Quantitätsverhältnisse im Auslaut des Litauischen. Archiv fürslavische Philologie 5, 188-90. . 1914. Grammatik der serbokroatischen Sprache. I.Teil: Lautlehre, Stammbildung, Formenlehre. Heidelberg: Winter. —. 1919. Grammatik der altbulgarischen (altkirchenslavischen) Sprache. 2. und 3. Aufgabe. Heidelberg: Winter. —. 1962. Handbuch der altbulgarischen (altkirchenslavischen) Sprache. Grammatik, Texte, Glossar. 8., verbesserte und erweiterte Auflage. Heidelberg: Winter. Liberman, Mark. 1975. The Intonational System of English. Ph.D. dissertation, MIT. — and Alan Prince. 1977. On stress and linguistic rhythm. Linguistic Inquiry 8, 249336. Lindeman, Fredrik Otto. 1972. Zur Akzentuation in den griech. Pradigmata von 56pu und yövu. Norwegian Journal of Linguistics (NTS) 26, 217-31. LIV: Lexikon der indogermanischen Verben: Die Wurzeln und ihre Primärstammbildungen. Unter Leitung von Helmut Rix und der Mitarbeit vieler anderer, bearbeitet von Martin Kümmel, Thomas Zehnder, Reiner Lipp, Brigitte Schirmer. Wiesbaden: Reichert, 1998. Lorentz, Friedrich. 1903. Slovinzische Grammatik. St. Petersburg: Academie Imperiale des Sciences. Lubotsky, Alexander. 1988. The System of Nominal Accentuation in Sanskrit and ProtoIndo-European. (Memors of the Kern Institute, No. 4.) Leiden: E.J. Brill. Liihr, Rosemarie. 1990. Reste der athematischen Konjugation in den germanischen Sprachen. Das Germanische und die Rekonstruktion der indogermanischen Grundsprache, ed. by Jürgen Untermann and Bela Brogyanyi, 25-90. Amsterdam/Philadelphia: Benjamins. Macdonell, Arthur Anthony. 1910. Vedic Grammar. (Grundriß der indo-arischen Philologie und Altertumskunde, 1. Band, 4. Heft.) Straßburg: Trübner. Magner, Thomas F. and William R. Schmalstieg. 1970. Baltic Linguistics. University Park, PA and London: Pennsylvania State U. Press. Mallory, J. P. 1989. In Search of the Indo-Europeans: Language, Archaeology, and Myth. London: Thames and Hudson. Marggraf, Wolf-Jürgen. 1970. Untersuchungen zum Akzent in Tocharisch B. Dissertation, Christian-Albrechts-Universität, Kiel. Mayrhofer, Manfred. 1970. Germano-Iranica. ZVS 84, 224-30. —. 1981. Laryngalreflexe im Indo-Iranischen. ZPSK 34,427-38. —. 1986. Indogermanische Grammatik. Band I, 2. Halbband: Lautlehre (Segmentale Phonologie des Indogermanischen). Heidelberg: Winter. —. 1989. Vorgeschichte der iranischen Sprachen; Uriranisch. Compendium Linguarum Iranicarum, ed. by Rüdiger Schmitt, 4-24. Wiesbaden: Reichert. McCarthy, John and Alan Prince. 1993. Prosodic Morphology I: constraint interaction and satisfaction. Ms., University of Massachusetts, Amherst and Rutgers University. —. 1995. Prosodic morphology. Goldsmith (ed.) 1995, 318-66. McCone, Kim. 1986. From Indo-European to Old Irish: conservation and innovation in the verbal system. Proceedings of the Seventh International Congress of Celtic Studies, ed. by D. Ellis Evans, J. G. Griffith, and E. M. Jope, 222-66. Oxford: Oxford U. Press.
