ERMANIC IN G r/n- S TEM H ETEROCLITE I NFLECTION R EMNANTS OF *

R

EMNANTS OF

*r/n-S

TEM

H

ETEROCLITE

I

NFLECTION

IN

G

ERMANIC

J.

K

LIMP

G.

K

ROONEN

J.-W.

Z

WART

MASTER THESIS FOR THE RESEARCH MASTER LINGUISTICS

UNIVERSITY OF GRONINGEN

Wakaiþ standaiduh in galaubeinai,

wairaleiko taujaiþ, gaþwastidai sijaiþ.

C

ONTENTS

Chapter 0. INTRODUCTION . . . . . 6

Chapter 1. PROTO-INDO-EUROPEAN GRADATION TYPES AND EARLY GERMANIC SOUND LAWS . . . . . . 13

§1.1 PROTO-INDO-EUROPEAN GRADATION TYPES . . . . 13

§1.1.1 NOMINAL GRADATION PATTERNS . . . . . 13

§1.1.2 THE PROTERODYNAMIC TYPE . . . . . . 15

§1.1.3 THE HYSTERODYNAMIC TYPE . . . . . . 19

§1.1.4 THE AMPHIDYNAMIC TYPE . . . . . . 22

§1.2 THE LENGTHENED GRADE . . . . . . 24

§1.3 AN OVERVIEW OF RELEVANT GRADATION TYPES AND LAWS OF VOWEL LENGTHENING . . . 30

Chapter 2. THE HETEROCLITE DECLENSION AND ITS REFLECTION IN GERMANIC . . . 31

§2.1 THE HETEROCLITE INFLECTION . . . 31

§2.1.1 FORMAL ANALYSIS OF HETEROCLITE INFLECTION . . . 31

§2.1.2 THE DIFFERENT TYPES OF HETEROCLISY . . . . 31

§2.1.3 ANOTE ON THE GRADATION OF HETEROCLITE INFLECTION . 33

§2.2 ASURVEY OF GERMANIC REFLEXES OF INHERITED HETEROCLITES . 34 §2.2.1 THE WORD FOR ‘DAY’ . . . 34

§2.2.2 THE WORD FOR ‘WATER’ . . . 38

§2.2.3 THE WORD FOR ‘UDDER’ . . . . . . 42

§2.2.4 THE WORD FOR ‘LIVER’ . . . 43

§2.2.5 THE WORD FOR ‘EXCREMENT’ . . . 47

§2.2.6 THE WORD FOR ‘SPRINGTIME’ . . . 47

§2.2.7 THE WORD FOR ‘FEATHER’ . . . 49

§2.2.8 THE WORD FOR ‘SHELTER,SHED’ . . . 52

Chapter 3. THE HETEROCLITE NOUN FOR ‘FIRE’ . . . 55

§3.1 THE GERMANIC MATERIAL FOR THE WORD FOR ‘FIRE’ . . . 55

§3.1.1 GOTHIC . . . 55

§3.1.2 OLD NORSE AND THE NORDIC DIALECTS . . . . 57

§3.1.3 OLD HIGH GERMAN AND THE HIGH GERMAN DIALECTS . . 62

§3.1.4 OLD DUTCH AND THE FRANCONIAN DIALECTS . . . 65

§3.1.5 OLD SAXON AND THE LOW GERMAN DIALECTS . . . 68

§3.1.6 OLD ENGLISH AND THE BRITISH DIALECTS . . . . 70

§3.1.7 OLD FRISIAN AND THE FRISIAN DIALECTS . . . . 71

§3.2.1 THE ETYMOLOGICAL CONNECTION OF

THE PIEHETEROCLITE STEM ‘FIRE’ . . . 72

§3.2.1.1 HITTITE . . . 73

§3.2.1.2 ANCIENT GREEK . . . 74

§3.2.1.3 SANSKRIT AND THE INDO-IRANIAN DIALECTS . . . 75

§3.2.1.4 TOCHARIAN . . . 76

§3.2.1.5 THE REMAINING IEDIALECTS AND THE

‘DOUBLE-ZEROED’FORM *puh2r . . . 77

§3.2.2 PRIOR THEORIES OF THE PIEDERIVATION AND

DEVELOPMENT OF THE HETEROCLITE STEM ‘FIRE’ . . . . 79

§3.3 THE DERIVATION OF THE PG FORMS FROM THE

PIEHETEROCLITE STEM ‘FIRE’ . . . 83

Chapter 4. THE HETEROCLITE NOUN FOR ‘SUN’ . . . 87

§4.1 THE GERMANIC MATERIAL FOR THE WORD FOR ‘SUN’ . . . 87

§4.1.1 GOTHIC . . . 87

§4.1.2 OLD NORSE AND THE NORDIC DIALECTS . . . . 88 §4.1.3 OLD HIGH GERMAN AND THE LATER HIGH GERMAN DIALECTS . 91

§4.1.4 OLD DUTCH AND THE LATER FRANCONIAN DIALECTS . . 92

§4.1.5 OLD SAXON AND THE LOW GERMAN DIALECTS . . . 93 §4.1.6 OLD ENGLISH AND THE BRITISH DIALECTS . . . . 96 §4.1.7 OLD FRISIAN AND THE FRISIAN DIALECTS . . . . 97 §4.2 THE PIEORIGINS OF THE WORD FOR ‘SUN’ . . . . 100

§4.2.1 THE ETYMOLOGICAL CONNECTION OF THE

PIEHETEROCLITE STEM ‘SUN’ . . . 100

§4.2.1.1 ANCIENT GREEK . . . 101

§4.2.1.2 SANSKRIT AND THE INDO-IRANIAN DIALECTS . . . 102

§4.2.1.3 LATIN . . . 104

§4.2.1.4 BALTO-SLAVIC . . . 106

§4.2.1.5 CELTIC . . . 107

§4.2.2 THE DERIVATION OF THE PG FORMS FROM THE

PIEHETEROCLITE STEM ‘SUN’ . . . 108

Chapter 5. THE HETEROCLITE NOUN FOR ‘WELL,SOURCE’ . . . 115

§5.1 THE GERMANIC MATERIAL FOR THE WORD FOR ‘WELL, SOURCE’. . 115

§5.1.1 GOTHIC . . . 115

§5.1.2 OLD NORSE AND THE NORDIC DIALECTS . . . . 115

§5.1.3 OLD HIGH GERMAN AND THE LATER HIGH GERMAN DIALECTS . 116 §5.1.4 OLD DUTCH AND THE LATER FRANCONIAN DIALECTS . . 117

§5.1.5 OLD SAXON AND THE LOW GERMAN DIALECTS . . . 117

§5.1.6 OLD ENGLISH AND THE BRITISH DIALECTS . . . . 118

§5.1.7 OLD FRISIAN AND THE FRISIAN DIALECTS . . . . 118

§5.2.1 THE ETYMOLOGICAL CONNECTION OF THE

PIEHETEROCLITE STEM ‘WELL,SOURCE’ . . . 118

§5.2.1.1 ANCIENT GREEK AND ARMENIAN . . . 119

§5.2.1.2SANSKRIT . . . 119

§5.2.2 THE DERIVATION OF THE PG FORMS FROM THE

PIEHETEROCLITE STEM ‘WELL,SOURCE’ . . . 121

Chapter 6. THE DEVELOPMENT OF HETEROCLITES IN GERMANIC . . . 123

§6.1 THE PARALLEL DEVELOPMENT OF SOME GERMANIC HETEROCLITES . 123

§6.2 ON THE DEVELOPMENT OF THE LARYNGEAL IN THE SEQUENCE * - 125 §6.3 THE WORD FOR ‘PALATE,GUMS’ . . . 127 §6.4 THE WORD FOR ‘POLE,PROP’ . . . 130 §6.5ON THE RELATIVE CHRONOLOGY OF SOME RELEVANT SOUND CHANGES 131

