indo·

european

by William F. Wyatt, Jr.

 The Seventh Publication
in the Haney Foundation Series
University of Pennsylvania
Indo-European /a/
BY WILLIAM F. WYATT , JR.
 Publicationofthia book has been
madep!:*iblebyagrantfrom
the Haney Foundation of the
t:ntvusityofPennsylvania

Cof17ntlrt © 1970 ~ till Tausrus Ol' TH!: UNIVI:IWTY 01' P!:NNIYLVAm.t.

Ubrary of C<>~~tress Catrdot C4rd Nwnbn 73:83/40
SBN:8l22-759-i--2

Printed in the United States of Amerlc&

Indo-European / a/
 Abbreviations

R_ f:.mik Boisacq. Dictionna ire hymo logique de Ia langue

g rccque . H eidel berg, 19 14.

E. \ 1 Alfrcd ErnoUI and A nt oine Meillet. Oictionnairc

ttymologiquc de Ia lang ue latine. ·hh cd. Paris, 1939.

Hjalmar Frisk . (;ricchisches ctyrnologischcs \Vi:irterbuch .

ll eidclberg, 1954 .

IF l ndogerrnanisch<' Forschun gen . Berlin .

]AOS Journal of th t: American Orient a l Societv. :'\ew
Haven

/\/. Ze itsc hrift fUr vergleichende Sprachforschung GOt-

tin~en .

M. \tanfred M ayr hofer. K urzgefasstes ety mologische;

\\'Orterhuch des Ahindisc hen. H eidelberg , 1%3- .

P. J ul ius Pokur ny. lndogerrn anisch c~ etymologisches W(ir-

terbuch . Vol. I. Bern & .\1uni1:h, 1959.

C C. C. Chlenhcrk. Kurzgefasstt>S etymologisc hes \\'ilrter-

huch d er :\ ltindischt'n Sprache. Amsterdam, IH9B

\\"-H :\ Waldt·. l.ate in isches etymologisdws \\"ilrlt"rbuch .

1d ed. rev., by J. B. Hofmann. H eide-l berg, 1938-
19:>4.
,' a / has always been of central concern 10 lndo-Europcan-
ists. The first scholars in the last centul~' seriously to at-
tempt the reconstruction of the Indo-European parent
tongue deriwd all the non-high vowels of the European
languages, that is to say i c a o / , from an original / a
which contrasted in Indo-EuroJJean only with the hig-h
vowels / if and / u / in a thret'-vowd systt'm. Later, as com-
parative methodolog-y improved , and as the hold of San-
skrit on the minds and hearts of linguists began to weaken ,
Indo-European ,' a / was split up, first into two \"Owels ,'e l
and ,'o 1 , still nostalgically written n 1 and az by Brugmann
in 1876 and de Saussure in 1879; and then finally into lOur
(/e a o / and the import from the Semitic world, sdw ·a ).
This stage of six short vowels, save for later isolated aber-
rations, reprt'sented the highwater mark of the Indo-Euro-
pean vocalic system , a point from which that system ha"
since been receding. Though / e / and 'o/ together with
/ i/ and / u/ , frequently regardc:-d as the vocalic allo-
phones of .' y/ and ,'w /, have successfully maintained their
status as original clements of the vowel system, both ; a '
and .' J ; have been expelled hy many, / J/ alone by some. 1
In what follows I shall discuss these displaced vowds. in
the hopes of findin g a home for at least one of them in our
grammars of Indo-European. I shall not be primarily con-
cerned with how they were realized phonetically, but
rather with their position in the system, that is to say, with
 Indo-European /a/
the number of paradigmatic oppositions into which they
entered. Did the Indo-European vowel system contain
four distinctive vowels to be written / i u e o/ as laryn-
gealists would have it; or five: / i u eo a/, as in Burrow's
system; or six, as in the traditional analysis w hich includes
/ i u eo ad/?
The answer to this question of course derives ultimately
from an inspection of the data, namely the correspond-
ences between words in the daughter languages. For the
sake of simplicity I cite forms only from Greek and San-
skrit: cognates in other languages could easily be cited,
but Greek and Sanskrit are here sufficient. 2
Cognates Solutions
Gru k. Sanskrit Traditional Laryngeal Burrow
LjuP imal:t
tvy6P yugim
cpipw bhfu-ami
Ots iviJ:t
'e/Ha-
ii:yw ijami e/Hz-.
wa"Tljp piti H/G_C
Since there are six cognate pairs, there is a maximum of
six vowels to be restored to the parent language. Beside the
cognates I list the three phonemic solutions to these cor-
respondences which seem to be in the field and worthy of
consideration today. The first , the traditional, directly and
simply accounts for the correspondences by rewriting one
symbol in the protoform with one symOOl in the various
daughter languages. The laryngeal theory, in one of its
less complicated form s, likew ise accounts for the corre-
spondences, but by combining two symOOis (or phonemes)
alread y in the inventory (namely / H / and /e/) to provide
one of the correspondences: phonetic [a] is phonemic
 Indo-European / a/
/ H2e/. /o/ existed in the system a lso, but I have not given
an example of it in the table: / H 2 o/ would presumably ap-
pear a lso as [a].3 Burrow's solution, a five-vowel solution,
is more radical and does not directly account for the
oorrespondences.
Burrow ( 1949) was moved by the small number of cor-
respondences requiring /d/ to deny / d/ to Indo-European,
not by proving it a variant or development of some other
sound, but rather by reanalyzing the correspondences on
the basis of which /d/ was postulated. Thus one of the cor-
nerstones of the doctrine of /d/, warftp = pitli 'father' is
removed by assuming ( 1949: 38- 39) that both pit!i and
duhi!A 'daughter' ( cf. Gk. i}vy6:T71P) contain an 1-1 suffix-ilti,
and thus are to be analyzed etymologically as p-itli and
duh-itiJ, rather than as pi-t!i and duhi-tiJ, the most usual ana-
lysis. And in The Sanskrit Language ( 1955 : 139) he suggests
that pitli. may contain the weak grade of the root pii-(y)- 'to
protect.' But whether or not one accepts this etymology,
Burrow holds that these words and all others formerly
thought by other scholars to contain IE */d/ contain IE
*/i/.4
Though Burrow's instinct was correct, and though
mar:ty of his individual arguments are stimulating, none-
theless his main oonclusion has rightly found little favor.
Denying /d/ to Indo-European is one thing, but denying
the relatedness of the co~:,>nates on which it is based is an -
other. In the former case we discard a phoneme, in the lat-
ter we discard an entire method together with its body of
theory. And yet that is what Burrow in effect has been
forced to do. One might as well propose (e.g.) that the / o/
ofGk. yivo<; 'race' does not correspond directly to the / a/
of S kt. ;anab 'race' because the Skt. form contains a suffix
-ab seen also in mana~ 'mind' and other words, and that
 Indo-Eu ropean /a/
therefore the suffix is secondary in Jana~ . And this means
assuming that on ly the *gen-part of the words is related,
or in the case of 'father', that only the initial / p/ 's corre-
spond. It is indeed possibl e, certain in fact, that Sanskrit
has occasionally introduced /i/ into positions in which it
did not originally appear, but a wholesale substitution of
Iii for 0" in kinship terms ( pitfl), in participia l forms
(sth llti/:1.), and in disyllabic roots (punimd~) seems most un-
likely. Burrovv's solution will not do.
The simplicity of the traditional solution is sufficient rec-
ommendation for it, but what of the laryngeal solution
with its denial of /a/ to Indo-European (or, in the think-
ing of some lal)'ngealists, to an earlier stage of it)? Why do
laryngealists feel that [a J results from a cluster / H 2e/ and
was not a single phoneme? Th ere are various theoretica l
reasons for this assumption , all deriving from the theory
itself and not from the comparative data. These reasons I
have discussed in a n article in Language (Wyatt 1964), and
hope to have indicated that they are not sufficient to sup-
port the doctrine. But Kurylowicz ( 1956 :187- 195, 1962:
114) has perhaps most recently expressed what he consid-
ers good practica l and statistical reasons for adopting the
view that [a]= / H 2 e/ . He points ou t first that / a/ is dis-
tinguished from /o/ in only a few languages, and that /a/
is of secondary origin in the most important cases in those
languages. The implication is that /a/ is a lways of secon-
dary origin. This is of course not a serious argument for his
conclusion, but leads up to others which are. [a ] ten ds to
appear in words of concrete character which do not in gen-
era l belong to the basic vocabulary, words like goose bean
nose salt. Few words of this type can be connected with
verbal roots, most appear in nonablauting paradigms, and
OOrrowings from a/o-languages into o-languages (and vice-
 Indo-European / a/
versa) may well ha ve played a role in causing us to recon-
struct a n / a/ where we should reconstruct /of . Third, the
other side of the same coin, !a ] does not occur in pronouns,
numerals, names of parts oft he body, k inship terms- th us
not in the basic vocabulary. Both these statements con-
cerning the distribution of /a/ in the vocabulalJ' are cor-
rect, but they do not really prove anything. It would be
pleasant to think that we are able to explain why the IE
languages did not utilize [a] in pronouns, etc., but we can-
not, and we might with as much justification object to the
doctrine of / Hd tha t / H 2 e/ does not occur in pronouns,
etc. We cannot in general explain the absence of a ten-
dency or artifact.
The fourth criterion is the most important, and so far as
I know is original with Kurylowicz . Neither he nor anyone
else denies the validity of correspondences involving initia l
[a] such as: Skt. djat1 'drives', Gk. li.yw ' lead '; Skt. djroi}
'field', Gk. tiyp6s 'field'; Skt. ti4a~ 'axle', G k.li.~wv 'axle'.
But, says Kurylowicz ( 1956:189), if the [a] reconstructed
fium these form s and others were really a separate pho-
neme / a/, we should expect to find ten times as many in-
stances of it in the interior of the word as initially: words
with a n initial vowel constitute only about 10 per cent of
the IE vocabulalJ'. We should then expect to find some
220 or 230 roots with internal / a / , but in fact find only
about thirty. This, accord ing to Kurylowicz, proves that
/ a/ did not exist as a separate entity in Indo-European,
and t hat the init ial correspondences are to be analyzed
othenvise, the otherwise being / Hze/ in keeping with the
doctrine that all IE words began with a consonant, and
that every verbal root contained the / e/ vocalism. Thus
/H 2 ego:/ bcromes precisely parallel to / !ego:/, /bhero:/,
and so forth.
 !'fl.do-European /a/