236
—. 1991. The Indo-European Origins of the Old Irish Nasal Presents, Subjunctives and Futures. (Innsbrucker Beiträge zur Sprachwissenschaft, Band 66.) Innsbruck: Institut für Sprachwissenschaft. —. 1994. Zum Ablaut der keltischen r-Stämme. Rasmussen (ed.) 1994, 275-84. Meillet, Antoine (Mel'e, Antuan). 1902. O nčkotoryx" anomalijax" udarenija v" slavjanskix" imenax". Russkij Filologičeskij Věstnik" 48:3-4, 193-200. —. 1902-5. Etudes sur l'etymologie et le vocabulaire du vieux slave. (Bibliothěque de l'Ěcole des Hautes Etudes, IVe section: Sciences historiques et philologiques, fasc. 139.) Paris: Bouillon. Melchert, H. Craig. 1984. Studies in Hittite Historical Phonology. (ZVS Ergänzungsheft, Nr. 32.) Göttingen: Vandenhoeck und Ruprecht. —. 1987a. PIE velars in Luvian. Studies in memory of Warren Cowgill (1929-1985): Papers from the Fourth East Coast Indo-European Conference, Cornell University, June 6-9, 1985, ed. by Calvert Watkins, 182-204. Berlin and New York: Walter de Gruyter. —. 1987b. Reflexes of *h3 in Anatolian. Die Sprache 33, 19-28. —. 1988. Word-final -r in Hittite. A Linguistic Happening in Memory of Ben Schwarz, ed. by Yoėl Arbeitman, 215-34. Louvain: Peeters. —. 1989. New Luvo-Lycian isoglosses. HS 102, 23-45. —. 1992. The middle voice in Lycian. HS 105, 189-99. —. 1994a. Anatolian Historical Phonology. (Leiden Studies in Indo-European 5.) Amsterdam: Rodopi. —. 1994b. 'Cop's law'in Common Anatolian. Rasmussen (ed.) 1994, 297-306. —. 1994c. The feminine gender in Anatolian. Früh-, Mittel-, Spätindogermanisch: Akten der IX. Fachtagung der Indogermanischen Gesellschaft vom 5. bis 9. Oktober 1992 in Zürich, ed. by George E. Dunkel, Gisela Meyer, Salvatore Scarlata, and Christian Seidl, 231-44. Wiesbaden: Reichert. —. 1997. Denominative verbs in Anatolian. Studies in Honor of Jaan Puhvel. Part One: Ancient Languages and Philology, ed. by Dorothy Disterheft, Martin Huld and John Greppin, 131-8. (JIES Monograph 20.) Washington, D.C.: Institute for the Study of Man. —. 2000. Review of Hackstein 1995. TIES 9, 145-7. Mikkola, J. J. 1950. Urslavische Grammatik: Einführung in das vergleichende Studium der slavischen Sprachen. III. Teil: Formenlehre. Miletič, Ljubomir. 1903. Das Ostbulgarische. (Schriften der Balkankommission, Linguistische Abteilung, 2.) Vienna: Höldner. Morpurgo-Davies, Anna. 1970. Epigraphical-cpi. Glotta 47,46-54. Morpurgo Davies, Anna. 1976. The -essi datives, Aeolic -ss-, and the Lesbian poets. Morpurgo Davies and Meid (eds.) 1976, 181-97. —. 1982/3. Dentals, rhotacism and verbal endings in the Luwian languages. ZVS 96, 245-70. — and Wolfgang Meid, eds. 1976. Studies in Greek, Italic, and Indo-European Linguistics Offered to Leonard R. Palmer on the Occasion of His Seventieth Birthday, June 5, 1976. Innsbruck: Innsbrucker Beiträge zur Sprachwissenschaft (Band 16). Narten, Johanna. 1964. Die sigmatischen Aoriste im Veda. Wiesbaden: Harrassowitz. —. 1968. Zum "proterodynamischen" Wurzelpräsens. Heesterman et al. (eds.) 1968, 919. 237
Neu, Erich. 1980. Studien zum endungslosen "Lokativ" des Hethitischen. (Innsbrucker Beiträge zur Sprachwissenschaft, Vorträge und Kleinere Schriften 23.) Innsbrusck: Institut für Sprachwissenschaft. Neweklowsky, Gerhard. 1986. Die Akzentuation der o-,jo- und /-stämmigen Substantive in der Sprache dca Protopopen Avvakum. Die Welt der Slaven 31:1, 160-73. Noyer, Rolf. 1997. Attic Greek accentuation and intermediate derivational representations. Derivations and Constraints in Phonology, ed. by Iggy Roca, 501-27 (Ch. 16). Oxford: Clarendon Press; New York: Oxford U. Press. Olsen, Birgit Anette. 