Chapter 7. CONCLUSION . . . 135

§7.1 SUMMARY AND CONCLUSION . . . 135

§7.2 OVERVIEW OF GRADATION TYPES AND SOUND LAWS . . . 137 §7.2.1SURVEYABLE REPRESENTATION OF PIEGRADATION TYPES . . 137 §7.2.2FORMULATION OF PIESOUND LAWS . . . 137

§7.2.3LAW OF LARYNGEAL LOSS . . . 137

§7.2.4THE FORMULATION AND RELATIVE CHRONOLOGY OF

SOME GERMANIC SOUND LAWS . . . 138

A

BBREVIATIONS

L

Dan. modern Danish Dor. Doric

Du. modern Dutch EG East-Germanic Far. Faroese

Go. Gothic

Gr. Greek

HG modern High German Hitt. Hittite

Hom. Homeric

Icel. modern Icelandic Lat. Latin

Latv. Latvian

LG modern Low German Lith. Lithuanian

MCorn. Middle Cornish MDu. Middle Dutch

MHG Middle High German MLG Middle Low German MPers. Middle-Persian MW Middle Welsh NeG Northeast-Germanic NFris. North Frisian NG North-Germanic Norw. modern Norwegian

NwG Northwest-Germanic OAv. Old Avestan

OBret. Old Breton OCorn. Old Cornish

OCS Old Church Slavonic ODan. Old Danish

ODu. Old Dutch OE Old English

OELF Old East Low Franconian OFris. Old Frisian

OHG Old High German ON Old Norse

OPru. Old Prussian OS Old Saxon OSw. Old Swedish

OW Old Welsh Pers. Persian PG Proto-Germanic PGr. Proto-Greek PIE Proto-Indo-European PIIr. Proto-Indo-Iranian PT Proto-Tocharian Russ. Russian SCr. Serbo-Croatian Skt. Sanskrit Sw. modern Swedish Toch. Tocharian (A and B) Ved. Vedic Sanskrit

W Welsh

WFris. West Frisian

WG West-Germanic

Chapter 0.

I

NTRODUCTION

§0.1

I

W

A

The present thesis was created over a period of about a year, divided naturally yet somewhat unevenly into two parts. The first part was during a stay of four months in Copenhagen, in the vicinity of my supervisor Guus Kroonen. The bulk of this thesis, however, was written upon my return in the Netherlands. My greatest appreciation goes out to my supervisor Guus Kroonen. Guus has been an inspiration and a humbling source of knowledge. He has given unending support for which I am deeply grateful. Secondly, I wish to express the greatest thanks to my second supervisor and teacher Jan-Wouter Zwart. He has never given up on me, despite my erratic study path and he has supported me at many occasions over the years. I moreover wish to thank him specifically for giving me a number of enormously valuable books that I have literally used daily ever since, most notably Jan de Vries’ Altnordisches etymologisches Wörterbuch and Winfred Lehmann’s A Gothic

Etymological Dictionary. I also wish to express my gratitude to the people of the

Indo-European Faculty at Copenhagen University. I spent almost four months daily at the faculty, even though I was there unofficially and unplanned—a decision on the spur of the moment. I was nevertheless welcomed with more than open arms. The faculty has provided me full access to the impressive reading room and to the great minds of the staff. I wish to single out Birgit Olsen, whom I visited on many occasions and with whom I had such wonderful talks about historical linguistics and pretty much everything else in life. I cannot single out any other members there, for they have all been very friendly and helpful without exception. Upon my return in the Netherlands I have dearly missed the nights out drinking together in Copenhagen.

§0.2

A

S

have once formed an important part of the lexicon. We find some of the most fundamental and historically stable vocabulary within this class. Heteroclite nouns range in meaning from natural phenomena such as ‘sun’ and ‘fire’, to body parts and organs like ‘liver’ and ‘blood’, and even to seasonal terms such as ‘springtime’. Formally, the inflection of heteroclites is quite peculiar. Whereas typical nouns attach a single suffix to the root within a paradigm, heteroclites show alternation of two different suffixes to the root. This is doubtlessly a most archaic feature. This study does not aim to be an exhaustive and definitive study of Germanic remnants of heteroclites, however; its aims are more modest. There are several types of heteroclite inflections, but I limit myself in this thesis to the class that alternates *-r- (or *-l-) and *-n-. Although a substantial number of r/n-stems are discussed, many of which provide interesting details and clues to their prehistoric development, others may still be recovered. Furthermore, a large part of this study is devoted to the parallel development of a small set of heteroclites that must have partially developed together.

Remnants of heteroclites can be found in all of the Indo-European daughter languages, although the older dialects such as Greek and Indo-Iranian still continue intact heteroclite inflection. Hittite is considered of particular importance for the reconstruction of heteroclites, because of its great number of intact heteroclite nouns. I have found that the Germanic languages and dialects have more than a few heteroclite remnants and even some fully functioning heteroclites in Gothic. As I show in chapter four, for instance, Gothic is the only Indo-European language in which the heteroclite noun for ‘sun’ is found intact and productive—a highly archaic situation. Germanic has much to offer to the comparative Indo-European study of heteroclites. And although Indo-Indo-European comparson is of great importance, I must stress that in this thesis I present a Germanicist perspective on the development of heteroclite r/n-stems in Germanic and provide new interpretations as well as new material from the Germanic dialects. I hope, then, that this thesis is may be of use to both Germanicists and Indo-Europeanists.

§0.3

P

O

C

In the second chapter I define nominal heteroclisy and briefly discuss some noteworthy properties of heteroclites. The rest of this chapter is devoted to discussing a number of Germanic remnants of old heteroclite nouns, namely the words for ‘day’, ‘water’, ‘udder’, ‘excrement’, ‘spring’, ‘feather’, and finally ‘shelter, shed’. This etymological survey concludes an important part of this thesis. It is interesting to find that Germanic actually provides much material for this very old class of nouns.

The third, fourth, and fifth chapter discuss three words that have at a certain point in time developed together. These chapters thus belong together in an important sense, even though they stand on their own (and can certainly be read as independent chapters). In the third chapter I investigate the word for ‘fire’. This is the bulkiest chapter of this thesis, enriched with dialectal material and details. As in the two chapters that follow, chapter three starts by gathering all relevant forms from the Germanic territory. Each Germanic language group is discussed and judged separately. Then, the comparative Indo-European material is adduced and an original, Proto-Indo-European paradigm is established. The final section of the chapter concludes with a section on how the Germanic forms derived from the Proto-Indo-European word. The fourth chapter treats of the word for ‘sun’ in a similar way, as does the fifth chapter on the word for ‘well, source’. In these three chapters I have reached the conclusion that the words must have shared in certain developments and I moreover require a specific law of the loss of laryngeals to account for their development. These chapters, too, show that Germanic and its dialects provide a huge amount of relevant material for the study of heteroclite nouns, and more generally for comparative Indo-European linguistics.

The sixth chapter discusses and surmises the details of the parallel development of the reflexes of the heteroclite stems for ‘fire’, ‘sun’, ‘well, source’, and ´shelter, shed’. A tentative law of laryngeal loss is deduced for which two other forms are adduced and discusses: the word for ‘gums, palate’ and the word for ‘pole’. Finally, some matters of the relative chronology of sound laws are examined. The seventh chapter provides a conclusion and short survey of this thesis as well as a list of relevant PIE gradation types, the formulation of PIE sound laws of vowel lengthening, and a list of formulations of important sound changes in a relative chronology.