But are these figures and percentages corroborative of

the conclusion they are meant to support? Frequency of
phonemic occurrence is likely to be erratic in any event
(SzemerCnyi 1964:9), and one need look no further than
Latin in his search for languages with relatively frequent
initial /a/ but /a/ rare internally (in noninitial syllables).
But to stay closer to Indo-European, suppose one were to
apply the same calculations tononapophonic /e/ and /o/,
that is to say to /e/ that does not alternate morphopho-
nemically with /o/, and /o/ that does not alternate
morphophonemically with /e/. Surely he would find al-
most exactly the same situation: a number of /e/ and /o/
appearing initially in very well attested roots, but next to
no such cognate sets internally. 5 And yet no one would
wish to deny /e/ and /o/ to Indo-European. Kurylowicz's
argument is not compelling, but endeavoring to account
for the distribution of [a J which he stresses leads one to an-
other conclusion which both elucidates his observation and
leads to a proper appreciation of the IE vowel system.
Indo-European [;;~] has almost precisely the opposite
distribution from [a]. It never occurs initially, but is ex-
tremely frequent in the interior of a word, generally unac-
cented. It seems indicated, on the basis of the distribution
of[a] and [:;)J to join them in one protophoneme which ap-
peared as [a] initially, but which was raised (or weakened)
to[:;)] in the interior of the word, at least in Indo-Iranian. I
offer the hypothesis that [a] and [:;)] are allophones of one
original phoneme which we may write /a/. There were
then five oppositions in the IE vowel system, oppositions
which can be best represented by the phonemes /i u eo a/,
the vowels, that is, which arc found later on in Latin,
Greek, Celtic, and Armenian.
There are a number of prima facie reasons which tend
 /n.do-European / a/
to support the assumption that [aJ and [;)J were originally
members of the same phoneme.
I) First, Indo-Iranian is the only branch of the IE fam-
ily to show a distinct reflex of[;}]: all others show the same
reflex of [a] and[;)]. This fact might tend to suggest that
Indo-Iranian has innovated in this matter, a suggestion
rendered the more likely by the fact that Indo-Iranian is
also the only language group to show a merger of / e/ and
/a/, a merger which might well have involved only the
accented or lower allophones of / a/. The merger may be
displayed as follows:
I ii / u/
[>] /il [>] / u/ /il / u/
/e/ /o/ ~ /e/ [a] /o/ /a/
[a]
That is to say, an original IE system of three distinctive
vocalic levels was collapsed into a two-level system . Upon
the merger of all low vowels, [;}J, the higher allophone of
/a/ , was split off from /a/, and, because it was a high
vowel, merged with /i!.
2) Second, from a purely IE point of view, we are led
to reconstruct six short vowel oppositions if we include
schwa, and only five long vowel oppositions. We might
rather expect either the same number of oppositions (as in
Latin) , or fewer short than long (as in Greek). The laryn-
geal theory, to be sure, handles this asymmetry easily by
assuming that [J] is merely the vocalic allophone of / H / .
3) As Kurylowicz and others have pointed out, and as I
have already mentioned, [aJ is rare internally in the word,
but frequent initially. This fact should indicate that / a/
appears in another guise in the interior of a word. And
since [;)J never occurs initially, it seems indicated to assume
10 lndo-l!.:urop(an Ia/
that [a] and[;)] are different allophones of the same pho-
neme. We shall attempt a more formal proof of this hy-
pothesis later on.
The idea that [a] and [d] are variants ofthe same pho-
neme is far from new. De Saussure in fact suggested it back
in 1879, though his pronouncements on the subject arc not
altogether clear.6 As is well known, he established for Indo-
European two new 'coefficients son antiqucs' A and g
which combined with/eo/ to yield / C a 6/ (< eA, oA;
<cA, oA; < cq, OQ rcspectively)just as / i/ and / u / com-
bined with these same vowels to yield /ei oi eu ou/. A, for
we may omit g here, between consonants and initially is
continued in all the daughter languages by /a/ (or its con-
tinuation /o/ ) (1922: 110). Thus Ag yields Lat. tigo, Gk.
liyw, Skt. d]ati. But under certain other conditions it ap-
peared as A, and it was this A which accounlS for the cor-
respondence Skl. sthitdl}, Gk. oTaT6~, Lat. status where we
fmd Skt. / i/ = Eur. / a/ rather than / a /= / a / (1922:
110- 111): 'On pourra done sans crainte etablir Ia rCgle
que, Jorsque les langues curop6ennes ant A, en syllabe ou-
verte comme en syllabe fermee l'arien montre a bref Mais
ceci veut dire simplement que \'a n'est pas rm a long: il
arrive en effct que dans certaines positions, par example a
Ia fin des racines, ce n 'est p lus du tout un a, mais bien i
ou i', au mains en sanskrit, qui se trouvc place en regard du
phoneme A des langues d'Europe. ' And further ( 1922:
226): ' Nous avons reconnu dans ce demier (sc. Skt./i/ in
rthitd.l}) lc descendant d'unf: voyelle faible proethniquc
designee parA, voyelle qui n'cst elle-meme qu'une modifi-
cation de I 'espece d'a, au des especes d'a autres que a 1 et a2
(A q).' And (1922:167): 'Nous crayons que cctte voyelle
i:tait une esptce d'e mutt, provenant de l'aitiration des phunlmes
A el e.' He was quite insistent on the fact that this ·1 arose
 Tndo -European /a/ II
from a full vowel ( 1922:1 67): 'Dire que Ia voyelle faible
proethnique d'oll dCrive l' z de sthitd, ristd, n'a point CtC
d'abord une voyelle pleine serait renonccr a expliquer l'a
de sthtiman, r&ti, dont elle forme Ia seconde partie.' He was
not clear, though, as to what brought about the alteration
of the phonemes A and o to A_ H e simply states ( 1922:167,
141) that this change was genera l at the end of roots, par-
tial in roots ending in a consonant. H e explicit ly denies
(1922:166) that the position of the accent could explain
the change completely, though he did agree that accented
A appeared always as Skt. / a/. What was needed at this
point was an analysis of the distribution of A showing un-
der what cond itions it passed to Skt. /a/ and under what
conditions it passed to /i/. TI1is de Saussure did not
provide.
De Saussure also posited two origins for his A, and it
was the origin of A, and not its nature, which prevented
later scholars from accepting his ideas. H e maintained that
the /i/ of the weak forms ofSkt. ninth-class verbs (punimd~
'we cleanse') and th e /i/ of the agent nouns (pavitdr) were
phonologically identical with the /i/ of sthitd~: al l of these
/i/ derive from A, which in turn derives from A. But the A
of the ninth-class verbs stems from an original full-grade
vowel, while the A of sthitd~ is the reduction of a long
;a/( 1922:226): 'D'autre part il y a entre l'f ou A de sthitd,
pitd, et l'l ou A de pam-, grabhi, cette importante diffCrence
morphologique, que le premier rbiulte de Ia reduction d'un
a{a 1A), tandis que lc second parait cxister de fondation a
l'Ctat autophtongue. S'il se combine avec a 1 , dans le prC-
sent en -mi, il n'en preexistait pas mains ace prb>ent'. Thus
some IE A are reductions of ;a; ( <*eA), while others arc
vowels A quite on par with/eo/. Th e type represented by
Ag- with A in initial position de Saussure considered to be a
12 Indo-Europmn /a/
weakening of /fl/ ( 1922: 150). A schematic representation
of de Saussure's ideas would show (I922: 164ff.):

II III
A A > I a/ (genemlly)
A>
A A
>Iii (at the end of roots)

The reason for de Saussure's a'isumption that the A of

ninth-class verbs was an origina l vowel and not also the
weakening of / 3/ was a morphological one. In Skt.
seventh-class verbs there occurs an alternation -na- - -n-,
e.g. inyumikti- yu~jmd~ 'join'. De Saussure felt that these
forms were to be analyzed as root + -nt - - -n- + ending
(yu-rui-k-ti- yu-il-j-mdl}) and not root + -n- + ending
(yu-n-dk-ti- yu-il-j-md~)- In other words the ablaut varia-
tion affected the infix and not the root, and the root re-
mained constant throughout asyuj-. He felt that the same
-ne- - -n- occurred also in fifth - and ninth-class verbs, in
the shapes -ne-u- - -n-u- (st(TJ6ti ,.....,. stroumd~) 'strew' and
-ne-A-- -n-A ( punfiti- punimdiJ) 'cleanse' respectively.
And since the root form appears as such in the plural with
only a nasal infixed, the root forms of these verbs must
have been stru- andpuA-. If so, the / ft./ of ninth-class singu-
lars must have resulted from the contraction of -ne- and
-A-, two vowels. In this case, then, the vocalism ofthe weak
forms is original, and that of the strong forms secondary.
In the case of stii- _. stA-, stii- ( <*steA-) was original and
<
stA - ( * stii-) secondary.
The doctrine of the original full-grade vowel A was only
so strong as the analysis of nasal-infix verbs, and some
scholars cou ld not accept de Saussure's views. Thus Streit-
berg (1915:208) favored analyzing these words as more
 Indo-European /a/ 13
literally nasal-infix verbs: i.e. verbs with only a nasal in-
fixed, and concluded that punb.ti:puninui}; were to be ana-
lyzed: pu-n-fl-ti:pu-n-i-md};, in which event -i- again could
be interpreted as the reduction of -d- : 'Man kann jedoch
mit gleichem Rechte auch folgende 11-ennung vomehmen :
*bln-n-id-mi, *pu-n.fl-mi, *k11i-n-iu-mi. Bei dieser Auffassung
existiert das auslautende i dersi{-Wurzeln nicht dtfondation
a l'itat autophtongue (S. 242), sondem ist das Ergebnis dcr
Rcduktion einer betonten Liinge, ist also wie das i von
sthi-ta- zu beurteilen'. Thus all A are to be regarded as the
reduction of ;a;.
The notion that A was the product of the reduction of
an unaccented long vowel also caused t rouble for de Sa us-
sure. H i.ibschmann ( 1885: lff., 65ff., 1900:27- 28) quite
correctly poi n1ed out - something known a lready to de
Saussure-that long vowels were reduced only in unac-
cented syllables, and that therefore there should be no
cases of accented A ( 1900:28): 'Bestand aber ein hoch-
toniges idg. a ( =skr. a, griech. a), das natiirlich ,...-eder mit
dem hochtonigen idg. 1 = e<:L ( =skr. a, gricch. il) noch
mit dem daraus (in unbetonter Silbe!) en tstandenen q. (;))
idemi.sche sein konntc, so muss an diesem a das ganzc
System de Saussu re's scheitem , d as a llen Ablaut auf den
Wechsel von o: e: - zu riickfLihren will. Denn idg. ig 16
kann wedcr = 6ag 1 6 noch eag 16 noch -ag6 sein .' 7 But of
course there are cases of accented [a]. For these Hi.ibsch-
mann .rcserved the symbol a, and symboli7.ed the reduced
long vowel by J . He was the first to my knowledge to posit
the two sounds as distinct phonemic entities.
It is well a t th is point to emphasize again the fact that
Hiibschmann was talking about the origin of A and not its
nature , and concluded som ething about its n ature from its
origin. It is of course quite JXISSible for a single sound to
result from the merger of two quite clearly distinct earlier
14 Indo-European Ia/
sounds, and arguments about the origins of sounds have
no place in discuss.ions concerning tlteir nature. What
Hi.ibschmann in fact did was to establish two morpho-
phonemes, a and J, morphophonemes which enable one to
infer the ablaut series to which the form in question be-
longs: forms with il always have beside- th em other forms
with a long vowel, while those vvith a do not, at least not
necessarily. This may indeed by a useful sort of conven-
tion, but it has nothing to do with the more or less phone-
mic analysis of reconstructed systems.
Bechtel next took up de Saussure's idea ( 1892:240 ff.),
and endeavored to get around HU.bschmann's objections
to it by positing an intermediate stage a in the develop-
ment from a to J. H e, too, accepted de Saussure's position
that there was no original IE accented a, and that ; was
merely the result of unaccented a . But he added an accent
shift to the picture ( 1892: 252): 'So driingt sich die Ver-
mutung auf, class das arische a iiberhaupt nicht durch das
folgende y, sondern durch den Accent bcdingt sci, der in
Folge spiiteren Versch icbung das a zu ciner Zeit getroffen
babe, als di e Vcrwandlung des arischen a in z noch nicht
vol lzogen war'. He thus still he ld to the doctrine that a
arose only from the reduction of an original long vowel.
His view can be represented schematically:
1) a > a
2) Some a receive the accent
3) a> a
a>:::~(Skt. i -1)