1988. The Proto-Indo-European Instrument Noun Suffix *-tlom and Its Variants. Copenhagen: Commissioner, Munksgàrd. Pedersen, Holger. 1925. Le groupement des dialectes indo-européens. (Kongelige Danske Videnskabernes Selskab, Historisk-filologiske Meddelelser 11, 3.) Copenhagen. —. 1926. La cinquième déclinaison latine. (Kongelige Danske Videnskabernes Selskab, Historisk-filologiske Meddelelser 11,5.) Copenhagen. Pinker, Steven. 1999. Words and Rules: The Basic Ingredients of Language. New York: Basic Books. Pinault, Georges-Jean. 1989. Introduction au tokharien. Lalies 7,3-224. Prince, Alan. 1983. Relating to the grid. Linguistic Inquiry 14, 19-100. Rasmussen, Jens Elmegârd. 1978. Zur Morphophonematik des Urindogermanischen. Die Erklärung qualitativer, quantitativer, und akzentueller Alternationen durch vorindogermanische Lautgesetze. Collectanea Indoeuropaea I, ed. by Bojan Cop, Varja Cvetko, and Jens Elmegârd Rasmussen, 59-143. (Series Comparativa, Vol. 3.) Ljubljana: Univerza, Filozofska fakulteta, Oddelek za primerjalno jezikoslovje in orientalistiko. —, ed. 1994. In Honorem Holger Pedersen: Kolloquium der Indogermanischen Gesellschaft vom 25. bis 28. März 1993 in Kopenhagen. Wiesbaden: Reichert Reinhart, Johannes. 1988. Eine Innovation bei der Aoristbildung im KroatischGlagolitischen. Zeitschrift für Slavische Philologie 48:2, 298-303. —. 1992. Die Geschichte des slawischen sigmatischen Aorists. Rekonstruktion und Relative Chronologie. Akten der VIII. Fachtagung der Indogermanischen Gesellschaft, Leiden, 31. August-4. September 1987, ed. by Robert Beekes, Alexander Lubotsky, and Jos Weitenberg, 367-81. (Innsbrucker Beiträge zur Sprachwissenschaft, Band 65.) Innsbruck: Institut für Sprachwissenschaft. ReSetar, Milan. 1931a. Les changements d'accent des substantifs en serbo-croate: les masculins. Revue des Études Slaves 11:1/2, 12-40. —. 1931b. Les changements d'accent des substantifs en serbo-croate: neutres et féminins. Revue des Études Slaves 11:3/4, 187-200. —. 1937. Les changements d'accent en serbo-croate: le verbe. Revue des Études Slaves 17,46-70. Ringe, Donald A., Jr. 1987. On the prehistory of Tocharian B accent. Studies in Memory of Warren Cowgill (1929-1985): Papers from the Fourth East Coast Indo-European Conference, Cornell University, June 6-9, 1985, ed. by Calvert Watkins, 254-69. Berlin and New York: Walter de Gruyter. —. 1990. Evidence for the position of Tocharian in the Indo-European family? Die Sprache 34, 59-123. —. 1991. Laryngeals and Sievers' Law in Tocharian. MSS 52, 137-68. 238
—. 1996. On the Chronology of Sound Changes in Tocharian. Vol.1: From ProtoIndo-European to Proto-Tocharian. (American Oriental Series, Vol. 80.) New Haven: American Oriental Society. —. 2000. Tocharian class II presents and subjunctives and the reconstruction of the ProtoIndo-European verb. TIES 9, 121-42. —. 2002. Syncopated present indicative forms in Old English. Verba et Litterae: Explorations in Germanic Languages and German Literature, ed. by Alfred R. Wedel and Hans-Jörg Busch, 125-56. (Festschrift for Albert L. Lloyd.) Newark, DE: Linguatext. —, Tandy Warnow, Ann Taylor, Alexander Michailov, and Libby Levison. 1998. Computational cladistics and the position of Tocharian. The Bronze Age and Early Iron Age Peoples of Eastern Central Asia, ed. by Victor H. Mair, Vol. I, 391-414. Washington, D.C.: Institute for the Study of Man in collaboration with the U. of Pennsylvania Museum Publications. Risch, Ernst. 1975. Remarques sur l'accent du grec ancien. Melanges linguistiques offerts à Emile Benveniste, 471-9. (Société de Linguistique de Paris, Collection Linguistique 70.) Louvain: Peeters. Rix, Helmut. 