§0.4

A

N

S

C

D

In my reconstructions I make ample use of the concept of a ‘transpose’. My conception of a transpose is loosely based on the concept of Heiner Eichner. In the absence of any book or article in which the concept is defined, I take the liberty of proposing a slightly modified use of the concept.1 I call a transpose a reconstruction from a proto-language to a

later proto-language or hypothetically to an actual language by sound law alone. All analogical processes, etc. are thus excluded. A transpose, then, provides a purely mechanical and ‘blind’ reconstruction. The value of a transpose reconstruction is that it provides a useful measure of comparison to comparatively deduced reconstructions (on the basis of a correspondence set) and may reveal where attested forms deviate (and may have innovated, etc.). Moreover, a transpose explicates our knowledge of sound laws and on the other. A down-side is therefore that the value of the transpose is limited by our (sometimes fragmentary) knowledge. The reconstruction of a proto-form from a correspondence set is, of course, the principal tool of comparative linguistics (besides internal reconstruction) and the concept of a transpose is meant to merely assist this principal method. The deviation of reconstruction that is deduced from a correspondence set to a transpose may (or may not) reveal interesting patterns and (a)symmetries. I must emphasize that this is not a new concept at all, for the method is often implicitly used. Once the reader is familiar with the definition its implicit used can readily be found everywhere. I have found it useful to make the concept explicit and more strictly defined. I indicate a transpose reconstruction with a superscript ‘plus’, ‘+_{’, to distinguish it from a regular deduction, denoted by the usual asterisk ‘*’.}

The Tocharian word A, B or ‘wood’ may be used to illustrate the value of the transpose (the example is taken from the text on §1.1.2). The word can be connected to Ved.

and Av. ‘wood’ (GEN.SG Ved. dróḥ, YAv. aoš), but the Tocharian form has

mysteriously lost its initial consonant. That is, we would expect **tor, not or. The word can be derived as follows (following Ringe, 1996:98, after a personal suggestion from Schindler):

PIE PT1 PT2

*dór-u > +tó2r-ä → *ó2r-ä

* - - > + - w-

The transpose of PIE * - - is +r-äw- which has lost the pre-consonantal *d by the law

*#dC- > *#C- (cf. Ringe, 1996:64ff.). The initial *t was eliminated everywhere in the paradigm to maintain innerparadigmatic consistency. We can now explain the attested forms of Tocharian A, B or. We can use the transpose as a tool that helps explain the origin of the analogical spread of the loss of *d throughout the paradigm. That is, in itself the transpose does not offer anything new to our methods—the solution itself was suggested by Schindler without ever having needed this concept—but the concept of a transpose provides a heuristic tool that can help explicate or assist us in clarifying some of our reconstructions.

I typically denote phonological forms that underlie the (ambiguous or unclear) notation of an attested word between slashes: ‘/.../’. Contrary to modern convention, I use a cursive notation between slashes, in line with writing all reconstructions and (normalized) attested forms in cursive. Exact attested instances of a word are given between triangular brackets: ‘<...>’. I have chosen to abbreviate the reference to the many dialect dictionaries by giving the first four letters of the last name in small capitals. This is done to aid the readability. The Vedic books a is abbreviated to RV and the Atharvaveda to AV. The structure of nouns is denoted as follows: root as R, suffix as S, and case ending as E. The

vowel grades /e, ē, o, ō, / occurring in the root R, suffix S, and/or ending E are given between contiguous parentheses, e.g. R()-S( ), meaning a zero-gade root with a suffix in the lengthened o-grade (the accentuation on the vowel is optional). The gradation types of nominal paradigms (discussed in depth in chapter one) are given an abstract matrix notation, where the rows denote the grade of the root, suffix, and ending respectively, and the lines denote the strong, middle (if there is one), and weak stem respectively. For example the gradation matrix gmn can be given as follows (where gij denotes an instance of the the set of vowel grades /e, ē, o, ō, / and the first subscript, m, indicates the rows and the second, n, the lines):

|

|

An arrow ‘→’ (or its reverse) is used to indicate morphological or analogical change, whereas ‘>’ (or its reverse) indicates a phonological change (which may or may not include analogical changes, etc.). A reconstructed form is marked by a preceding asterisk ‘*’ and a false form (especially falsely reconstructed) is indicated by two asterisks ‘**’, as is conventional. Finally, a word boundary is indicated by ‘#’.

Furthermore, all etymologies have been checked against the invaluable database

etymologiebank, found at http://www.etymologiebank.nl/. This online resource provides fully

digitalized versions of the following four etymological dictionaries:

Philippa, M., Debrabandere, F., Quak, A., Schoonheim, T., & Sijs, N. van der

2003-2009 Etymologisch woordenboek van het Nederlands, AUP: Amsterdam.

Vries, J. de

1971 Nederlands etymologisch woordenboek, Brill: Leiden.

Wijk, N. van

1936 F anck’s tymolo sch woo nbo k N lan sch taal, met een Supplement van C.B. van Haeringen, Martinus Nijhoff: Den Haag.

Vercoullie, J.

1925 Beknopt etymologisch woordenboek der Nederlandsche taal, Van Rysselberghe

& Rombaut Uitgevers, Gent.

Chapter 1.

P

ROTO

-I

NDO

-E

UROPEAN

G

RADATION

T

YPES

AND

E

ARLY

G

ERMANIC

S

OUND

L

AWS

§1.1

P

ROTO

-I

NDO

-E

UROPEAN

G

RADATION

T

YPES

In this chapter I discuss patterns of vowel gradation within the paradigm of certain nouns. The stem (consisting of a root and zero, one, or more suffixes) of athematic nouns displays innerparadigmatic vowel gradation, which means that while the consonants of the stem remain invariant the vowel pattern varies. Each athematic noun originally had two or three distinctly graded stems. Each stem formed different cases by adding case endings. The strong stem always formed the nominative case as well as the accusative if there was no middle stem. If there was a middle stem, it formed the accusative and locative case. The weak stem formed the remaining cases. For instance, the word for ‘father’ is reconstructed to have a strong stem in the NOM.SG *ph2t , a middle stem ACC.SG *ph2tér-m , and a weak stem GEN.SG

*ph2tr-ós. This form shows the consonantal pattern *ph2tr- which remains intact despite the

variation in (or absence of) the vowels. The vowel gradation is * , *é, and  (that is, no vowel at all) respectively between the *t and *r, and *ó in the genitive ending. Interestingly, the variation in vowel grades can be shown to have originally been markedly constrained; a vowel grades either *e, *o, or . Note that the long * in *ph2t is the result of lengthening,

as I discuss later in this chapter.

There are several patterns of paradigmatic vowel gradation and they have been the subject of much debate. The patterns must be considerably old and most languages have generalized a fixed vowel pattern throughout the paradigm, although Hittite, Greek, Indo-Iranian, and marginally in some other languages still show living patterns of gradation. The paradigmatic patterns of vowel gradation are very important for the reconstruction of heteroclite nouns, because they can help explain and unite the attested variant forms in the Indo-European dialects, and, more specifically for this study, the attested forms of the Germanic dialects. In the chapter at hand I discuss some important and relevant patterns of vowel gradation that have been discussed in the literature and I will review some of the most important, probative evidence. Furthermore, some laws of lengthening of originally short vowels are discussed. The reason for this is that lengthened vowels occur in some of the proposed reconstructions and this has to be explained.

N

G

P

It had already been observed by the Indian grammarians that the vowels of words belonging to different stem classes often show the same underlying pattern in their vowel gradation. In the 19th century it was further realized from a historical, comparative point of view that the patterns of vowel gradation often are, or were, identical across the different stem-classes (a.o. Bartholomae, 1882:25; de Saussure, 1879, etc.). Such patterns in nominal and verbal vowel gradation reflect a deeper unity within the morphological system, transcending stem classes based on suffix identification (i.e. as r-stems, n-stems, etc.). Pedersen (1926), in his landmark study of nominal gradation types, recognized two distinct patterns of gradation in the nominal system. Pedersen called these two fundamental gradation types “flexion hystérodyname” and “flexion protérodyname”. The first type is exemplified by the canonical form (in modern reconstruction) *ph2tē ‘father’ and the latter type by the

canonical form (in modern reconstruction) *suHnús ‘son’. In order to account for the variation in gradation in the attested languages, Pedersen has to assume a plethora of analogies and innovations which degraded the original two patterns of gradation in very early PIE times. Pedersen’s work fell on fertile ground with Kuryłowicz (1935) and especially Kuiper (1942), who in many ways furthered and refined Pedersen’s views on patterns in gradation and laid some of the foundations for the modern conception of gradation types.