·nwugh Bechtel's assumption of an intennediate stage a

should have freed him from attacks based on the origin of
a, that is, on questions of ablaut series, he was nonetheless
so attacked,s and it remained for Pedersen (1900:75-86)
finally to free the question from its ablaut bondage. Peder-
 lnd(!-European I a/ IS

sen quite clearly and correcdy saw that ( 1900: 75) the dif-
ferent ablaut series can prove on ly that /a/ had several
sources, not that it had two different pronunciations. And
since Indo-Iranian is a lone in showing a divergent devel-
opment of a and~ ( 1900:76): 'so blcibt nur zu unter-
suchen, ob d ie doppdhcit 1 :a auf arischen boden durch ein
speziell arisches lautgesetz entstanden sein kann. Falls cine
solche mOgl ichkeit existiert, ist jede andere erkllirung
abzulchnen; besteht die mOglichkcit cines arischcn laut-
gesetz nicht, so darf auf cine idg. doppelheit geschlossen
werden.' With ablaut considerations resolutely banished
from the field, only phonological equations were con-
sidered. And on the basis of purely phonological evidence
Pedersen concluded (1900:85): 'ein un betontcs idg. a in
ofTener (nicht auf sonorlaut ausgehender) silbe geht im
Arischcn in i tiber, wcnn cs wedcr im an laut noch im
auslaut stcht, nicht auf;· v k. (g) folgt und nicht unmittelbar
vor y steht.'
Pedersen's viev-..-s have found little favor. Brugmann
( 1904: 80), without arguing the matter, simply states that
he is unconvinced; Hirt (1900: 148-149) demonstrated
that the rule that a does not pass to 1 after v is contravened
by Skt. tai.N;. 'strong' but tavl.fli~ 'id.'; and Gi.intcrt (19 16 )
devoted some twenty pages to a detailed attack on Peder-
sen's formulation. But Pedersen attempted to establish his
opinion further ( !90S: 398-402 ), included it in his dis-
cussion of Semitic affinities of IE ( 1907: 349), and main-
tained it still in his large Celtic grammar ( 1909: 30). So
far as I know he never gave it up, though he did modify it
in one important regard (1926: 27), as we shall see below.
And it is clear that certain aspects of Pedersen's formula -
tion are incorrect or unnecessary. Preceding /y/ can have
had nothing to do with the preservation of / a /. The rule
is in the first place unnecessary because needed to account
16 Jndo-Eurapean / a/
only for the two correspondences: Skt. ydtati ' attach',
Gk. t 1J'Tiw 'seck' and Skt. ydjatt 'worship', Gk.liKop.at 'stand
in awe of'. But the first of these equations must be dis-
carded {see App. [J t26), and the / a/ of the Skt. form in
the second equation is sufficiently accounted for by the
accent. Furthermore there are counterexamples which
show that the Skt . outcome of IE • / ya/ was something
other than /ya / . Giintert ( 1916:6- 7) mentionsjita < jyii-
'overpower', Gk. {3titw ' constrain' andpind < pyii- 'fill up'.
To these can be added the internal Skt. evidence of words
like ipsati ' wishes to obtain', the desiderative of iip- in
iipn6ti 'obtains' There are also the excellent cognate pairs:
Skt. bharantl, Gk. q,ipovoa < *bheronlia 'carrying (fern.
part. ); Skt. tri, Gk. 1pla < *tria 'three (ntr. plur. )' ; Skt.
krltd.~ 'bought', Gk. '1Tptm6~ < *kwri.at6s. All these words
show that IE* / ia/ (or /ya /) passes to / i: I in Skt.
A preced ing semivowel can have had no effect on the
preservation of / a / . Nor can a preceding / k/. It is indeed
interesting and significant that / a / appears with some
frequency in in itial syllables after /k/ and that [d] does
not , but this fact has more to do with the question of the IE
gutturals than with the IE vocalic system. / a/ > [d] after
the other velar consonants (Pedersen 1900:77-78) and it
is unreasonable to su ppose that the voiceless velar had an
effect on the succeeding phoneme not shared by its voiced
and aspirated counterparts. Hence we must strip Peder-
sen's foiTilulation at least of the conditioning effect of the
preceding phoneme: we shall reserve judgment on the
other aspects of his formulation till after we have examined
the ev idence.
Later scholars seem by and large to have retained
HU.bschmann's view of the IE vowel system, however they
may differ in their interpretation of it, and to have ignored
Pedersen 's view. I presume that t he reason for this neglect
 Indo- European / a/ 17
of Pedersen is that many scholars felt that Giintcrt'scriti-
cisms were suffit:ient to destroy his position, and that
furthermore some scholars at least felt that the laryngeal
theory, buttressed an ew by the evidence of Hittite, pro-
vided a more powerful explicatory device. In any event,
the more traditionally-minded lndo-Europeanists, such
as Krahe (1962:54), retain both /a/ and j;J / . Laryngeal-
ists are split into two camps on the question. T he more
orthodox view holds that Ja J arises from / H 2 e/ (and / H 4 e/
for those who accept /H4/ ), while (;J] develops from any
/ HI between consonants. A somewhat rad ical view main-
tains that there was no schwa in Indo-European and that
consonantal / H / did not develop to vocalic schwa. Those
who subscribe to this view (Burrow 1949) derive [a] from
/H~e/ in some cases, and also from /a/ . I propose in what
follows to examine the evidence for IE / a/ and schwa, in
hopes of providing an acceptable p roof that IE [a] and [;J]
are allophones of the same phoneme, and that /a/ de-
veloped to [cl] only under certain conditions.
I. According to Pedersen's view IE* / a/ in absolute initial
position appears as / a / everywhere. To support this view
he gave ( 1900:76-77 ) Brugmann's list ( 1897 : 158-1 63).

I. Skt. d;ati 'drives', Gk.liyw ' lead' (M . 1. 23, F. 1.1 8).

2. Skt. Ujral} 'field', Gk. &yp6~ 'field' (M. 1.23, F. 1.1 6).
3. Skt. dkgY_1 'axle' , Gk. lX~wv 'id.' (M. 1.1 6, F. Ll 16).
4. Skt. dpa 'away, off', Gk. &'ll"6 'from' (M. 1.37, F.
l.l 22) .
5. Skt. dnti 'before', Gk . iwri 'id.' (M . 1.36, F. 1.113-
114).
6. Skt. dfma 'stone', Gk. liK11-wv 'anvil' (M. 1.60, F.
1.54).
7. Skt. catur-a.ird/.1 'four-cornered ', Gk. liKpm 'end '
(M . 1.61 , F. 1.59- 60).
IS lnd!J-f:uropmu / a/
8. Skt. dTiJha/J 'anxiety', Gk. liyxw 'press tight' { ~. 1.1 4,
F. 1.17- 18).
9. Skt. tiytf; 'iron, metal ', Lat. aes 'copper' (M. I.46,
W-H 1.19-20).
10. Skt. dniti 'breathes', C k. iivl/-W'> 'w ind', Lat . animus
'mind' (M. 1. 33, F. I.IO.i, W-H 1.49-50).
II. Skt. amid& 'sou r', Lat. am&rus 'bitter' (M. 1.46, W-H
1.35).
To this list Pedersen himself adds:
12. Skt. dim 'teat', Gk. OOKpv •;d.' (M. 1.60, F. 1.344).
13. Skt . arghdiJ 'worth', Gk. &Aqxhw 'acquire' (M. 50,
F. 1.81 ).
14. Skt. dsrk ' blood', Lat. asser ' blood ' (M. 1.66, W-H
1. 72).
Still more examples are prov ided by Ku rytowicz ( 1956:
137- 188)•
15. Skt. ditka& 'curve', Gk. liy Ko~ 'bend' (M. 1.19, F.
1.10- 12).
16. Skt. ajd& ' he-goat', Oir. ag 'cow, deer'(P. 6- 7, M.
1. 23).
17. Skt. dnu 'after', Gk. O:vO: ' up' (M. 1.34, F. 1.100).
18. Skt. dndhal; ' herb', Gk. liv{}o., ' Rower' (M. 1.36, F.
1.108-1 09).
19. Skt. ard/,1 'sJX>ke of a wheel', Gk. Opo:pioKw 'join ' (~1.
l.48,F. I.J 28- 129).
20. Skt. drjuna~ 'shining', Gk. 0iJY6~ 'id.' (.~v1. 1.50-51,
F. 1.1 32-133 ).
21. Skt. drdat1 'flows', Gk. /i.pt3w 'water (cattle)' (M. l.Sl ,
F. !.135).
22. Skt. civa 'down, off', Lat. au- 'away' (M. 1.56, W-H
!.79).
23 Skt. iiyU/.t "living', Gk. aiWv 'li fe' (M. 1. 77, F. 1.49).
 lndo-Eu.rapran /a / 19
24. Skt. avi~ 'open ly', Gk. tYiw 'perceive' (M. 1.82, F.
1. 48~49).
25. Skt. i.)ati 'seeks', Arm . ai( 'enquCte' (~ - 1.130, 35).
26. Skt. ina~ ' crime', Gk. aiVvtuxt ' take' (M. Ll 28, F.
1.4 1).
27. Skt. idlvl& 'fuel', Gk. aH}w ' kind le' (!vi . I.l 28, F.
1.37~38) .
28. Skt. Oja.{l 'strength ', Gk. aV ~D: vw 'increase' (M. 1.1 3 1,
F. 1. 187 ~ 188).