1976. Historische Grammatik des Griechischen. Darmstadt: Wissenschaftliche Buchgesellschaft. —. 1990. Review of Watkins (ed.) 1987. Kratylos 35,41-8. —. 1995. Einige lateinische Präsensstammbildungen zu Set-Wurzeln. Kuryiowicz Memorial Volume, ed. by Walter Smoczyriski, Vol. I, 399-408. (Analecta Indoeuropaea Cracoviensia, 2.) Cracow. —. 1999. Schwach charakterisierte lateinische Präsensstämme zu Set-Wurzeln mit Vollstufe I. Compositiones Indogermanicae in Honorem Jochem Schindler, hrsg. von H. Eichner und H. C. Luschützky unter redaktioneller Mitwirkung von V. Sadovski, 515-35. Prague: Enigma. de Saussure, Ferdinand. 1894. À propos de l'accentuation lituanienne. Mémoires de la Société de Linguistique 8,425-46. (Reprinted in de Saussure 1922,490-512.) —. 1896. Accentuation lituanienne. IF Anzeiger 6, 157'-66. (Reprinted in de Saussure 1922, 526-38.) —. 1922. Recueil des publications scientifiques de Ferdinand de Saussure. Lausanne/Genève/ Neuchâtel/Verrey/Montreux/Berne: Payot. Scatton, Ernest A. 1993. Bulgarian. Comrie and Corbett (eds.) 1993, 188-248. Schindler, Jochem. 1966. Bemerkungen zum idg. Wort für „Schlaf '. Die Sprache 12, 67-76. —. 1967. Zu hethitisch nekuz. ZVS 81, 290-303. —. 1969. Die idg. Wörter für "Vogel" und "Ei". Die Sprache 15, 144-67. —. 1972a. Das Wurzelnomen im Arischen und Griechischen. Dissertation, Würzburg. —. 1972b. L'apophonie des nom-racines indo-européens. BSL 67, 31-8. —. 1975a. L'apophonie des thèmes indo-européens en -r/n. BSL7Q, 1-10. —. 1975b. Zum Ablaut der neutralen s-Stämme des Indogermanischen. Flexion und Wortbildung: Akten der V. Fachtagung der Indogermanischen Gesellschaft, Regensburg, 9.-14. September 1975, ed. by Helmut Rix, 259-67. Wiesbaden: Dr. Ludwig Reichert Verlag. —. 1976. On the Greek type iTTireûç. Morpurgo Davies and Meid (eds.) 1976, 349-52. —. 1977. A thorny problem. Die Sprache 23,25-35. 239
. 1994. Alte und neue Fragen zum indogermanischen Nomen. Rasmussen (ed.) 1994, 397-400. Schmidt, Johannes. 1889. Die Pluralbildungen der indogermanischen Neutra. Weimar: Böhlau. Schmidt, Klaus T. 1985. Beiträge zur Kenntnis der tocharischen Verbalmorphologie. Grammatische Kategorien: Funktion und Geschichte. Akten der VII. Fachtagung der Indogermanischen Gesellschaft, Berlin, 20.-25. Februar 1983, ed. by Bernfried Schlerath and Veronica Rittner, 424-34. Wiesbaden: Reichert. —. 1992. Archaismen des Tocharischen und ihre Bedeutung für Fragen der Rekonstruktion und der Ausgliederung. Rekonstruktion und Relative Chronologie. Akten der VIII. Fachtagung der Indogermanischen Gesellschaft, Leiden, 31. August-4. September 1987, ed. by Robert Beekes, Alexander Lubotsky, and Jos Weitenberg, 101-14. (Innsbrucker Beiträge zur Sprachwissenschaft, Band 65.) Innsbruck: Institut für Sprachwissenschaft. Schmitt, Rüdiger. 1977. Grammatik des Klassisch-Armenischen mit sprachvergleichenden Erläuterungen. Darmstadt: Wissenschaftliche Buchgesellschaft. Schrijver, Peter. 1991. The Reflexes of the Proto-Indo-European Laryngeals in Latin. (Leiden Studies in Indo-European 2.) Amsterdam: Rodopi. Schwyzer, Eduard. 1939. Griechische Grammatik auf der Grundlage von Karl Brugmanns Griechischer Grammatik. Band I: Allgemeiner Teil, Lautlehre, Wortbildung, Flexion. München: Beck. Senn, Alfred. 1966a. Handbuch der litauischen Sprache. Vol. I: Grammatik. Heidelberg: Winter. —. 1966b. The relationships of Baltic and Slavic. Ancient Indo-European Dialects: Proceedings of the Conference on Indo-European Linguistics Held at the University of California, Los Angeles, April 25-27, 1963, ed. by Henrik Birnbaum and Jaan Puhvel, 139-51. Berkeley and Los Angeles: U. of California Press. Sihler, Albert. 