Today there are principally two schools of thought on PIE gradation: the Erlangen school and the Leiden school. There is considerable overlap in views between the schools even though disagreement exists on many details. I shall first focus on the Erlangen school. The Erlangen school originated with Hoffmann’s seminar on heteroclite r/n-stems in 1964 where gradation patterns were extensively discussed (cf Rix, 1965, and see Meier-Brügger, 20028:205 for a survey and on the historiography). The Erlangen school basically reconstructs two distinctly graded stem-forms for each gradation type, called the strong stem and the weak stem, and possibly a different third stem for the locative. The absence of a middle stem in the reconstruction of gradation types is a distinctive difference with the Leiden school.

Characteristic of the Erlangen school is the reconstruction of a greater variety of gradation types than the two recognized by Pedersen and Kuiper. Furthermore, there is some (and often considerable) variation between individual views on the nature and number of gradation types. This has resulted in the inherent lack of unity within the Erlangen school (which has since spread all over the globe). For example, Eichner (1973) developed Pedersen’s terminology further, positing the following seven (!) gradation types:

The Leiden school of gradation types is codified mainly in the work of Kuiper’s student Beekes (1985, cf. the survey in Beekes, 20112, and rececntly Kloekhorst, 2013). The Leiden school is characterized by a more reductive argumentation in which the proliferation of gradation types is constrained by positing two original types: hysterodynamic (HD) and proterodynamic (PD). The PD and HD types are posited as the two original gradation types from which other types developed over time. The Leiden school does allow for a considerably greater variety of gradation types for later PIE as it does for early and pre-PIE, however. That is, the Leiden school accepts other gradation types such as posited by the Erlangen school, but only as later, degenerated types of the more archaic HD and PD types. Another peculiarity of the Leiden school is that instead of reconstructing two forms for each gradation type, the Leiden school posits a separate and distinct graded stem form for the accusative and locative of the HD inflection which I call the middle stem. The Leiden HD inflection thus comprises a total of three differently graded stems in contrast to two graded stems in the Erlangen HD (disregarding the Erlangen reconstructions of an independent locative stem), a fact to which I return below.

Note that the present discussion of gradation types excludes not only non-nominal gradation such as verbal gradation (which hardly goes without saying), but also excludes root noun gradation. Root nouns follow their own patterns of gradation, but they have no bearing on the present study. Therefore, I discuss only athematic stem gradation in the following paragraphs.

§1.1.2

T

P

T

The proterodynamic (or: proterokinetic) gradation type, PD for short, is of great importance for the heteroclites in the following chapters, as will become readily apparent. The PD inflectional type is uncontroversial and well established, and it is reconstructed by both the Leiden and Erlangen schools as having a strong stem R(é)-S()-E() and a weak stem R()-S(é)-E() (cf. Kuiper, 1942:4; Schindler 1975a and b, Rix, 19922:123; Rieken, 1999:6; Widmer, 2002:36, Beekes, 20112:202ff.). A probative form for this gradation type is the heteroclite noun for ‘blood’. The word is attested in Hittite as - š- a < *h1ésh2- (cf. Gr.

ἔαρ, Skt. s -k), with a weak stem e.g. GEN.SG š- a-an-aš from *h1esh2-én- which leveled

R(e) throughout the paradigm (thus HIL:307f.). The weak stem is retained in the state R()-E(é)- in the Latin form san ēs ‘diseased blood; pus’ < *h1sh2-én-ih2- ‘blood-like’). Old

Irish continues the archaic paradigm for ‘woman’: ben < virtual *gw

én-eh2 for original

*gwén-h2, with the archaic GEN.SG mná < *gwn-éh2-s. Sanskrit continues the strong and weak

stem of this word as independent nouns. The strong stem *gwén-h2 ‘wife’ became the i-stem

an ḥ based on the vocalized laryngeal -i (cf. the transpose of *gw

én-h2 > +jan-i). The weak

stem *gwn-éh2- was continued in n - ‘queen’and developed into an -stem. The form is also

anachronistically reconstructed form *gwén-eh2. The form *gwén-eh2 may be interpreted as

continuing strong R(é) together with weak S(e), from which it became an ō-stem paralleling the Sanskrit -stem. Following both the Erlangen and Leiden school, I posit a PD1 paradigm (not to be confused with PD2 discussed below):

PD1 = | ́  

 ́ |

‘blood’ ‘wife’

NOM.SG R(é)-S() *h1ésh2- *gwén-h2

GEN.SG R()-S(é)-E() *h1sh2-én-s *gwn-éh2-s

More controversial, however, are PD forms with R(ó) in the strong stem instead of R(é). Kuiper (1942, 30ff.) already drew attention to the Vedic and Avestan neuter u-stems with R(ó) and classified them as PD: Ved. / ḥ ‘wood’ (Av. , GEN.SG YAv.

aoš), s n /sn ḥ ‘summit, mountain ridge’, y ‘life, lifetime’ (with the pattern intact in

Avestan yū/yaoš), and n ‘knee’ (all with open syllable lengthening of *ó by Brugmann’s Law). These neuter u-stems have cognates in most if not all Indo-European languages and are certainly old. However, the neuter u-stems stems are classified as acrostatic (AS)in both the Leiden and Erlangen schools (following Schindler, 1975a, Beekes, 20112:208, Widmer, 2002:35, etc.). Schindler (1975a:4ff.) posits two types of AS gradation of which the first type interests us here (for the second type contrasting R( : é) see the discussion of the word for ‘liver’ §2.2.4, p.43). The AS pattern is characterized by static root gradation R(ó) in the strong stem and R(é) in the weak stem, whereas suffixes and endings are in the zero-grade. The most probative examples of AS include the word for ‘water’, but as shown by Kloekhorst (fthc.a) and in detail in §2.2.2, p.38, this word cannot have been AS. The neuter u-stems that are adduced as forms that would prove AS gradation actually militate against the reconstruction of this type for the athematic nouns.

First of all, the form *dór-u/*dr- -s ‘tree’ can be reconstructed on the basis of Indo-Iranian forms: NOM.SG *dór-u for Ved. and Av. ‘wood’, and the GEN.SG *dr- -s

for Ved. dróḥ, YAv. aoš. Thus, the weak stem *dr- - has R()-E(é)- and does not correspond at all to the AS stem R(é)-E()-. The pattern may be explained by analogies, but the Indo-Iranian neuter u-stems together, as a class unto themselves, incontrovertibly showcase a consistent pattern that does not implicate analogy or leveling. The proposition that the pattern of the locative stem was leveled to the genitive is unconvincing, for there is no evidence for a locative stem of the pattern R(é)-E()- in the first place.2 The GEN.SG Ved.

dróḥ, YAv. aoš simply does not have a model for analogy. The gradation pattern of the

weak stem cannot be explained otherwise and must be original. This consistent and systematic pattern is furthermore corroborated by other Indo-European languages. The Tocharian word

2 _{In fact, the}

LOC.SG is attested as e.g. and (with - i from n-stems, which strongly influenced the

for ‘wood’ A, B or may be derived as follows (following Ringe, 1996:98, after a personal suggestion from Schindler; the transpose is, of course, my addition):

PIE PT1 PT2

*dór-u > +tó2r-ä → *ó2r-ä

* - - > + - w-

That is: * - - > +r-äw- by loss of pre-consonantal *d via the law *#dC- > *#C- (cf. Ringe,

1996:64ff.). The loss of *t was leveled over the paradigm to maintain innerparadigmatic consistency (i.e. the PToch. initial *t was eliminated everywhere in the paradigm). This means that we can explain virtual *o2r-u (with u-mutation of *ë < *ó, yielding *o2) by assuming

analogical spread of the absence of *d from the weak stem +r-äw- < * - -. This provides

indirect but compelling evidence for the existence of a weak PD stem with R()-E(é)-, for only this form provides the environment for loss. Thus, the Tocharian forms also point to a PIE paradigm *dór-u/*dr- -s, corroborating the pattern that is found in Indo-Iranian.