Most of these comparisons arc sound and can be re-

tained, but the list must nonetheless be pared d own a bit.
dfru. cannot be included because the question of its init ial
vowel is not secu rely answered, a nd the precise relation of
dfru to Gk. 00Kpv, Lat. lacrima, etc. is uncertain . Though
Szemerenyi ( 1964: 8) counts this correspondence as sure,
Kurylowicz (1956 : 194) does no t incl ude it in his list of
forms pointing to IE */ a/ , and considers &i:Kpv a 'Euro-
pean word'. Much the same difficu lty arises with dsrk, the
initial vowel of which is rendered uncertain by Gk. ~'ap
(d.ap, ~ap) and Hitt. d&ar ( F. 1.432). tiyzi& and tivi~ must be
excluded from the list because they show a correspondence
/ a: /:::: /a/, not /a/:::: /a/ . Though it is likely that the
forms are in fact related morphologically, they cannot be
u~ed for comparisons on th e phonological level . The same
must be said of drdati : i.ipSw, for the vowel in the Gk. form
is long (Herodian 2.1 09, F. 1.135 ). Perhaps also, in view of
Frisk's cri ticism s ( 1938:8-1 2), it is best to give u p t he con-
ncctidn of bw~ and Gk. alVv,ua:t.
Num erous forms remain, however, enough to show that
an */ a/ must be reconstructed for Indo-European in ini -
tial syllables. As K urytowicz says ( 1956: 189), one cannot
cast doubt on t hese correspondences without contesting
the very principles of comparative lin,!.,>uistics. We can
20 Indo-European /a/
therefore establish the ru le that some IE */a/ appear in
in itial position as / a/ both in the European languages and
in Indo-Ira nian as well. Pedersen felt that all IE */a/ in
initial position remained in all languages, but he may have
been overenthusiastic. If we look again at the now some-
what shorter list of cognates, we fmd that almost all the
Skt. forms have the / a / accented. The suspicion therefore
arises tha t perhaps only accented IE */a/ - (or accented
1-1 */a/-) appears as /a/ in Indo-Iranian, and that the
reflex of unaccented */ a / is something else. The only
counterexamples to such a narrower formulation of the
rule arc: catur-aiuib; ajdf}., ajfnam; anib; arghdl;; amldb, five
examples out of twenty-two. And whereas accented • / a /-
occurs in etymologically unexceptionable morphemes,
some of th e correspondences displaying unaccented */a/-
are questionable.
catur-airdb, however, is probably related to liKpos in ap-
pearan ce on ly . Wackernagel ( 1905: 119), followed by
Mayrhofcr (M. 1.6 1}, felt tha t -airtib is not a form inherited
as such from Indo-European, but rather is the regular
replacement of dir0 in composition. Now dir0 is most
likely to be related to Gk.liKpt'> ' hill-top' (M. 1.61, F. 1.59-
60, Kurylowicz 1956: 188}, and hence rather supports
than contradicts the rule that only accented IE */a/ -
passes to Ind. /a/. But it is possible a lso that tiir0 is rather
to be compared with Gk. 0Kp£" 'jagged point', Lat. oms
'mons confragosus', in wh ich event it is irrelevant to the
discussion at hand (F. 1.59-60, 11 .374; W -H 11.1 99).
There are difficu lties also with the equation Skt. amldl}.
'sou r', Lat. amdrus ' bitter ', PGme. *ampra- < *ambro-'sour,
bitter', a n d Kurylowicz d oes not even mention it in his list
(1956 ; 188). In the first place the Skt. form is not attested
until fairl y late, being epic and classical only (Wackernagel
1954:862), and second ly it never OCt."Urs in an accented
text, so that we cannot be sure of the accentuation (llirwe
 Indo-European fa/ 21
1956: 198). am/db therefore lacks the secure credentials one
hopes for in Skt. words. And there are formal difficulties.
Frisk ( 1934:14) includes this word among fonnations in
-ro- which show goocl semantic agreement but 'slight for-
mal discrepancies'. He is, however, inclined to accept the
equation, though he is uncertain about Lat. amtirns, saying
that if it is related, it must have been refashioned on some
unknown moclel. The difficulty is that the earliest Ger-
manic fonn reconstructable is *ampra- which could de-
velop from *ambro- with a -b-. But the Skt. forms amltil,1 and
ambldJ,1 (Lex. ) presumably must derive from a form without
-b-, such as *am-ro-. And the Lat. form not only lacks a -b- ,
but has a long vowel between the -m- and the -r-. One
could indeed assume an original *amro- continued directly
by Sanskrit, but with epenthetic -b- in Germanic, but the
Lat. form still proves recalcitrant.
There is also difficulty with the initial vowel. In addi-
tion to Skt. amid!;. and Lat. amtirus, the following forms arc
usually cited as cognate:9 Alb. tam/l'(saure) Milch',ambde,
tmble ' si.iss', timble 'Galle' (with the / t/- being an article);
PGmc. *ampra-; Olr. om 'raw', W. of ' id.' ; Lett. amuofs
'Sauerklee' As can be seen from the list, even if we assume
that a ll these form s are indeed related, only two languages,
La tin and the Celtic languages, can give any evidence as to
the quality of the original vowel, and that evidence is con-
tradictory. Celtic points to original* /o/, Latin, to original
*/a/ . Pedersen (1909:32) felt that Olr. om contains the
original vocalism, and that Lat. amtirns has a secondary
/ a/ . The original IE root form was therefore* / om/ -. Po-
korny seems to agree (777 - 778), for he lists all the forms
under *om-, and not *am-. None of this proves the assump-
tion of an IE *am-ro wrong, but makes rather more likely
a fom1 such as *om(a)-ro as starting-point for the develop-
ments in the daughter languages.
Skt. ajd!J. 'he-goat', 0:ifl. 'she-goat' also contravene my
 Indo -European /a /
m le, t hough they are in accord with Pedersen's, il; that is,
they are cognate with Olr. ag. And , though ag is generally
connected wit h Skt. ahi 'cow', Ave. a.;:f'p rcgnant', and not
with a;df:i (:vi. 1.68 ), it seems most likely that Kury!owicz
is right in connecting it rather with ajd/:1. In addition to ag,
the family o f ojd/:1 includes: Lith . oZjs 'he-goat ', of.kil 'she-
<
goat'; Alb. dhi( *a.r;i ) 'she-goa t'; a n d more impressively:
Skt. a;fnam 'skin'; Lith. oZ ~1115 'zum Ziegenbock gehOrig':
OCS () )azno<*a.;:tmo ' skin'. But the Slavic and Lith. forms
must be removed from consideration on the phonological
levd, for all form s in these languages point to an original
*/a:g/ -, not */ ag/- : on ly Alb. dhi, Olr. ag and Skt. ajh,
O:Jdf:i point to* / ag/ -. And the suffix differs in each of these
languages : O lr. -rs; Skt. -ii, -os; Alb. -i. H ence thcycannot
all derive from one a nd the same ancestral word. The best
that can be said for t hem is that they a ll derive from a sin -
gle root- noun *ag-, homophonous with the verbal root *ag-
'drive, lead' , o rig ina lly id en tical with it, and of course siryce
monosyllabic, originatly accented. Thus all the forms in
the daughter languages are derivatives of* /ag/ - fOrmed
independently, all mean ing ' the driven one', a meaning
applied to goats in Albanian and Sanskrit, but to cows in
Old Irish. Th ere is, so far as I can st.'"C, no reason to assume
with Thiem e (1953:580-581) a n IE word *ag 'goat', and
there is of course sti ll less reason to posit an IE word *ag6s
'goat' ancestra l to Skt. ajd/:1.
The other two cases, arghd/:1, and ard/:1, though clearly
counter-examples, and though a lmost certainly cognate
with European forms with initial / a / , are not sufficient to
destroy the ru k . Though the forms are cited in these
shapes, accent ed derivatives of the same roots d o occur in
Sanskrit, and \VC may presume, faute d e micux, that it was
these accen ted form s (or other accented derivatives now
lost ) that preserved the /a/, unless indeed ·we assume an
 lndo-Europwn /a/ 23
accent shift wit h in lnd ic. In the case of arghd/:1 there occur
tiryati 'acquires' and tirhati 'is worth', and for ard/:1: dram
'readily, enough'; the regular unaccented form of this root
is seen in rtdb 'right, true'. l-Ienee we may reformulate
Pedersen's rule to read: IE ;a; remains initially in all IE
languages.
The rule just given seems unexceptionable as stated , but
one wonders what happens to unaccented /a/, and also
how one is to account for cases for which scholars have in
the past been tempted to reconstruct hi in initial position.
There are three classes of examp les which a re relevant
here, each w ith its own problems, and each of which must
be dealt with, however cursorily, before my rule can be
considered ready for serious consideration. I) A Skt. form
like ipsatl 'wishes to obtain' (instead of*yapsati) from apn6tz
favors an analysis •iJpseti (or *iHpsat1) over *wpsttz: * /ya/
should appear as /ya/ in Sanskrit, as in yti)ate, rather than
a~ /i: / . 2) The Gk. prothetic vowel, seen in linjp ' man' as
opposed to Skt. nr 'id .' again might demand an analysis
•Hner or *mer with the rule that */H/ or */:J/ passed to
/ a/ in Greek but to 0 in Sanskrit (and elsewhere ): with-
out this rule one might expect Skt. *anr instead of nr.
3) Even if this / a / = 0 correspondence be attributed to a
literally prothetic vowel in Greek, we nevertheless find cor-
respondences, such as Lat. aurora, Lith. aufr/i, Skt. u.~d~t
'dawn' in which all European languagc:s show / a /, while
Indo-Iranian alone shows no vowel at all. Again one is in-
dined to reconstruct something other than •aus- as the IE
foim, possibly *JUs - , on the assumption that, though IE
*/a/>/a/ cveryvvhcre, IE */;;,/>Eur. /a/ but Skt. 0.
We shall take up these difficulties one by one.
1) The first of these difficulties was raised by Glintert
(19 16 :5 - 6) as an objection to Pedersen's view that IE ini-
tial */a/ remains in Sanskrit. H e held that the following
24 lndo-Europran /a/
countercases are sufficient to invalidate Pedersen's rule:
fpsati 'wishes to obtain', desiderative to yap in iipn6tl
'obtain' <*i + .JP + sr- (M. 1.95).
i4att 'sees', a reduplicated present to the af0- seen in
ak~i 'eye' < *i + :~k + sr- (M . 1.95).
dvipd!J ' isl and', a compound of dvi- + the weak form of
apai:J 'water' <*dwi + "P- (:vi. Il.86-87).
fhatr 'desi res', connected with Ave. iiz.,if 'Begicrde', and
perhaps further with Gk. fixr'jv 'needy', hence from
*i + ,gh- (M . 1.97, F. 1.200 ).
Gi.intert reasons ( 1916:6): 'diese Faile sind gewichtig
genug urn die Unhaltbarkeit der Behauptung, i = europ.
a begegne niemals im Anfang arischer Worte, zu erweisen .'
But in fact these cases are not relevant to Pedersen 's rule
at all. They show that IE */a/ could appear in Sanskrit
as som ething other than /a/ in morpheme initial position,
but have no bearing whatsoever on the question of the fate
of IE"' / a/ in word-initial position. They do, however, in-
dicate that IE • /a/ (and "'!of in ik..yatr) 10 when unaccented
(cf.y4f-: 11.43 below), and when not prott-"Cted by accented
forms elsewhere in the paradigm, appeared in Sanskrit as
length after / i/ (and /u/ as well: cf. aniipd!J. 'situated near
the water'). This distributional statement will later prove
to hold in all positions in the word. Hence unaccented IE
/a/ disappears after a sem ivowel with compensatory
lengthening of that semivowel: unaccented /a/>1:/

/[~]_ c.
2) The prothetic vowel in Greek is a large and comp li-
cated problem, too complicated in fact to be used as evi-
dence of a contrast between /a/ and / d/ (/H / ) in initial
position. Nonetheless it has been so used, and in particular
in cases like Gk. O:lfl]p, Skt. nr 'man '; Gk. Ovof..UX, Skt. ndma
 lndo- Eumpean /a/ 25
{Arm. anun) 'name' . From these words scholars have con-
cluded that IE "/d R/ (or /HR/) contrasting with / aR/
(or / H 2eR /) developed to VR- in Greek (and Armenian )
but to HR- elsewhere .
Prothesis is, however, too regular, and at that too regu-
lar in terms of purely Gk. phonological conditioning, to
allow for such a conclusion . In the first place, as is well
known, */r/- always receives a prothetic vowel in Greek,
and relevant instances can therefore be found only before
/ 1m n w/ (there are no cases before /y/) . Furthermore,
even before / 1m n w/ prothesis occurs on ly, though not
always, in two major categories of words: I) before a reso-
nant followed by /e/ in a closed syllable; and before / a /
in either an open or a closed syllable, a position in which
there is variation between forms with prothesis and fonns
without: 2) as the result of the analogical extension of a
syllabic resonant to positions in the paradigm where the
resonant was consonantal. These two environments may
be exemplified in the first instance by a mere listing, in
the second b y a concrete example (examples from Lejeu ne
1955: 127- 129, 148):

a) t:ll\dcpw but l\o{;w

l.l\£iJ{J£po'> AO~o ..
&l\d'T1js l\dww (l\il\omo:, ~l\mov)
&pi.l\yw but J.l.f)TYjp
Op.dxw ,Ul\t
Ovu8os but vlo<;
tlv£1ft0<> vitpo<;
iAo:xtJs beside Ao:x£ia
&,uaAOs po:l\o:K0'>
UA&mt beside k'A8£Tm
UptYIJ ~ptJYj
26 lndo- Eurojmm I a/
The quality of the vowel varied, apparently in terrns lxHh
of the follow ing consonant and of the succeed ing vowel or
diphthong.
b ) The prothetic vowel in t:il"ljp is the result of a level ing
of the irregular parad igm * / ne: r / , */arbs/ by extension
of the / a / (</n/) of the oblique cases to t he nomi native
and accusalive cases: thereu pon the stem o f the word was
taken to be / an /-, and th is / an / - was introduced to the
oblique cases. T he original IE paradigm of this word in the
singular was (restricting myself to cases preserved in
Greek):
/ nC:r/ / nrOs/ /n ri/ / nCrm /
From this there developed in Greek, after the vocalization
of the sonant resonants, the pa radigm:
/ ne:r/ / ar6s/ / arl/ / nera/
TI1is irregular paradigm was leveled by introducing / a /
to the nominative and accusat ive:
/anf:r/ / ari'is/ / arl; /anCra/

whereufX>n /an/- formed the base of the paradi).{l11 and

was exte nded to the ob liq ue cases, and the classica l para-
d igm resulted: II
/ane : r/ / andr6s/ / andri; / anera/

Th ere never was a vocalic element preceding the initial

*I n/ of IE */ ner/-, and the Gk . prothelic vowel has no
bearing on the question of I E in itial */a/.
2) Cases in which Eur. / a/ corresponds to I -I 0 initially
cause no difficulty for those who believe in an I E ablaut
/a/- 0, whether they express the alternat ion in this way
or as /H 2 e/ - - /I-121- >0. Nor does this alterna tion
cause difficulty for my posi tion because the rule expressed
 Indo -European / a/ 27
above holds tha t in itial / a/ remains in Indo- Iranian on ly
when accented : we need merely suppose that unaccented
/ a /> (l. :'\oncthelcss these equat ions, a ll absolutely ce r-
tain , do cause trouble for Pedersen's position and are im-
{Xlrtant enough to warrant separate treatmen t.
'dawn ' Skt. urdh, Ave. uJJ, Gk . ~W~ < "ausOs, Lat.
aurora, Lith. auSr/1 (M . I.l 13, F. 1.605- 606,
W-H 1.86).
'grow' Skt. Ukrat1, Ave. ul].fyrit1, Gk. ai.l~w Lat. augeo,
Lith . dugt1 ( M.l.98, F. 1.187- 188, W-1-1 1.82-
83).
'ear' Ave. uJi 'both cars', Lat. auris (P. 785, F.
11.448-449, W-H 1.85 - 86) .
' voice' Skt. uditri~1, Gk . o:Vlhj (M .l.l 04, F. 1.1 84).
'and /or' Skt. utd, Ave. ultl, Lat. aut (M. 1.101 , 'vV-H
1.87).