1995. New Comparative Grammar of Greek and Latin. Oxford: Oxford U. Press. Sims-Williams, Nicholas. 1989. Sogdian. Compendium Linguarum Iranicarum.ed.Ъy Rüdiger Schmitt, 173-92. Wiesbaden: Reichert. Stang, Christian S. 1942. Das slavische und baltische Verbum. (Det Norske VidenskapsAkademi i Oslo, Hist.-Filos. Klasse, 1942, No. 1.) Oslo: Dybwad. —. 1952a. Eine Bemerkung zur slowakischen Präsensflexion. NTS 16, 271-5. —. 1952b. Sur l'accentuation des verbes causatifs et itératifs en slave. NTS 16, 263-70. —. 1957. Slavonic Accentuation. (Det Norske Videnskaps-Akademi i Oslo, Hist.-Fil. Klasse, No. 3.) Oslo. (Republished 1965, Oslo/Bergen/Tromsö: Universitetsforlaget.) —. 1966. Vergleichende Grammatik der baltischen Sprachen. Oslo/Bergen/Tromsö: Un i versitetsforlaget. Stankiewicz, Edward. 1973. The accentuation and grammatical categories of the -a stems in South Slavic. Južnoslovenski Filolog 30, 193-202. (Reprinted in Stankiewicz 1986c, 195-206.) —. 1978. The inflection of Serbo-Croatian substantives and their genitive plural endings. American Contributions to the Eighth International Congress ofSlavists, Zagreb and
240
Ljubljana, September 3-9, 1978, ed. by Henrik Birnbaum. Volume 1: Linguistics and Poetics, 666-82. Columbus: Slavica. (Reprinted in Stankiewicz 1986c, 215-29.) —. 1979a. Studies in Slavic Morphophonemics and Accentology. (Michigan Slavic Materials, 16.) Ann Arbor: Michigan Slavic Publications. —. 1979b. The oxytonic accent of the present tense in Serbo-Croatian and other Slavic languages. Stankiewicz 1979a, 147-58. —. 1986a. The place and function of stress in Russian nominal forms with a zero in the ending. Stankiewicz 1986c, 327-36. (Originally published in Russkoe i slavjanskoe jazykoznanie. K 70-letiju člena-korrespondenta AN SSSR R. I. Avanesova, 235-43. Moscow, 1972. —. 1986b. The Slavic athematic (nominal) stems and their accentuation. Stankiewicz 1986a, 395-409. —. 1986c. The Slavic Languages: Unity in Diversity. Berlin, New York, and Amsterdam: Mouton de Gray ter. —. 1993. The Accentual Patterns of the Slavic Languages. Stanford: Stanford U. Press. Starostin, S. L. and S. A. Nikolaev. 1981. Paradigmatičeskie klassy indoevropejskogo glagola. Balto-slavjanskie Issledovanija, 261-343. Steriade, Donca. 1988. Greek accent: a case for preserving structure. Linguistic Inquiry 19,271-314. Stiles, Patrick V. 1984. On the interpretation of older Runic swestar on the Opedal stone. NOWELE (Northwest European Language Evolution) 3, 3-48. —. 1985. The fate of the numeral '4' in Germanic. NOWELE (Northwest European Language Evolution) 6, 81-104. —. 1988. Gothic nominative singular bröpar 'brother' and the reflexes of Indo-European long vowels in the final syllables of Germanic polysyllables. TPS 86, 115-43. Streitberg, Wilhelm. 1896. Urgermanische Grammatik: Einführung in das vergleichende Studium der altgermanischen Dialekte. Heidelberg: Winter. Strunk, Klaus. 1967. Nasalpräsentien und Aoriste. Ein Beitrag zur Morphologie des Verbums im Indo-Iranischen und Griechischen. Heidelberg: Winter. —. 1994. Relative chronology and Indo-European verbal-system: the case of presentand aorist-stems. JIES 22:3/4,417-34. Szemerényi, Oswald. 1956. Latin rĕs and the Indo-European long-diphthong stem nouns. ZVSIЪ, 167-202. —. 1957. The problem of Balto-Slav unity — a critical survey. Kratylos 2:2, 97-123. —. 1962. Trends and Tasks in Comparative Philology: An Inaugural Lecture Delivered at University College, London, 23 October 1961. London: H.K. Lewis. —. 1967a. The perfect participle active in Mycenaean and Indo-European. Studi Micenei ed Egeo-Anatolici 2, 7-26. —. 