Furthermore, Hittite has the form t [n.] ‘wood’ (Old Script plene spelling

NOM/ACC.SG ta-a-ru, cf. CLuw GIŠt - ‘wood’) which goes back directly to the expected

*dór-u (cf. HIL:980f.), as does Gr. δόρυ ‘spear’. In Germanic, the transpose of the weak PD stem * - - > +t - - survives in the thematic neuter *t -an, cf. ON tré, Go. triu, OE t ēo,

t ēow, OS trio, etc. ‘tree, wood’. These forms provide additional support for the consistent

Indo-Iranian pattern. In this light, the Germanic formation *ter- a- > OE teoru, teru ‘tar’ (cf. Lith. dervà ‘burned wood, tar’, also cf. Let. darva ‘tar’ to ON tjára < *der- eh2), does not

corroborate an AS pattern R(é)-E()-, but rather reflects an independent formation. The forms are formally and semantically further removed, meaning ‘tar’ not ‘tree, wood’. That is, the form that formally matches the weak stem of the pattern in *dór-u/*dr- -s, i.e. PG *t -an meaning ‘tree, wood’, also has an incontrovertible semantic correspondence. Thus, these secondary formations are no evidence for an AS paradigm.

Another word that is used as probative evidence for an AS paradigm is * on-u, * n-u ‘knee’. The form should in my opinion be of the PD type with strong R(ó), i.e. * on-u/ * n- -. We find evidence for the stem * n-u with R(ó) in Vedic n , Av. z n ‘knee’, Gr. γόνυ ‘knee’, Toch.A kanw-em, Toch.B keni (NOM.PL k nīn ) ‘knees’, etc. The corresponding

weak stem is found in Go. kniu, OHG knio, ‘knee’ etc., a thematization of *kn-eu- < * n- -. The word is nevertheless classified as AS on the basis of the e-grade in the root in Latin genu ‘knee’ and Hittite genu ‘knee’, which point to * én-u. Kloekhorst (2013:109) rightly rejects the AS gradation of this form on the basis of the Hittite forms, which according to him is PD. However, the form ganu- occurs as well. But, as Kloekhorst (HIL:540f.) convincingly shows, the paradigm clearly has R(é) everywhere and the form ganu- (occurring in the verb

kan šša azi

Latin gena [f., -stem] means ‘cheek, side of the face’, but it is still an u-stem in the fixed collocation ntēs n -īnī ‘molars’ (cf. EDLIL:257f., where it is suggested that the form became an -stem in analogy to m la ‘cheekbone, jaw’). The form must be connected to a very similar PIE form derived from the same root, but with a consistent difference in meaning: * én-u-s [m.] ‘chin, cheek’, cf. Greek γένυς, hánuḥ 3, Avestisch zanauua (PL) , etc., all meaning ‘chin, cheek’.4

The form is attested in Germanic as well where we find Go. kinnus ‘chin, cheek’, ON kinn ‘cheek’ < PG *kinn- < *k n - < * én - (when followed by a vowel in the oblique cases and with restitution of the suffix *-u- to the new stem *kinn-). Otherwise we find o-stem forms formed to *kinn- < * én -, cf. OHG kinni, OS kinni, OE cinn, cin, OFri.

zin-, all ‘chin’. The form * én-u-s [m.] ‘chin, cheek’ doubtlessly contains the same root as the

PD form * n-u/* n- - [n.] ‘knee’, but both formations must be old and existed side by side in the proto-language. I conclude this from the fact that many languages (e.g. Vedic, Avestan, Greek, Germanic, Tocharian, etc.) continue reflexes of both the neuter * n-u/* n- - [n.] ‘knee’ as well as of * én-u-s [m.] ‘chin, cheek’, proving that both formations must go back to the proto-language. It is clear that the masculine and neuter forms must be strictly separated on the basis of the consistent difference in both form and meaning. In this light, the form * én-u-s ‘chin, cheek’ must have yielded Latin gena ‘cheek, side of the face’, which is perfectly acceptable when we assume analogical transfer to the -stem in analogy to m la ‘cheekbone, jaw’, and the PD form * n-u/* n- - ‘knee’ must then have yielded genu ‘knee’. The root vocalism in the form genu, however, must reflect R(é), which is unexpected when we reject an AS paradigm. This has caused much confusion.

However, the fact that only Latin and Hittite do not fit the pattern of * n-u/* n- -, and the fact that Hittite must have been proterodynamic (at least PD1), leaves the Latin form isolated. It is then likely that the Latin form took the vowel grade from somewhere else. It is possible that other neuter u-stems (such as pecu ‘cattle’) influenced the stem or that the very similar form * én-u-s ‘chin, cheek’ (continued in gena) influenced the form. Alternatively, it cannot be ruled out that the form was actually a zero-grade: * n -u-, with vocalization of the nasal.5 Since the form is securely attested with long -ū already in the oldest attestations (Weiss, 1993:92ff., who suggests it continues a dual), it may represent an old dual * n -u-h1 or

plural * n -u-h2 (cf. the formation in Gr. δρῦς ‘tree (oak); timber’ if from *dr-u-H-). The Latin

3

Kuiper (1942:49) reconstructs * h1én-u-s “chin” with * h1- to account for the Vedic initial h of hánus “chin, jaw”. This may explain the Indic form nicely, but is not desirable because of the odd root structure *CHEC- (and may arguably have produced Greek +χένυς instead of γένυς as well). The Vedic initial consonant remains inexplicable, but the form can hardly be separated from the masculine form * n-u-s.

4

It is tempting to connect the root in the * n-u/* n- - [n.] ‘knee’ to * n-u-s [m.] ‘chin, cheek’ semantically to a hypothetical * n- indicating some kind of joint if the meaning of * n-u-s ‘chin, cheek’ developed from ‘jaw’. This is especially appealing in light of a connection to e.g. PG *kn -ka- (cf. Norw.dial. Norw. dial.

knjuka ‘knuckle’, ON knjúkr ‘rounded mountain top’) to the weak stem * n- - plus the Germanic suffix *-(u)k-

used for body parts (cf. Krahe & Meid, 1967:213), and the zero-grade *knu-ka- (MDu. cnōk ‘bone, articulation, lump’, MLG knōk , kn k ‘bone’, Norw. knoke, Dan. dial. knoge, ‘knuckle, joint’, formed to the collective stem * n-u-). This suggests an original meaning of some protruding, rounded joint or articulation (although it cannot be ascertained to which and if it referred to one specifically). Note that it is not uncommon historically for words for body parts to be applied to other (somehow similar or proximate) body parts.

5

form can in any way be explained as secondary. Note that the reverse is not true: the securely attested gradation pattern of the stems * n-u/* n- - cannot be explained from analogy or leveling and must be original. Thus, the Latin form must be either analogical or a zero-grade, and the Hittite form was PD1 (which could easily have replaced original R(ó) with R(é)). I conclude, then, that the words for knee and chin are consistently separated in both form and meaning in all attestations. This consistent distinction in meaning and form between the two words and the occurrence of this pattern all over the Indo-European world prohibits any connection between the two into a single paradigm—the form for ‘chin, cheek’ with R(é) simply does not belong in the paradigm for ‘knee’, even though it probably derives from the same root * én- (‘joint’?). This piece of evidence for an AS pattern, then, must also be rejected.