There are other examples, bu t these suffice to show the re-

lation Eur. / a/ = 1-1 0, at least before / u/, a nd they also
provide the condi tioni ng for t his relation.
There a re a lso inner-Skt. examples which buttress the
conclusion that unaccented /a/> 0 before / u/:
ijal} 'strength':ugrd~ 'powerful' (compa rative 4fiya.5-) to
ulq 'grows' and cognate with Lat. augustus 'venerable' (.:\1.
1.1 3 1, 98-99: W-H 1.82-33).
dvati 'is p leased' :un6ti 'encourages', cognate with Lat.
aveo 'desire' (M . 1.57, 104; W-H 1.81 ).
6tum infinitive : ut~ past pa rticiple of vd)"at1 'weaves' cog-
nate with Lith. at-audal 'Einschlag' dtLJti 'weave' (M .l.J 32,
111.1 47).
All show th at IE "'/a u/- rema ins in Indo- Iranian when
accented, bu t lose.<; the " / a /- when unaccented. And a
28 Indo-European /a/
check of Mayrhofer's pages (M. l.130- 133) shows that in
fact there are no cases of unaccented initial /o/ equatable
directly with IE forms containing */au/. We may there-
fore account for correspondences like aurora: 14b4 most eco-
nomically by assuming a peculiarly I-I sound law in ac-
cordance with which IE • / a/- > 1-I 0 in initial position
before / u /.
The same accentual relation exists within Sanskrit a lso
\Vith the vowels /e/ and /i/, save that certain deictic words
(etdt 'this', evd 'so, exactly', e.~d 'this') have unaccented /c/.
But these must be secondary forms, a strengthened form
of /i/ in the case of evd; tat, and sa prefixed by /e/ ( <"'ei)
in the case of eMt and t.[d (M. l.l27, 128). All other cases
of initial /e/ in Sanskrit which are equatable with IE
forms have the accent (M. 1.126-130). The following
equations are relevant here, though i'IUJ/] may not belong
(see above 1.26):
idhal] 'fuel', Gk. o:l{)os 'fire': inddhi'sets on fire' (M. 1.1 28,
88, F. 1.37- 38).
lna!J 'crime', Gk. ai'PVJLat 'take':in6ti 'advances upon'
(M . 1.1 28, 87; F. 1.41).
i~atz 'seeks', Arm. aic; 'enquete':icchdti 'seeks for' (M.
l.J 30, 85).
And the same relation Eur. */ai/-::::; Skt. /i/-, but with-
out an accented Skt. /e/-, is seen in:
Gk. al()oiWL 'be ashamed': Skt. iur 'venerates' <*idtai
(M. 1.95, F. 1.34-35 ).
We may therefore extend the rule: IE */a/-> 0 / - / u /
in Indo-Iranian t o include /i/ as well. And since we have
found no cases of unaccented ini tial /a/ in Indo- Iranian ,
it is legi timate to present the rule governing IE */ a/- as:
/3./- >;a;-, b ut /a/> 0.12
 lruW-F.u.ropwn /a/ 29
II. I n t he inte rior of the word things are more difficult,
for to Eur. /a/ (or /o/ < */a/) corres(Xlnds both /a/ (Skt.
harh.~dl}, Lat. anser 'goose') and / i / (Skt. sthitti&, Lat. sla/u.s,
past participle of stii- 'stand'). Furthermore, though the
/a/ ::::: /i/ equation is well established by many secure
etymologies, the / a/ ::::: / a/ equation is frequently most
problematic . Scholars will d iffer as to what co~:,rnates they
will admit as evidence for JE /a/ (Eur. /a/= 1-1 /a/),
but the fo llowin g list at least will likely be accepted by all ,
regardless of how they chcx.Jse to symbolize the recon-
structed segment:

I. Skt. harhsti& 'goose'. Gk. xilv 'id.', Lat. ansa ' id.' <
*ghans (U. 356, B. 1058, W-H 1.52, Szemcrenyi
1964:8).
2. Skt. nas- 'nose', OCS nosr, OHG nasa, Lat. nlisus,
nrlres< "'nas- ( M . 11.1 46, VV-H 11 .143- 144, Peder-
sen 1900:82, SzcmerCnyi 1964:8).
3. Skt. devtir- 'brother of husband', Gk. l5aijp, Lat . levir
< *daiwir- (M . IL 6S , F. 1. 338-339, W -H 1. 787-
788, Kurylowicz 1956:191, Szemerenyi 1964:8) .
4. Skt. bhdjati 'deal out', Gk. (q,o.yov ' cat' <
*bhag- (M.
11.463- 464, B. 1010, Wackernagc\ 1896: 78, Peder-
sen 1900:82, Ku rytowicz 1956: 1:J I ).
5. Skt. grtisalt 'swallow, devour', Gk. ypb.w 'gnaw, eat'
< *gras- (M. 1.352, F. 1.326, Wackernagel1896 :78,
Pedersen 1900: 82 ).
6. Skt. hi,m& 'arrow ', Gk. xo.to~ 'shepherd's staff '
<*ghms- (U. 362, B. 1046, Kurylowicz 1956: 192).
7. Skt . tvak 'skin', Gk. a6:Ko~ ' shield' <*twak (M.
L537 , B. 849, Kurytowicz 1956: 190 ).
30 lndo -t:uropemz j a/
All these examples contain /a / in the in itial syllable of the
word, and that / a/ is eithe r accen ted (4- 7); or appears in a
monosyllable ( 1- 2; *ghans must have been the IE word for
'goose', and the Skt. and Lat. forms la ter extensions of it);
or in a syllable closed w ith / y/ (3), from which we may
generalize to: a syllable closed by a resonant.
Given the above environmenta l specifications we can
then admit the following as evidence for IE */ a/:
8. Skt. darilsa~ 'marvelous power', Gk. c'h'ww 'counsels,
plans, arts' < *dansos (M . II .9, F. 1. 382, Pedersen
1900:78 ).
9. Skt. ddfa11 'bites', Gk. Oo:Kflv 'id.' <
*dak- (M . ll .27,
F. 1.343- 344). But it may be that the / a/ in both
the Gk. and the Skt. fmms derives from* / n/.
10. Skt . dtiyate 'divides, imparts', Gk. Oaio}J.O'.l 'divide,
distribute ' < *day- (M . 11.20- 21, F. 1.34 1-342). Cf.
also Skt. PPP. dmd~, ditd}_1 and Gk . 8a.Ti.oJW.t 'divide
among th emselves'.
11. Skt. dhdyat1 'sucks, drinks', Goth. daddjan 'suck',
OCS dojr 'suck' < *dhay- (M . 11. 93, Hi.ibschmann
1885 : 79). But the vowel in these cases may have
been • /of as expected in causative formations (Bur-
row 1949:43, Kurytowicz 1956:166).
12. iats)"ati 'fall (fut.)', Lat. cado 'fal l' < *kad- (P. 516,
E -M 82, Kurylowicz 1956: 19 1).
13. Skt. kadanam 'd estruction', Gk. Ko\.a8Wv 'deprive of'
< 'kwl- (M . l.149, F. 1.8 11 ).
14. Skt . fiifdda 'excel', Gk. KiKa.8)J.O.t 'excel'< *kad- (P .
516- 517, F. 1.811 - 812, Pedersen 1900:82). Kurylo-
wicz ( 1956: 192) holds that the attested forms of fad-
do not exclude a root with long vowel (fad-).
 lndo-Europeml /a/ 31
15. Skt. kekara/J 'squint-eyed ', Lat. caecus ' blind', Goth.
haihs 'one-eyed' < *kaiko- (W-H 1.1 29, Kurylowicz
1956:190). Mayrhofcr (1.264) feels that this rela-
tion is in no way certa in because of t he late attesta-
tion of kekara~1, the by-forms not exam pled in litera -
ture (kedarafJ, terakal;), and because European
cognates come only from the western European
area. Szcme rCnyi ( 1964:8) regards this as one of
the surest exam ples of IE */ a / .
16. Skt. kisaram 'hair', Lat. caesaries 'the hair'< *kais-
(M. 1. 268, W -H 1.1 33, th ough there are phonolog-
ical difficulties signaled by Kuryfowicz 1956: 193).
17. Skt. kiva{a 'cave, h ollow', Gk. Ko.iam·OpVypn.m Hes.
< *kaiw- (F. 1.753, Kurylowicz 1956: 190 ; .\1ayr-
hofer 1. 267 feels that the connection of these words
is unlikely).
18. Skt. k evaia~1 'exclusively one's own', Lat. cadehs 'un -
married' <*katl- (.\1 . 1. 267, W-H 1.130).
19. Skt. kalyal; ' hea lthy', Gk. Ko.Miwv ' fa irer' <*ko.ly-
(M . 1.1 84- 185 , F. 1. 767, Pedersen 1900:77).
20. Skt. iamnite ' toils', Gk. K6.J.ww ' work' <*kam- (F.
l.773 - 774, Cardona 1960:502- 507). It is perhaps
more usual to derive these forms from * / km/-.
21. Skt. kaninaiJ 'young', kdniyiin 'younger', Gk. Ko:tv6~
' new' <*karry- ( !\1. 1.151 , F. 1.754).
22 . Skt. iankU- ' peg', Welsh came 'b ranch ', O f\ hdr
' Ruderk.lamp', OCS sok 'branch' <*kanku- (P.
523, Kurylowicz 1956:19 1, Szemeri:nyi 1964:8).
23. Skt. karka(a/; 'crab', G k. KapKivm 'crab', Lat. cancer
' id. ' (<*karkaros by d issimilation ) <*kar- ( .\1.
32 lnda-European / a/
1.169, F. 1. 789 -790, W -H 1.1 5 1, Pedersen 1900:77,
Kurylowicz 1956:190, 193).
24. Skt. karko.raf:t ' hard, finn ' , Gk. KixpKapos·TpaxVs Hes.
<*ka rkar- (F . 1.789; M . 1.170 is dubious) . Kury-
fowicz ( 1956 : 192) says the form rests on intensive
reduplication , and Burrow ( 1945 :9 7-98) feels that
th e Skt. word is connected \vith kharai}. 'hard, rough,
sharp', and that both are of Dravidian origin.
25. Skt. mddatz 'rejoice, be intoxicated ', Gk. JUX&'xw ' be
moist ', La t. madeo 'be wet, be drunk' < *mad- (F.
II.I 57 - !58, Wackernagcl 1896:78, Pedersen 1900:
83; E-M 377 exclude comparison with the Skt.
form, and Kurylowicz ( 1956: 189) rienies that the
Skt. form derives from *mad- because ofGk. !HOTO<>
which points to an */e/ vowel).
26. Skt. paillca~ ' mud, mire', Gaul. ana ' palus', Mlr. an
'eau', Goth. Ja ru 'Schlamm ' <*pan- (M. IL1 84,
Kurylowicz 1956: 191 ).
27. Skt . rdbhalt 'ergrcift ', G k. Mrpvpo: ' booty' <*labh-
(M. lll.42-4 3, F. ll.9 I , Pedersen 1900: 83) .
28. Skt. rribha~ 'violence, impetuosity', Lat. rabUs 'rage,
madness' <*rabh- (M. 111.43, VV-H 11.41 3, E-M
562, Pedersen 1900: 82).
29. Skt. skdndat1 'springs', Lat. scando 'climb', Gk.
oK<iv&x,\ov 'trap' < *skand- (B. 870, E -M 599, W-H
11.488, \Vackernage l 1896 : 78, Pedersen 1900: 78).
30. Skt . svddat1 'taste well to', Gk. &v&ivw ' please'
<*swad- (F. 1.104, Pedersen 1900 :77 ; Hi.ibsch-
mann 1885 :59- 60 fee ls that *swad- <
*swnd-, but
this seems unlikely).
31. Skt . lavili '€:tre fort, a voir Ia puissance, pouvoir', Gk.
 lndo~Eumpean /a/ 33
nri"wpiyo:s, Ta6am•p;(yaMvas , wAwvO:oa~ Hcs.
<*taw- (B. 945- 946, Kury!owicz 1956; 190; M.
1.490, though he accepts the equation, finds diffi-
culties with it).
32. Skt. vtiiicati 'totter, stagger', Lat. vacillart 'sway to
and fro' <"wok- (W-H 1.268, M. IlL 127 hesitantly;
E-M 710 regard vacillare as an expressive word of
obscure origin; Kurylowicz 1956;189 feels that
these forms cannot be related be<:ause of the com-
plete absence of the nasal in the Lat. word).
<
33. Skt . vastu 'place, thing', Gk. (F)6:aTv 'town' •wastu
U. 279, F. 1.173- 174, Kurylowicz 1956: 191 ).
34. Skt. yd;fai 'worships, sacrifices', Gk. O:kop.m 'stand
in awe of' <*yag- (M. IIL 3-4, Pedersen 1900:77;
F. 1.10 feels that the Gk. word is 'nicht sichererklirt'
and hence not certainly to be compared with Skt.
)'djati; E-M 5!:!7 favor connection of Gk. O:yw~ with
Lat. sactr).
All of the above are positive instances of a rule:

/ a/ > /a/ It '~c]

but there arc several counterexamples not accounted for
by the rule, words which contain unaccented /a/ in an
open syllable : Skt. faT)ti~t 'a kind ofhemp', vaffl 'cow, barren
cow', fafd/:t 'hare'.
Skt. faTJd/:1, Gk. K0.vvaf3ts (Lat. cannabiS) OE h(J!ntp, OCS
k.onoplj"a are clearly related (U. 301 - 302, F. 1.779, W-H
1.154, Kurylowicz 1956: 19 1), but are just as clearly all
loans from some non- IE source. This conclusion is indi -
cated both by the nature of the thing designated, as well
as by ( I ) t h e Slavic /k/ which excludes direct connection
34 Indo -European /a/
with Skt. Satpi~l, and by (2) the Skt. cerebral /Q/. Further-
more the word is not restricted to Indo-European , for the
Sumerian kunzbu must also be related. Hence this word is
not a true counterexample because it was not in the lan-
guage at the time that unaccented /a/ was passing to [o ].
Skt. va.lii, Lat. vacca (U. 278, W-H 11.722, Pedersen
1900:77, Kurylowicz 1956: 191) is more difficult, for the
lack of exact identity in the velars is not sufficient to cause
one to discard the equation. But there are various reasons
for doubting that the \VOrds are related. In the first place
the word occurs in only two IE languages, languages
which share vocabulary generally only in the religious and
legal spheres ( Palm er 1961 :25), and not in the agricul-
tural. Second, the meanings diverge: the Skt. word seems
to mean primarily 'barren', and can be used both of cows
and women, while the Lat. word d<X:s seem primarily to
mean 'cow'. Third, if the words arc cognate, and if the
original meaning was 'cow', then the IE word denoted
only the female of a domestic animal, a distinction accord-
ing to sex not made with most other domestic animals
(E-M 710). And finally there are other words within
Sanskrit with which vaSJ might better be compared: either
viiiatf 'heu lt , blOkt', or ulqJ 'bull' (U . 278), and certainly
Wiit/J 'rindcmde Kuh' (Wackernagel 1896: 226); and in
Latin we might connect vacca with vaccinium 'blueberry' on
the assumption that they are related Mediterranean
words, though vaccinium is usually connected rather with
Gk. iHXKtvlJo~ 'hyacinth' (W- H 11.722). None ofthcsc con-
siderations taken alone or together would be sufficient to
destroy the connection, but thev are sufficient when the
word ~!so goes against a ru le (if unaccented */ a/ > 1-1
*/ i/ be accepted as a rule) of IE phonology. 13
Skt. faJd~ 'hare', Lat. ainus (<*casnos) 'gray, white' ,
Osc. ca.mar 'old', \Velsh uinach ( < *kasnl) 'hare', OHG haso,
 Indo-European /a/ 35
OPruss. sasins (V . 306, F. 1.812, W-H 1.156, Kurylowicz
1956: 19 1). Though this equation has been considered
t:ertain by many (SzemerCnyi 1964 : 8), and indeed ap-
pears solid , it is in reality not so. In the first place the Skt.
form requires that we assume an assimilation of i-s to
1-i, an assimila tion nowhere else attested (Mayrhofcr
1952: 29 ), and further that we posit a thematic derivation
of *kas-, a derivative not to be found in any other lan-
guage. T his equation also leaves completely unexplained
the Gk. form K£Kijva<;·,\aywoU<;.Kp1jn<; (l-Ies.), a form which
one wou ld o rd inarily compare with the Sanskrit , deriving
both from *kek- (or possibly kekes-). .\tlayrhofer ( 1952) in
fact does just th is, and in turn connects these words with
faiati 'sprin~:,>s' (< *k eketl). H e further assumes that *kas-
'white, gray' is a European word only, an assumption
shared by Ernout-Meillet (94) . An IE **kas6sshou!d have
developed to Skt. * fud~1, not iaidh. 14
If * / a / occurs on ly in closed syllables or accented, schwa
must occur on ly elsewhere: the following cognates show
under what conditions [d] appears:l'•
35. Skt. damitdl; 'tamed', Gk. 0:00/Ul'TO'i" 'unconq uered',
Lat. domllor 'tamer' < *domat- (M. Il.1 9, F. 1.346,
W-H 1.367 - 368).
36. Skt. duhi!A, Gk. tfvyirTI}p, Goth. daUhtar 'daughter'
<*dhugatir (M. II .56, F. 1.690) .
37. Skt. iamzldr- 'preparcr, dresser' , Gk. Kcipa-ro<; 'toil,
trouble' < *kamat- (U". 303, 308; F. 1. 773 ).
38. The -n i- of Skt. ninth-cia~ verbs (e.g. mnumdh
'crush' beside mr~m1) and the -va:- ofG k. verbs like
&ipvctf.LlV ' subdue' beside OOpvliJlt (de Saussurc
1922o224 ).
39. Skt. pitdr-, Gk. '7iaTftp, Lat. pater 'father' < *fJalir.
36 lndo- Europta n / a!
40. Skt. sthitd}_1 'standing', Gk. amn}s 'placed', Lat.
status 'set, fixed' <*stat6s ( U. 34 7, F. 1.739, W-H
11.596-599).
4 1. Skt. ii{!ti& 'taught', Lat. autus 'morally pure'
<*kastOs (U. 308-309, E-M 104; W-H 1.179-180
deny any connection between these words).

Thus schwa occurs only unaccented , whether in initial or

noninitial syllables, and in initial syllables only before a
single consonant /t/ , or before / st/, a cluster which may
well not have closed a preceding syllable in Indo-Euro-
pean. There is therefore no contrast benvcen [a] and [;}) in
any environment, and h ence no reason to distinguish t hem
phonemically. We need only the rule: IE unaccented* /a/
passes to[;>] ( I-I / i/} in syllables not closed by a resonant .
Further cognates, some of them questionable, support
the rule just given:
42. Skt. l.p.rd& 'refreshing, fresh; flourishing, vigorous',
Gk.lo:p6~, Up6~ ' holy' <*uar6s (M. 1.93, F. 1.712-
714).
43. Skt. madirf0 'intoxicating', Gk. }.ta0ap6s 'wet'
<*madar6s( M. 11.569, F.ll.l 57- 158).
44. Skt. ptirlman- 'plentifully', Gk. 'ir"D..avo~ 'thick liquid
substance' < *pita- (M. 11.219, F. 11.494).
45. Skt. prathimf1 'extension, width ', Gk. 'ir"Aa.Ta}J-Wv
'broad flat body' < *plata- (M. 11.364, F. II.554).
46. Skt. tdmisra ' a dark night', Lat. lmrhrae 'darkness'
<*timas- (M. 1.479, W-H 11.664).
47. Skt. fad& 'sharpened' (PPP of iii -), Lat. calus 'sharp,
shrill' < *kat6:. (U. 3 11, W-H 1.1 83- 184).
48. Skt. ftirito& 'crushed' (PPP ofir-), Gk. Kfpai!w ' rav-
age, plunder' < *kira- (U . 315, F.1. 822).
 Indfi-European Ia / 37

49. Skt. chitd~ 'cut up' (PPP of chii- ), Gk. ox&w 'slit open'
<*ska- (M . 1.410, B. 931 -932, W-H II. 495-496;
Burrow 1949:47 analyzes the root as *skeH- ,
*skHi-).
50. Skt. vdmiti 'vom its', Gk. ipiw (if from *i!Jiiw) 'id.',
Lat. vomitus 'id.' <*(w)imn- (M. III.l46, F. 1.505,
W-H 11.835).
Just as there were apparent instances of unaccented / a /
in an open syllable, there are also cases of accented Skt. / i/
which would seem at first sight to cast doubt on the distri-
butional statement just made, and which must be disposed
of before we can accept my ru le. The first group of excep-
tions, cited by Glintert (1916 :13-14) and Bru~:,>mann
(IB97: 173), is composed of cases of accented / i/ deriving
from schwa in the in itial syllable, a position in which we
would expect /a/ . Such cases arc: sthftz~ 'standing' from
sthii- 'stand'; d[tll} 'generosily, d istributing' from dri- 'give';
sidhyatz 'succeed' beside siidh- ' reach one 's goal'. Un fortu-
nately Glintert wrote at a time when the philological in-
vestigation of Sanskrit had not yet reached a very ad -
vanced level, and hence relied on form s not sufficient to
support hi s arguments. To take the forms in reverse order:
Renou (1964: 164- 165) has recen tly called into question
the connection of th e late Vedic hapax sidlryatz with the
root siidh-. And even if the roots arc related , they can not
be used as evidence for accented schwa, for the words are
Indo-Iranian only, not Indo-European. And one of there-
quirements for establishing IE schwa for a morpheme is
that it both contain Sk t. / i/ and show / a / in a European
lailguage. Mayrhofer (M. 439-440) doubts that t here ever
existed in Sanskrit a form ditil} beside the rq~,'ular dfzt1 and
- tti; Glintert thus was using a vox n ihili. sthftiiJ, though a
0

real form, is not exampled until the Satapatha Brahma!}a, a

period in which the accent of many such forms was being
38 Indo -European Ia /
dra\\-Tt back from the ending to the stem (Wackernagd
!954:622ff., esp. 63 1), thus a llowing for the possibility at
least that in early Vedic times sthitiiJ was in fact oxytone.
But in any event abstracts in -t1-, regard less of the attested
place of the accent, were formed with the reduced grade
of the root (Wackernagcl1954 :629) and the reduced grade
appears only unaccented. Too, analogy with sthzta!;, in-
voked by Pedersen (1900: 84) is not impossible. These ex-
amples do not invalidate the ru le.
The other part of Giintert's argument ( 1916: 13- 14)
concerns nouns in -tra- (Wackernagel 19S4:70 1ff. ) like
(5 1) arilra- 'oar' which is usually compared with Gk. (p(fi'/~
'oarsman', but is in fac t more likely to be related to Gk.
V...arijp 'driver' (Bechtel 1892: 205 ), both deriving from
something that could be written in traditional orthogra-
phy as *errt- (or *c&-t-).lnstancesofthis sort do constit ute
an exception to Pedersen's ru le , at least at first blush, but
do not as yet constitute an exception to my formulation,
since I have assigned a role to the accent only in initial
syllables: it is possible that, accented or not, all */ a/ in
interior syllables passed to(;}] and thence to / i/ in Indo-
Iranian . Such an assumption would be phonologically ac-
ceptable, bu t the cases in question do not favor this conclu-
sion. Simi lar fo rmations in Greek, such as TiptTpov ' borer,
gimlet' and /i.poTpov ' plow' are accented on the initial syl -
lable, not on the schv.'a, and many scholars in order to posit
a single protoform have assumed a secondary accent shift
to the second sy llable well within the history of Sanskrit
(Hi rt 1900:7, Kurylowicz 1958:68). And Pedersen { 1900:
84) , without positing an accent shift, invokes the influence
of the agent noun an'td r- 'rower'. Thus it St."t':ms that these
cases neither constitute an exception to Pedersen's rule nor
favor a rule: */ a/ (regardless of accen t) > [ :~] > / if in
interior syllab les in Indo-Iranian.
 Indo-European /a/ 39
Much the same can be said a lso of the pair:
52. Skt . kravf&, 'raw flesh ', Gk. Kpias ' meat' ( <*krewJs
F. IL12, M . 1.277, W-H 1.294- 295) .