1967b. The history of Attic ouç and some of its compounds. Studi Micenei ed EgeoAnatolici 3,47-88. —. 1968. Mycenaean: a milestone between Indo-European and historical Greek. Atti e memorie del Primo Congresso Internazionale di Micenologia, Roma, 27 settembre - 3 ottobre 1967, Vol. 2, 715-25. Rome: Edizioni dell'Ateneo. —. 1970. Einführung in die vergleichende Sprachwissenschaft. Darmstadt: Wissenschaftliche Buchgesellschaft. Taylor, Ann. 1990. Clitics and Configurationality in Ancient Greek. Ph.D. dissertation. Universiity of Pennsylvania. 241
—. 1994. The change from SOV to SVO in Ancient Greek. Language Variation and Change 6:1, 1-37. Tedesco, Paul. 1923. a-Stämme und aya-Stämme im Iranischen. Zeitschrift für Indologie und Iranistik 2,281 -315. —. 1948. Slavic /ie-presents from oldery'e-presents. Language 24:4, 346-87. —. 1951. Review of Mikkola 1950. Language 27:2, 165-77. —. 1968. Sanskrit dehiigi\e\ Language 44:1, 1-24. Testen, David. 1994. The spread of Q in the Avestan w-stem paradigm. CLS 30: Papers from the 30th Regional Meeting of the Chicago Linguistic Society, Vol. I: The Main Session, ed. by Katharine Beals, Jeannette Denton, Robert Knippen, Lynette Melnar, Hisami Suzuki, and Erica Zeinfeld, 418-27. Chicago: Chicago Linguistic Society. Thieme, Paul. 1929. Das Plusquamperfektum im Veda. (ZVS Ergänzungsheft Nr. 7.) Göttingen: Vandenhoeck und Ruprecht. Thurneysen, Rudolf. 1946. A Grammar of Old Irish. Revised and enlarged edition, translated from the German by D. A. Binchy and Osborn Bergin. Dublin: School of Celtic Studies, Dublin Institute for Advanced Studies. Vaillant, André. 1934. L'élimination des présents athématiques et le slave *moÉ. Revue des Études Slaves 14:1-2, 26-35. —. 1942-5. Hittite sakhi, latin sciö. BSL 42, 84-8. —. 1950. Grammaire comparée des langues slaves. Tome I: Phonétique. (Les Langues du Monde, Série Grammaire, Philologie, Littérature, Vol. VI.) Lyon: IAC. —. 1958. —. Tome II: Morphologie. Première Partie: Flexion Nominale. (Les Langues du Monde, Série Grammaire, Philologie, Littérature, Vol. XI.) Lyon: IAC. —. 1966. —. Tome III: Le verbe. Paris: Klincksieck. van Wijk, Nicholas (Nicolaas). 1923. Die baltischen und slavischen Akzent- und Intonations système. Ein Beitrag zur Erforschung der baltisch-slavischen Verwandtschaftsverhältnisse. (Verhandelingen der Koninklijke Akademie van Wetenschappen te Amsterdam, Afdeeling Letterkunde. Nieuwe Reeks, Deel 23, No. 2.) Amsterdam. —. 1926. Die slavischen Partizipia auf -to- und die Aoristformen auf -tu. IF 43, 281-9. —. 1931. Geschichte der altkirchenslavischen Sprache. Erster Band: Laut- und Formenlehre. (Grundriß der slavischen Philologie und Kulturgeschichte, v. 8.) Berlin/Leipzig: de Gruyter. Vasmer, Max. 1986. Etimologičeskyj slovar' russkogo jazyka. Perevod s nemeckogo i dopolnenija O. N. Trubačeva, pod redakciej i s predisloviem B. A. Larina. 2. izd., stereotipnoe. Tom 1: A-D. Moscow: Progress. —. 1986b. —. Tom II. E-Myž. —. Vine, Brent. 1999a. Greek Brjpóc, Armenian erkar 'long': notes on full-grade ^ r e formations in Greek and Indo-European. Paper presented at the 18th East Coast IndoEuropean Conference, University of Texas, Austin, 27-30 May 1999. —. 1999b. On "Cowgill's law" in Greek. Compositiones Indogermanicae in Honorem Jochem Schindler, hrsg. von H. Eichner und H. C. Luschützky unter redaktioneller Mitwirkung von V. Sadovski, 555-600. Prague: Enigma. Wackernagel, Jacob. 1877. Der griechische Verbalakzent. ZVS 23,457-70. (Reprinted in Kleine Schriften, Band 2, 1058-71. Hrsg. von der Akademie der Wissenschaften in Göttingen. Göttingen: Vandenhoeck und Ruprecht, 1953.)