I then follow Kloekhorst (2013) in that the AS gradation type probably did not exist, for the evidence adduced in favor of this gradation type does not hold under scrutiny.6 Instead, the consistent pattern in the neuter u-stems of the type *dór-u/* - -s shows another type of PD paradigm. I therefore posit another PD paradigm which I simply call PD2. This gradation type is supported by the comparative evidence and can be schematically represented as follows:

PD2 = | ́_  _{́ }|

‘blood’ ‘water’7

NOM.SG R(ó)-S() *dór-u * -r GEN.SG R()-S(é)-E() * - -s *ud-én-s

Note that this pattern is important in some reconstructions, because an athematic stem with R(ó) strongly implicates (but does not yet prove) a PD2 paradigm.

§1.1.3

T

H

T

In the Erlangen school the hysterodynamic or hysterokinetic (HD) form is exemplified by the word for ‘father’. The word is reconstructed as *ph2-ter- and has cognates in all

Indo-European branches, cf. Gr. πατήρ, Skt. pita, Av. pt , Lat. pater, Osc. patir, OIr. athir, Arm.

hayr; Toch.A, p ca , B pacer, etc., all ‘father’. The paradigm can be reconstructed as follows:

6 _{Note that this does not count for root nouns, for which AS gradation appears very plausible. However, the} matter of root nouns lie outside the scope of this thesis.

PIE Skt. Gr.

NOM.SG *ph2-t r < pita πατήρ

ACC.SG *ph2-tér-m < pitáram πατέρα

LOC.SG *ph2-tér-i < pitári —

GEN.SG *ph2-tr-ós < pitráḥ πατρός

DAT.SG *ph2-tr-ó < pitré πατρί

We find that the root is everywhere R() and that the suffix shows gradation S( : é : ). If we assume that short vowels lengthened at the end of a word before resonant (see §1.2, p.24), we can reduce this type to showing this pattern in the suffix: S(é : ), reflecting a strong and weak stem. The weak stem shows a full grade in the ending which is traditionally reconstructed as E(é). However, there is no direct evidence for an e-grade, but rather only of E(ó).8 I therefore follow Beekes and reconstruct o-grade (cf. Beekes, 1985:180).

In the Leiden school (Beekes, 1985:7, Kloekhorst, 2013:116f.) the nominative R()-S( )-, in e.g. *ph2-t , is as not regarded as original to the HD paradigm. The

nominative *ph2-t must under this view have originated analogically from the accusative

stem. There is not much evidence for this view, however, as evidence for e.g. a **péh2-t that

was replaced by *ph2-t is attested nowhere. This conception is purely out of theoretical

deliberations. The hypothesis that *ph2-t is secondary is admittedly not posited for PIE, but

for an earlier stage of the proto-language, however. This earlier stage is of no interest for my purposes and I base myself mostly on the actual evidence instead of on theoretical deliberations. The form of the word for ‘father’, then, is of PIE age and reflects a certain gradation pattern also found in other words, even if it once had a different nominative form and has a leveled vocalism in the nominative in PIE.

However, there is evidence supporting the view of a strong stem R(é)-S()-E() elsewhere. The evidence is always circumstantial, although by sound abductive reasoning (i.e. by inference to the best explanation), another type of HD gradation can be established. The word for ‘earth’ can be adduced as an example and has recently been discussed in detail by Kloekhorst (fthc.b:14ff., cf. Beekes, 1985:45f. and NIL:86ff.). Kloekhorst posits that the root of the form was *dh h- (earlier probably *dh -). The traditional reconstruction of * h - m/

* hð-m-ós (cf. the discussion in HIL:992f.) based on e.g. Gr. χθών/χθονός ‘earth’ must in my

opinion be reinterpreted as a collective *dh h- m/*dh h-m-ós (cf. §1.1.4, p.23 below on the

collective gradation).9 The important Hittite form NOM.SG tēkan [n.] ‘earth’ goes back to *dhé h-m presents us with a strong stem pattern of the singular. The Hitt. GEN.SG takn š must

go back to *dh h-m-ós, also found in Ved. jmás. This form provides evidence for a weak stem,

corresponding entirely to the HD pattern R()-S()-E(ó). Another stem is found in e.g.

8

The ON r-stems have a DAT.SG with i-mutation, e.g. feðr (to NOM.SG faðir ‘father’) and brøðr (to NOM.SG bróðir ‘brother’), which can be adduced to substantiate an original DAT.SG *ph2-tr-é with E(é). Nevertheless, this form may be mechanically created from the regular mutation in datives elsewhere and the existence of the alternant DAT.SG fǫ corroborates this. Moreover, the paradigm of faðir/fǫ points to a static gradation

paradigm with -S()-E() and thus follows the original gradation pattern of bróðir, which compromises the value ON data for HD gradation.

9 _{The strong collective stem *d}h h_{- m yielded χθών directly, from which the -n- < *m was introduced into the}

paradigm. The weak stem, e.g. GEN.SG *dh h-m-ós, is continued in χθονός with typical Greek innovation of a

CLuw. t amm-, Lith. m , OCSl. zemlja ‘earth, soil’, etc., which point to *dh h-ém-. This

stem corresponds in structure to the Ved. LOC.SG k ámi < *dh h-ém-i. In the HD paradigm of

*ph2-t , the ACC.SG *ph2-t -m corresponds to the gradation pattern of the LOC.SG *ph2-tér-i,

both exhibiting the pattern in R()-S(é)-E(). We can now understand the Vedic LOC.SG

k m in R()-S(é)-E() as reflecting HD gradation as well, from which we can deduce an early ACC.SG *dh h- m-m (perhaps continued in Ved. ACC.SG k m from *dh h- m <

*dh h- m-m by Stang’s Law). In all, then, there appear to be three stems: a strong stem

*dhé h-m on the one hand, and on the other a weak stem *dh h-m-' as well as *dh h-ém- which

correspond the the HD patterns described above. When we put the forms together, we obtain an alternative HD paradigm for ‘earth’:

NOM.SG *dh h-m

ACC.SG *dh h- m-m LOC.SG *dh h-ém-i GEN.SG *dh h-m-ós

It is, then, prudent to propose two variations of the HD type for PIE (at least for a later stage), which is in line with the evidence (as does Beekes, 20112:104, where the type *ph2-t is

called HD subtype 4). I shall refer to the two types simply as HD1 and HD2: HD1 = | ́   ́ 

  ́

| HD2 = |_ ̅́ _{́ }

  ́

|

In the HD1 paradigm we can distinguish three distinct stem patterns. The HD1 type has the pattern R(é)-S()-E(), which in HD2 was R()-S( )-E(), and is traditionally called the strong stem. The strong stem forms the nominative case. The weak stem has R( )-S()-E(ó) in both the HD1 and HD2 paradigms, occurring in the genitive, dative, and instrumental cases. Furthermore, the accusative and locative are formally identical in their pattern of gradation: R()-S(é)-E(). For the sake of convenience I term this pattern the ‘middle stem’. Thus, despite the proposals for different types of PD and HD, there are some fundamental differences between the PD types on the one hand and the HD types on the other. It is possible that originally the ablative case in the HD paradigms was also a middle stem. Kloekhorst (2013:115f.) gives the paradigm of ‘boundary, line’:

NOM.SG - a-aš ACC.SG a - a-an

GEN.SG a - a-aš

ABL.SG a-ra-a -za D/L.SG a - i

Disregarding the leveled endings with the theme vowel (grading as E(ó)) throughout the nominal system of Hittite, the NOM.SG - a-aš must go back to *h1ér-h2(-s) and the GEN.SG

a - a-aš to *h1r-h2-ós. We would expect a different ACC.SG, however, for a - a-an must go

back to a virtual *h1r-h2-óm, whereas we would expect a reflex of *h1r-éh2-m (i.e. **araa

back to *h1r-éh2-ti. As Kloekhorst points out, this pattern is inexplicable from the Erlangen

hysterodynamic model (HD2), and can only be adequately explained by the HD1 pattern. Note that it is therefore important to determine the original pattern of a ‘middle stem’ for the accusative, the locative, as well as the ablative whenever possible. In the example of Hittite

- a-aš, the ACC.SG took the stem pattern of the weak stem, but must have originally been of a different pattern which is still found in the ABL.SG. I conclude, then, that the HD1 and HD2

paradigms each consist of three distinct gradation patterns of the stem: the strong, middle, and weak stem.