The words differ in the place of the accent, and it is again

perhaps indicated to assume that the Skt. accent has
shifted from the stem to the ending after rocil} ' light', Socii;
'flame', etc., and that Greek has preserved the original
place of accent. This seems a better solution than assuming
the opposite, and there is therefore no need to deny con-
nection between these words as is done by Pedersen ( 1900:
77 - kravib < *km vis) and Burrow (1949:54- kravf;l <
"'krtuHis, Kpias < *krtuHos). Both the Gk. and the Skt.
forms derive from "'krlwas.
Neither of the above comparisons speaks in favor of as-
suming that a ll IE */ a / , accented or not, pass to / i/ in
noninitial syllables. What is more, two forms urge that we
retain Pedersen's formu lation and assign a role to the ac-
cent in internal syllables as well as initially:

53. Skt .jard& (V) 'old age', Gk. y(po:<> ' prize', yi)pm 'old
age' (M. 1.421, F. 1.299).
54. •-kds in Skt. parvaftib 'glicdweise', Gk. Q:pSpo:Kth
'man by man' {Wackcrnagel 1888: 144, Schwyzer
1939:630, F. 1.473).

In the first case there is no doubt that the forms are related,
but there is doubt as to the original shape of the word. Both
Frisk and Mayrhofer seem to assume an original "'gerUs
(< "'gmos ??) as the ancestor of the Skt. form, an assump-
tion for which I can see no need or justification, though of
course there is no argument sufficient to refute it. Rather
it seems that the Gk. form has once again preserved the
original accent (though we cou ld perhaps just as easily as-
40 Indo-European /a/
sum e the opposite) and that the ancestor fonn was IE
*giras. But in th is case apparently the shift of the accent
in Indo- Iranian from first syllable to stem vowel took place
early, before the phonetic change of unaccented /a/ to [;;,].
The regular development of unaccented */a/ is seen in
;arimfl 'decrepitude, old age' (M. 1.422) .
The second case seems to allow for no doubt whatsoever.
Schwyzer ( 1939: 630) compares Gk . iK&s ' afar off' and
&vSpaK&s ' man by man' with Skt. dvildh 'zu zweien, paar-
weise' andgaf}a.fd~l 'in Scharen'; Frisk (1.473) compares the
G k. words with Skt. iatairil} 'hundcrt fi.ir hundert, zu
Hunderten'; and Wac kernagel ( 1888: 144 ) followed by
Pedersen ( 1900: 82) compared parvafd& 'gliedweise' d .i.
'mit Sonderung der Glieder'. There seems to be general
agreement that the forms a rc related, and since they are
accented on the same syllable, we must reconstruct an IE
suffix *-luis, and must assume furthe r that it was the accent
which prevented the change of I-I */ a / to J;)] w / i/ . Hence
the rule for internal syllables wi ll be the same as that for
initial syllables: unaccented IE */a/ passes to I-I /i! in
open syllables.
T he above lists and discussions do not procluce the 250
cognates Kurylowicz ( 1956: 189) feels reasonably to be
expected from the number of initial correspondences. But
there are a number of reasons which can be al!egt.xl to ac-
count for this discrepancy, a d iscrepancy which in any
event is created by Kurylowicz's expectations and is far
from inevitable (Szemerenyi 1964:9). In the first place
there is the role played by historical accident : a number
of words attested only in European languages may well
once have existed in Indo- Iranian but have been dropped
because of the strikingly different cultural milieu into
which the 1-I peop les moved . Examples of this possibility
include (l ist from Szemerenyi 1964: 9; a longer list in
Kurylowicz 1956: 194- 195):
 Indo-European /a/ 41
IE *bhardhii 'beard' : Lat. barba, OHG bart,
OCS bradn (Russ. boroda)
IE *laiwo- 'left': Gk. i\m6~, Lat. laevus, OCS lfvr>
IE *skaiwo- 'left': Gk. aKm6~, Lat. scaevus
And furthermore, roms containing */a/, if of a morpho-
logical category generally displaying ablaut, may well
have been remodeled to /e/ - /o/ on the analogy of the
at the time most prevalent type (Szemerenyi 1964: 9 ).
Such may yet be visible in roots which seem to show an
*/a/ - */e/ variation, roots like: *kas--- *keJ 'cut' (Ap-
pendix II *27 ) *(s )kand--- *(s)kend- 'brilliant (vel sim.)'
(Appendix II *21), *kratos- *kretoJ 'power' (Appendix
II *30). Thus we find that the number of instances of* /a/
in interior syllables is greater than had been supposed, and
we have some reason to believe that it may in fact have
been greater still.
Ill. In final position there are relatively few cognate
pairs bearing on the question of* / a/ and schwa, but there
is no question about what cases are to be considered. They
include:

a) Skt. vida ' I know', Gk. ol8a ' id .'

Skt. vittha 'you (sg.) know', Gk. olrr&a ' id.'
OOth of which show an / a/ = / a / correspondence, and:
b) Skt. cibharrimahi 'we were carrying (mid.), Gk.
i.<jJEp6p.d}a 'id .'
Skt. bhciranti 'carrying ( ntr. plur.)', Gk. pipovm 'id.'
Skt. mtihi ' big (ntr. sg.)', G k. piya ' id.'

all ·ofwhich show an /a/= /i/ correspondence, the cor-

respondence usually subsumed under schwa. The two pho-
nemic solutions for these contrasts most generally enter-
tained clearly differentiate these correspondences by using
different symOO\s:
42 Indo-European /a /
b
laryngeal / H,c/ / H/
traditional / a/ / o/
The traditional solution, though, has been weakened by
our having shown above that [J] elsewhere in the word is
merely the unaccented allophone of / a / : it would be
strange indeed if Indo-European had a phoneme / J/ dif-
ferentiated from /a/ only in final position and only in
three words or grammatical categories. Hence the tradi-
tional solution must be given up. The laryngeal solution
suffers from this same weakness, for only fina lly would
/H 2e/ contrast with / H /; but not to the same extent, for
/H/ is supposed to occur after consonants elsewhere (as in
*stH!Os), and also before pause (as in "'p6tniH). But this may
be the only case of -CH #, and this fact wou ld tend to
weaken the laryngealist position.
Other scholars, however, have declined to reconstruct
two entities for this position in the word, and have sup-
posed that all those Gk. and Skt. forms containing / a/
derive from IE* /a/, while Skt. forms containing / i/ con-
tinue original IE (or 1-1) * /i/. This position of course in-
volves denying that the final segments of the cognates listed
above in b ) are related. In order to decide the question
whether [d]:;::::; / a / in final position, we shall have to dis-
cuss the relevant cases individually, perhaps best begin-
ning with /a/ = /i/.
I. Skt. dbhartimahl = Gk. l$E.p6~-tE.ita. Pedersen ( 1900:
80- 81) pointed out that the functional identity oft he end-
ings of t hese words is only a postulate, since the Gk. form
is both primary and secondary, the Skt. form seconda1y
only: Gk. -~-tE.fJ.a could just as easily be compared with
-mahe, the primary end ing. Furthermore the inner-Skt.
relation: 2d sg. mid. bhdrase, 2d sg. act. bhdrasi is parallel
 lndo- Eu.ropean / a/ 43
to that seen in -mahe, -mahi. And perhaps most important,
it is by no means certain that Indo-European possessed a
primary:secondary distinction in the firs t person plural
middle, and it is therefore possible that the Indic distinc-
tion is an innovation . He concludes that it would not be
unreasonable to assume that -mahe and -1-w'Jo: were phono-
logically d istinct forms of the same original morpheme,
and that -mah1 (cf. 1st sg.jUhve:djuhvi 'sacrifice') is just as
much an innovation as the subju nctive ending -mahm.
Burrow (1 949:5i), remarking simply on the series -mahi :
-mahe: -mahai, concludes that the / i/ is an original / i/ . And
Hirt ( 1928: 149) separates the forms by regarding -}Jd)o.
as -me- plus the particle -dha, -mahi as-me- p lus the particle
-dhi.
Gii.nterl ( 19 16: 12) objects to Pedersen's separating the
two forms on the very good grounds that they are so clearly
related that really impressive arguments are needed to
separate them: and Pedersen's arguments are not really
impressive. Gi.interl willingly concedes that Sanskrit has
innovated in introducing a primary : secondary distinc-
tion in this category, but feels that the innovation concerns
-mahe, not -mahi. And it does seem that this is the most rea-
sonable solution. For if we assume that the original ending
was -*medha, we can then account for the Skt. si tuation
quite simply by assuming that -i was optionally added in
the primary tenses (cf. Vedic -mas1 beside -mal} in the 1st
plural active) after the analogy of other primary middle
endings (-ai, -sai, -ta1; -ntaz), a development which resu lted
in:
pnmary secondary

{_-;::::J -medha

with -medha and -medhai in more or less free variation in

44 Indo-European f a/

primary tenses. Later final -/a/ passed to -/i/, and the

shorter form became -mahi, thus g iving the fonns:

przmary secondary

( -mahi) -mahi
-mo.lu

At this point the opposition between -mahe and -mahi was

grammaticalized , and -mahr, originally an optional vari-
ant, took over alone as the primary form, -mah1 as the sec-
ondary. Thus a primary : secondary distinction arose in
the first person of the plural as well as elsewhere. Such
seems the simplest solution, but is of course predicated on
the assumption that final 1-1 *-/ a/ > -/ i/. But it is al-
ready interesting and important that the simplest solution
point.'! to this conclusion, a conclusion quite in keeping
with our assumptions about unaccented */ a / in other po-
sitions in the word.
2. bhdrant1 = <P(povm. This equation, though accepted
for many years, has a lso bt.>en questioned for many years,
first by Schmidt (1889:2271T.), then by Pedersen (1900:
79- 80), and most recently by Burrow ( 1949:46), all of
whom assume that the final -/ i/ in Sanskrit is an original
IE */ i/ and not */ a/ or * j';J / . The reason for this, at first
sight, strange conclusion seems to be that the Skt. -Iii is
not obligatory in the Veda, and does not seem to be the
most ancient form of the neuter plural, hence presumably
not of IE date. And since not of IE date, not IE * /a/ , but
1-I */if, which is th en assumed to have derived, not from
IE */ a/ , but from IE */ i/ .
And indeed there is good reason to believe that in Indo-
European the neuter plural was not \Veil established as a
morphological category. Leaving aside the vowel stems,
we find that Vedic has the following neuter plural forms
(data from Lanman 1880):
 Indo-European / a/ 45
stem short form long form
aha, Sin:;a ahani, SiN?aQ.i (Lanman 538~539)
-nt santi (Lanman 510)
-ansi (Lanman 566)
Other stem forms are not represented in the Veda. \Vhat is
common to all these forms is the length of the stem-final
vowel, a mark common also to vocalic stems when in-
flected for the plural , and what is common to the longer
forms alone is the nasal and the /i/. It is therefore possible
to state that there are two types of neuter plural, both of
which involve lengthening of the stem-final vmvel, and
that the second involves also the addition of -/ni/ to the
stem-vowel thus lengthened:
1) lengthen stem-final vowel
2) (optional) add -/ni/ to stem-final vowel
But these rules cause trouble, slight though it be, with siinll,
which would come out by these rules as: l) *stint, 2) *siinint,
forms which would in turn become *siin (and possibly *sd)
and *siinin, respectively, because of the rules of sentence
sandhi. It seems, then, that this formulation is incorrect
and must be replaced. Since final -/n/ disappears in sen-
tence sandhi after a long vowel, it seems most economical
to state the first (obligatory) rule for the formation ofSkt.
neuter plurals as:
I) add / : n / to stem-final vowel
The effect of this rule would clearly be nullified by the
sandhi rules, and in fact produces the shorter fOrms ahd
aOd Sfr~d above: *stint wou ld appear again as *sdn (or *sd),
-iins would appear as -ds (generally). The operation of this
sandhi rule was, however, prevented by (optional ) rule 2)
which preserved th e morphological identity of these forms:
2) (optional) add / i/ to final consonant
46 /ndo-Europran /a/