242
Watkins, Calvert. 1969. Indogermanische Grammatik. Band III: Formenlehre. Erster Teil: Geschichte der indogermanischen Verbalflexion. Heidelberg: Winter. . 1970. Remarks on Baltic verb inflexion. Magner and Schmalstieg (eds.) 1970, 16570. —, ed. 1987. Studies in Memory of Warren Cowgill (J929-1985): Papers from the Fourth East Coast Indo-European Conference, Cornell University, June 6-9, 1985. Berlin and New York: Walter de Gruyter. —. 1993. Some Anatolian words and forms. Indogermanica et Italica: Festschrift für Helmut Rix zum 65. Geburtstag, ed. by Gerhard Meiser, 469-78. (Innsbrucker Beiträge zur Sprachwissenschaft, Band 72.) Innsbruck: Institut für Sprachwissenschaft. Weiss, Michael. 1994. On the non-verbal origin of the Greek verb νήφειυ 'to be sober'. HS 107, 91-8. Whitney, William Dwight. 1924. Sanskrit Grammar. Including Both the Classical Language and the Older Dialects of Veda and Brähmana. 5th ed. Leipzig: Breitkopf und Härtel. Winter, Werner. 1977. Internal structure and external relationship of two verbal paradigms: Tocharian Β wen-, A wen- 'say'. JIES 5, 133-59. —. 1978. The distribution of short and long vowels in stems of the type Lith. ésti : vèsti : mèsti and OCSjasti : vesti : mesti in Baltic and Slavic languages. Recent Developments in Historical Phonology, ed. by Jacek Fisiak, 431-46. (Trends in Linguistics: Studies and Monographs, 4.) The Hague/New York: Mouton. —. 1990. The importance of fine points in spelling: deletion of accented vowels in Tocharian B. Historical Linguistics and Philology, ed. by Jacek Fisiak, 371-91. (Trends in Linguistics: Studies and Monographs, Vol. 46.) Berlin and New York: Mouton de Gruyter. —. 1993. The development of underlying accented "shwa" before dental in Tocharian B. TIES β, 197-205. Wright, Joseph. 1910. Grammar of the Gothic Language and the Gospel of St Mark, Selections from the Other Gospels and the Second Epistle to Timothy, with Notes and Glossary. Oxford: Clarendon. Yoshida, Kazuhiko. 1990. The Hittite Mediopassive Endings in -ri. (Studies in IndoEuropean Language and Culture, Vol. 5.) Berlin and New York: Walter de Gruyter. Zaliznjak, Andrej A. 1967. Russkoe imennoe slovoizmenenie. Moscow: Izdatel'stvo «Nauka». —. 1985. Ot praslavjanskoj akcentuaciji k russkoj. (Akademija Nauk SSSR, Institut Slavjanovedenija i Balkanistiki.) Moscow: Izdatel'stvo «Nauka». Zee, Draga. 1988. Sonority Constraints on Prosodie Structure. Ph.D. dissertation, Stanford University. Zerdin, Jason. 2000. Studies in the Ancient Greek Verbs in -skö. Ph.D. dissertation. University of Oxford.
243