§1.1.4

T

A

T

The amphikinetic or amphidynamic type (AD) is posited by the Erlangen school (cf. Schindler 1975a and b, Rix, 19922:123; Rieken, 1999:6; Widmer, 2002:36, Meier-Brügger, 2002:218ff.). The form *h2 s- s ‘dawn’ is adduced as probative evidence for this type. The

word is attested in Gr. (Hom.) ἠώς, Dor. ἀϝὠς [f.], Ved. ah- [f.], Av. šah- [f.], and in Lat.

a ō a [f. -st] (explained by EDLIL:63, as an -stem build to the leveled strong stem, e.g.

ACC.SG *a sōs-em > *a ō -em), all meaning ‘dawn’. From these forms we can deduce the

following paradigm:

PIE Ved. Gr.

NOM.SG *h2 s-ōs < - h ἠώς

ACC.SG *h2 s- s-m < - s-am ἠῶ, ἠοῦν

GEN.SG *h2us-s-ós < - ḥ ἠοῦς

Although Kloekhorst (2013:118ff.) wants to reduce this type via an extremely complex series of steps from a HD type, he accepts that the type existed in (late) PIE. That is, there is a broad consensus for the existence of this type in the attested Indo-European languages and I follow the consensus. I wish to stress a difference between this type and the HD1 type in the middle stem. In the HD1 middle stem we find S(é), whereas in the AD type we find S(ó) (cf. the Vedic ACC.SG *h2(e) s- s-m < - s-am with lengthening of *ó in open syllable by

Brugmann’s Law). As to the reconstruction of a middle R(e) or R() I tend to R() because of the columnar suffix accent in Vedic implying S(ó), but unaccented R(e) cannot be ruled out altogether. Thus, I posit an AD type as follows:

AD = | ́ ̅ ́ 

  ́

|

typical reconstruction for ‘water’ as holodynamic cannot be upheld. Kloekhorst (fthc.a) has conclusively shown that the Hittite data point to what I term a PD2 paradigm: NOM/ACC.SG

a-a-tar, a-tar < * - and GEN.SG - 5-te-na-aš << *ud-én-s, and with a NOM.PL

ú-i-ta-a-ar, ú-e-da-ar < *ud- . The reconstruction of a collective * - (cf. already Schmidt, 1889,

on the collective of heteroclites) from the Hittite forms is proved by Greek (N/A.SG!) ὕδωρ, Ved. N/A.PL , and Umbrian N/A.PL utur. This means that the PD neuters exhibit the

following pattern in their respective collective: HD3 = | ̅́

  ́|

I term this paradigm simply HD3, because the gradation is more towards the end of the word and it constrains the unnecessary proliferation of terminology.

With respect to the endings of the HD3 collective, I wish to point out the conspicuous correspondence between the endingless strong stem of the singular (e.g. N/A.SG * - ) and the endingless strong stem of the collective (e.g. N/A.COL * - ) in the neuters. This indicates that the endings of the singular and plural may originally have been the same. This would have facilitated the transfer between singular and collective and this is exactly what we find in e.g. Greek. Thus, the generalized heteroclite ending GEN.SG -ατος in Greek (ignoring

the secondary -τ-) and GEN.SG -naḥ in Sanskrit—which are conspicuously not PD—must then be continuants of the HD3 collective ending GEN.SG *-n-ós.10 This hypothesis is inconsequential for the arguments presented in this thesis, however. Note that Greek and Vedic generalized the GEN.SG endings in the nominal system, having replaced old PD endings

everywhere, and therefore the Greek GEN.SG -ατος and Vedic GEN.SG -naḥ cannot prove an

HD paradigm.

From the above deliberation I reconstruct the neuter PD singular stems with an “internal” HD3 collective as follows (for the details on the derivation of the employed forms see §2.2.2, p.38, and §4.2, p.100):

SINGULAR COLLECTIVE

N/A.SG * - * -

GEN.SG *ud-én-s *ud-n-ós

DAT.SG *ud-én-i *ud-n-

SINGULAR COLLECTIVE

N/A.SG *péh2 *ph2 (→ *puh2 )

GEN.SG *ph2 n-s *puh2n-ós

DAT.SG *ph2 n-i *puh2n-

10 _{This may furthermore explain other facts, most notably the occurrence of a Hittite}

This concludes the discussion of the relevant gradation types. One remaning issue is the contested origin of the lengthened grade. In the following paragraph I look at a proposal and offer a minor refinement.

§1.2

T

HE

L

ENGTHENED

G

RADE

Most of the long vowels that were once reconstructed for PIE have been explained by lengthening of short vowels due to contiguous laryngeals. The vowel system of PIE was thereby radically simplified, although a few consistent cases of long vowels remained unexplained. There are actually two typical environments in which these remaining long vowels occur: in monosyllables and word finally before a resonant. The first environment is at the end of a word before contiguous resonant (extended and adapted from Beekes, 20112:176, cf. already Kortlandt, 1975:84ff.): 11

1. in the NOM.SG of the neuter HD2 (§1.1.3, p.19), HD3 (§1.1.4, p.22), and AD paradigms (§1.1.4, p.22)

2. in the endingless LOC.SG of the PD1 and PD2 paradigms (e.g. Ved. s na ‘on the

mountain/ridge’ << *sn- with analogical root vocalism, probably also reflected in the Go. u-stem DAT.SG -au < *- )

3. in the 3PL.PERF ending where we find either *-r or *-ē (never **-er, cf. Beekes, 1985:153).

The second environment is in certain monosyllables:

4. in the NOM.SG of sigmatic root nouns (cf. *p -s ‘foot’ as evinced by e.g. Greek

Dor. πώς, Att. πούς, Skt. NOM.SG p t, ON fótr, etc.) 5. in the N/A.SG of PD1 or PD2 s-stems (cf. §1.1.2, p.15)

6. in the IND.ACT of the root s-aorist (e.g. 3SG * h-s-t ‘to drive’)

7. in the IND.ACT of the root aorist (reflected in the Germanic preterit of class IV

strong verbs, e.g. *gw m- > Go. qem-, cf. Lat. ēn t, Toch. B m-, cf. Kortlandt, 2012)

8. in the static presents (e.g. Skt. t ‘he does carpentry’).

Since the long vowels are notably rarer than short vowels and the fact that they appear in two specific environments, it is to be expected that these long vowels were the result of regular

sound law. In essence I follow the accounts of Kortlandt (1975:84ff.) and Beekes (1985:152ff., 20112:176f.) on the origin of long vowels.

The long vowels in polysyllables have to be explained under any view. Beekes (1985:152) holds that *-ER develops regularly at word end to *- .

1. LAW OF FINAL LENGTHENING: Short vowels before contiguous resonant at word end

are lengthened, so that *-VR# becomes *- R#, where R = /r, l, m, n, , / or:

*V > * | __R#

By this law we can explain e.g. the NOM.SG of HD2 words like *ph2-t as well as e.g. the

NOM.SG of AD paradigms like *h2 k -m n as the regular outcome of final lengthening. An

alternative to this view is that a final *-s or *H was lost after -VR with subsequent lengthening of the vowel in compensation, known as Szemerényi’s Law (Szemerényi, 1970:109). However, Szemerényi’s Law does not take into account the two conspicuous environments. Moreover, as Beekes (1985:151f.) explains, the alleged loss of *-s with compensatory lengthening is often unfounded, because it cannot be proven that (at least some of) the forms ever had a sigmatic NOM.SG. Szemerényi’s argument, then, is unfalsifiable. A good example of an HD paradigm without NOM.SG *-s are the animate *h2-stems, which are known to not

have had a NOM.SG *-s (see further Beekes, 20112:199f.). That is, there is no evidence at all to

suppose that some forms ever had a sigmatic NOM.SG, but they are lengthend, and the

lengthening can then not be explained by Szemerényi’s Law. This conclusively refutes the existence of Szemerényi’s Law.

Although the law of final vowel lengthening proposed by the Leiden school offers a very attractive solution, some problems remain. The first problem is found in the endings of the o-stems. The endings that are securely reconstructed, such as ACC.SG *-om and DAT.SG

*-o must have been lengthened under the law of final lengthening, but they evidently didn’t. Beekes (1985:184ff.) therefore proposes that the o-stems were formed only after the working of the law of final lengthening. This effectively solves the problem, but stratifies the Indo-European proto-language. Nevertheless, the o-stems are generally accepted to be a late PIE phenomenon, which makes the presence of the law in the (older) athematic paradigms wholly acceptable. A corollary to this solution is that the oldest stratum in PIE morphology will consistently show this pattern of lengthening (which appears to be the case, as I discuss below) which makes it a verifiable hypothesis.

2. LAW OF MONOSYLLABIC LENGTHENING. – Monosyllables lengthen a short vowel when

followed by either a contiguous plosive P or resonant R, but not a fricative *H or *s, plus another voiceless dental T = */s, t/ (including the dental clusters /st, ts/), or:

*-V- > *- - | #Cn __ T#

Note that this excludes monosyllables ending in simple *s or *t (also *st?) as well as monosyllables ending in only a consonant. Furthermore, a final *-d# was devoiced early on and then also caused lengthening under these circumstances (cf. the word *k t < *k below). Laryngeals checked by plosives (i.e. *#PHP- or *#PHR- before the vowel) that could vocalize prohibited lengthening. That is, forms of the type GEN.SG *sh2 ns never got

lengthened (for the reconstruction of this form cf. §4.2, p.100), although the initial sequence *#HC- before a vowel did not cause problems. That is, the initial consonant sequence (whether or not including a resonant) appears to be inconsequential. This behavior of laryngeals should not at all be surprising as laryngeals often played an important role in syllabification (cf. the role of laryngeals prohibiting Brugmann’s Law in Indic) and could vocalize (i.e. become themselves syllabic). A more serious problem seems to be consistently found in the GEN.SG, however. For instance, besides positing monosyllabic lengthening in the

NOM.SG *p -s ‘foot’ and * kw-s ‘voice’, Beekes (20112:209) reconstructs the unlengthened GEN.SG *péd-s, * kw-s without scruples. Nevertheless, the consistency in which the genitive

does not yield to the law of final lengthening is perhaps indicative of rigorous leveling or blocking. Whereas the NOM.SG typically had no correspondent stems (other than the GEN.SG if it had the NOM.SG ending *-s) the GEN.SG of e.g. the HD paradigm had more pressure from other weak stem forms with the same gradation pattern (cf. also the presence of root noun

DAT.SG *péd-i, * kw-i, cf. Gr. ὀπί, which could not lengthen the vowel). It seems plausible,

then, that the GEN.SG was rigorously leveled (or blocked the law) by analogy to other (weak)

stems that did not undergo vowel lengthening. This admittedly remains a weak point and requires further investigation. Despite these qualifications, however, I believe that the law is tenable for it holds very well otherwise.

If, as the Leiden school maintains, the laws of vowel lengthening functioned only in early PIE, and the o-stems were largely exempted because they were formed at a later point in the development of PIE, then we should expect to find the laws working in paradigms and classes that are indubitably archaic. One form that must certainly be old is the heteroclite noun for ‘heart’, *k /*k - -ós.12

The Hittite paradigm shows the original situation best:

N/A.SG kē < *k (where final *d > *t and thus could lengthen *é > * , cf. above), GEN.SG

ka aš < *k - -ós, and the interesting LOC.SG kerti < *k -i with short vowel (see

EDHIL:543ff.). That is, we are dealing with a kind of noun inflection that alternates  with *-i- that is otherwise unattested. The pattern is intact in Hittite, endingless as well as i-stem cognates are found beside each other in other dialects, and since this extremely rare pattern

12 _{I am skeptical of the appurtenance of *k - in a compound *k -d}h_(e)h

1-, cf. Ved. a h - ‘trusting’, Latin.

c ē ō ‘believe’, etc. This form would prove that we are really dealing with a morpheme *ker- + (a suffix?) *-d-,

but this would be an entirely isolated suffix. The LOC.SG kerti < *k -i instead of **k r-éd-i also militates against this interpretation. Perhaps the form can be related, however, if Mayrhofer (KEWA III:387) is right that *k

-dh(e)h1- was changed to *k -dh(e)h1- to avoid the conconant sequence -rdzdh- (cf. *h2 -, in e.g. Go. aukan ‘increase’, but *h2 -s-, in Go. wahsan ‘to grow’, with relocation of the vowel due to the *-s-) But then this form does not provide evidence for an alternant **k - either.

can hardly be explained as an innovation it must be old. Sanskrit has the doubtlessly related but problematic form h ‘heart’ (with inexplicable initial consonant and -i from the weak case forms, adjoining the type a hi, on which see §2.1.2, p.31), GEN.SG h aḥ without *-i-,

but cf. Ved. h aya- ‘heart’ and h ya- ‘tenderly loved’ with *-i- (likewise OIr. cride [n.] ‘heart, middle, love’ < *k - -ó-, etc.).13

The long vowel survived in Gr. (Hom.) κῆρ ‘heart’, which formed an independent word (cf. Hom. LOC.SG κῆρι), cf. OPruss. seyr ‘heart’, etc. The

weak stem *k - - is also continued as an autonomous word in Gr. (Hom.) κραδίη, Ion. καρδίη ‘heart (as the seat of rage)’, etc. The form Gr. ἀκήριος ‘heartless, worn out’ combines the long grade with the suffix *-i-. Many more forms can be adduced to further corroborate what has already been shown (cf. NIL:417). The comparative evidence proves original heteroclisy of /i and must be reconstructed for PIE as follows:

N/A.SG *k

GEN.SG *k - -ós

LOC.SG *k -i

The difference of the vowel length of N/A.SG *k (or rather *k t, cf. above on devoicing)

and LOC.SG *k -i can now be easily explained as resulting from the law of monosyllabic

lengthening. This shows that this archaic noun used to have two distinctly graded stems, one with R(é)-S() and one with R()-S()-E(ó). This can only be aligned with the strong and weak stem of the HD1 gradation. However, an HD1 type normally exhibits a middle stem for the ACC.SG and LOC.SG with the pattern R()-S(é)-E(). Oddly enough, the Vedic ACC.SG

h is identical to the NOM.SG h and the LOC.SG gradation pattern of h aligns with the

other weak cases, but not with the ACC.SG. This is more reminiscent of a PD gradation type

and is therefore not likely to be old. It is important here to point out that the otherwise inexplicable Hittite LOC.SG kerti < *k -i (cf. Palaic k t < *k -i, NIL:417, 420) with R(é)-S() must continue the older form. This means that the ACC.SG and LOC.SG actually do share the same gradation pattern (i.e. R(é)-S()) as we would expect for an HD1 paradigm, but that this pattern is identical to the strong stem. The LOC.SG *k -i must have originally

graded like the N/A.SG *k t < *k . Note that all have R(é) and do not have the suffix *-i-

(not to be confused with the LOC.SG ending -i).14

In Germanic we find an n-stem *hert-ōn, cf. Go. hairto, ON hjarta, OS herta OHG

herza, OFri. herte, hirte, OE heorta, ODu. herta, etc. The form became part of the neuter

n-stems, a small but resilient class of nouns mainly denoting body parts, cf. *a -ōn, ‘eye’, * an -ōn ‘cheeck’, etc. The short vowel in *hert- can be explained as the result of Osthoff’s Law of vowel shortening before resonant plus consonant. Alternatively, we can think of the leveling of the locative stem, although it is hard to see why exactly this stem would become productive. There are no traces of the old /i-inflection in Germanic.