This -/i/ was optional in Vedic Skt., was an lndic -/ i/,

and nothing can as yet be stated about its o rigin.
A similar situation obtains in Avestan Persian, save that
here short forms are better represent ed (Bartholomae
1895 :1 32 -1 33, Brugmann 19 11 :235- 237):
stem short form long form
dam~n afsmlmi
ayiir 9
srava. var~ Ciihi
-nt mi:ldavfln

The only difference between this situation and the Vedic

involves the nasal, and the rules :
I) len gthen ste m -final vowel
2) (optional ) add / i/ to final consonant
will handle all these fonn s. The nasal is not present in
Avestan, and can easily be explained as secondary in Indic.
\Vc may therefore regard the Avestan ru les as essent ially
the 1- I ru les, confirmation for which comes from the nasal -
less lndic catvdri 'four'.
The fact that the addition of -/i/ is an optional ru le even
in Ind o-Iranian does n ot of itse lf prove that -/i/ derives
from something other than IE • / a / ( = [;)]). But there is
evidence supporting such a conclusion. Burrow ( 1949 : 46),
followed by Palmer ( 1961: 246), feels that the case for • I ii
is clinched by the sit uation in Hittite. For Hittite, it is clear,
beside cases of l-1 rule I ) (which must now be taken as IE
rule I ) such as kururtfl.A 'enemies', a lso has fomts which
seem to conform to rule 2), such as kurun1_11.A, thus suggest-
ing that rule 2) also is of IE date. If one identifies the 1-1
/ i/ with t his Hittite / i/, then clearly qi.poPTo: and bharanti
are no longer d irectly comparable, but are different adap -
tations of something that may have been *bherr!nl. That is:
rule l) will have applied everywhere, but Greek will have
 /nda-Eumpean /a! 47
rewritten rule 2), now obligatory, to: add /a/. The reasons
for the Gk. substitution o f / a / for /i/ will remain uneluci-
datcd by this theory.
Hut the Hittite evidence is in fuct not conclusive, for
Hittite also has neuter plurals in -a, and it is therefore not
justifiable to compare I-I -/ i/ solely with Hitt. -i. One
must take into account also Hitt. -a, the more so because
i-forms are restricted to a) i-stems and b) -r, -l, and -r/-n
stems, and do not occur (save for ljumanti 'all ') in -nl- stems.
In -r, -!,and -r/-n stems they alternate with no ending at
all, while in i-stems th ey alt ernate with -a. Brosman ( 1962:
63-65) has suggested that the -i in 1-stems is the original
stem of tlte word without any ending, and that the -a of
i-stems, occurring mostly in adjectives, represents a con-
traction of -aja, an ending which also occurs. In the
liquid stems -i replaces ,0, and hence provides a means of
distinguishing the singular from the plural in these stems.
The reason that -a, the ending of other consonant stems,
was not chosen fort h is function is that in the i-stems a d is-
tinction had arisen between -i, which occurred mostly in
nouns, and -a < -aja, which occurred m ostly in adjectives.
And since -r, -1, -rj-n words were mostly nouns, -1 was ex-
tended to them; and conversely, since -nt- stems were
mostly adjectives, the -i ending made no inroads on the
original -a. Hence the -i is a secondary entry into the
neuter plural, not an inherited ending, a nd not to be com-
pared phonologically with I-I -/ i/. On Brosman's inter-
pretation, therefore , Hittite also had an -/a/ in neuter
plurals, and Hitt. -/i/ does not provide support for inter-
preting 1-1 -/i/ as IE* /i/. We should compare bharanti =
q,i.pona = !Jumanta, deriving all from original *-Onta.16

3. Skt. mahz = Gk. piya. This comparison, which has

frequently been made, and which provides one of the cor-
nerstones for the d octrine of final -/d/, is the weakest of
48 Indo-European /a/
those thus far discussed. Evidence for · / a/ , aside from the
Gk. fonn, is provided by Arm. mec 'gross', instr. mecaw;
and for -/ i/, aside from the Skt. form, by Goth. mik1ls
'gross' (<PGmc. *mekila;;: Szemerenyi 1952:48~49 ) and
Hitt. mekki'sehr' , mekkis 'gross' . There can be no doubt that
the>e words are related, but there is doubt as to the quality
of the second vowel.l1 It rather seems as if different lan-
guage groups have adapted an uninflected *meg- to their
own inflectional tastes in various ways. In view of this un-
certainty, it is perhaps best not to use this comparison as
evidence for the development of final -*I a / . But at the very
least ~Jiyo: = mahi d oes not prove incompatible with a rule
which provides that final -/ a/ > [d] > 1-l-/ i/.
And it docs in fact seem to be a ru le that IE final */ a /
appears as / i/ in absolute final position in Sanskrit. An~
other case is provided by:
4. Skt. iti 'so, aufdiese Weise', Lat. iw 'thus, so' (M. 1.86,
E~M 325; W-H 1.722- 723 deny the connection ).

The question then arises whether all fina l " /a/ pass to / i/ ,
or whether some rema in -/a/, a question that is answered
by the equation:
5. Skt. thd 'here' (Pali idha, Ave. :&, OP ufO. ), Gk.
lOo.ytvrj~ ' aboriginal' (M. 1.94, F. 1.715). C( Lat. ibi
( < *zdhai), though some take this from *idhei (W-H
1.722-723).
From th is one case we can see that the same rule applies
in fi nal position as elsewhere: unaccented / a /> / i/.
Two cases remain, however, two important cases which
yield a result counter to ru le and to expectation :

6. Skt. vida 'I know', Gk. olOa ' id .' < "woida (P. 1125,
F. 1!.357 ).
 lndo-Europeml / a/ 49
7. Skt. vittha 'you (sg.) know', Gk. oloiJa 'id', the sec-
ond jXrson sin gular of *woida, <
*wotlstha ( =
I /woirndha/ /) .

Neither of these words contains accented / a / , so both are

exceptions to my rule.18 Various solutions to the opposi-
tion veda * bharanti have been proposed. 1) The tradi-
tional explanation again posits the two phonemes / a/ and
j -;Jj , while 2) the laryngeal explanation, in doing in fact the
same thing, derives the forms from *woidH2e (and
*w oidstHze) and *bherontH respectively. And the laryngeal
hypothesis finds impressive support for *wmdH2 t in the
Hitt. -hi conjugation frequently , though not always, com-
pared with the IE perfect. For the endings of the first and
Sl.'COnd persons singular of this conj ugation are -bz (Luvian
preterite -ba) and -li, with, in the first person, the /h/ de-
manded by the theory . And in the second person the / h/
is supposed once to have been present because the /i/ did
>
not palatalize the / t/ to / z/, as it did for instance in / til
/ zi/ in the third person singular of verbs of the -mi con -
jugation. But the second part of this argument is incorrect,
for -li must derive from earlier *taz (or *thai') (Kurylowicz
195Ba: 236), and so the absence of palatalization is to be
expected. But the /h/ (though not the / e/!) of the first
person is assured, and it is therefore not impossible that IE
*/ h / , once present in the IE form, prevented the passage
of IE */woidha/ to Skt. */ vedi/ . Thus far we might be
inclined to go along with the !aryngealist argument , an
argument originally framed in other tcnns. But the */ h/
will have prevented the passage of* / a / (attested in Greek,
Sanskrit, and L uvian) to / i/, not the passage of (unat-
tested ) "' / e/ to / i/, or indeed have caused the passage of
*/e/ to* /a/ . And it \\'ill have been an I E phoneme* / h/,
not an IE morphophoncme I / HI I with the properties of
50 Indo-European /a!
appearing now as [ ], now as[;} ], now as [h]. We shall be-
low provide anoth er explanation for the preservation of
-/a/ in Indo-Iranian, without ta king a stand on whether
the IE form was " woidha or *wozda.
3) Pedersen (1900:78 - 81) and Burrow (1949:45 - 46,
57), as we have seen, deny the connection ofSkt. forms in
-/i/ with European forms in -/a/, and derive /a/= /a/
from IE* /a/. But we have seen a lso that their denial of the
correspondence -/i/ =-/a/ is unjustified. 4) Another ex-
planation derives both Skt. reflexes from IE */a/. Thus
Wackernagel (1896:6) states: 'Dunkel blcibt vorliiufig,
warum von den Personalendungen, welche griechisch en
auf a entsprechen, die einen i haben: -mahi, -vahi, die an-
dern a: -a, -tha des perf.'. And Pedersen, abandoning his
position of 1900, holds ( 1926:27): 'La fin de mot -a(-;,) a
donnC en sanskrit tant6t -a, tant6t -i (mahi = gr. pi.yo.);
c'est a tort que j 'ai voulu n ier une des a lternatives, KZ
36,76ss. J e crois main tenant que -a est le traitement de Ia
pause, -i le traitement du sandhi.' What Wackemagel
found inexplicable, Pedersen found explained in phono-
tactic terms, sentence sandhi.
And it seems to me that, however theoretically unsatis-
factory and d isquieting their conclusion may be, Wacker-
nagcl and Pedersen are correct in deriving a ll forms show-
in g final -/a/ in European languages from IE-*/a/,
regard less of whether Sanskrit has -/a/ or -/i/ . Pedersen's
conditioning factor , however, is not sufficient, being only
a guess, for one wonders why -/a/ should remain -/a/
before a pause but pass to -Iii in sentence sandhi. lt is im-
possible for us at this distance in time to determine which
was the prepausa l form , an d even whether fina l position
in the sentence had any affect on vowel color. S ince we
cannot know about these things, we must d iscard Peder-
sen's attempt at providing phonological con ditioning, or
 Indo -European / a/ 51
at least hold it in abeyance because itself requiring expla-
nation and justification. It is best simply to admit (with
Wackernagel) that no phonological conditioning is dis-
cernible. But we can at least state one thing about the
distribution of unaccented -/a/ and -/ i/ which is not sub-
ject to dispute: -/a/ occurs in the singular of the perfect
active, and only in the singular ofthe perfect active, while
-Iii occurs everY'vhere else. Since we can specify the
morphological environment, we are free to suppose that
the morphological, not the phonological, environment is
relevant in stating the distribution of word-fmal -/ a / .
And we can then state in rule form: unaccented -/ a /
passes to -/ i/ except in the singular perfect active.
Those who feel that morphological conside rations can
affect phonological d evelopments will have no difficulty
in accepting this type of explanation, while those who
allow only phonological conditioning for sound change
will of course be unable to accept it. 19 To answer objections
from this quarter, I ,...·ill state first that the rule given above
seems almost inevitably demanded by the data, for setting
up a new phonological enti ty solely to account fOr two
morphological clements is most uneconomical. And
though I shall not in fact provide a truly satisfactory ex-
planation, I would suggest that certain factors were at
work other than simply that *wotda and *wottslha are
perfect forms.
There is the fairly strong possibi lity that the second
singular ending of the perfect is the same originally as the
ending of the first person plural middle, that is to say, in
both cases* / dha/ . But in the one case the* / dha / passed
to / dhi/ in Sanskrit, in the other it did not. The reason,
or a reason, for this divergence may well be the fact that
-/dha/ of * /mcdha/ was not opposed to any other ending
-mtdh-, while -dha of the second singular perfect active was
52 /11do -European / a!
opposed to the -dh1 of the imperative. To be sure the vo-
calism of the root differed, olo&a beside tO&t, but there was
nonetheless an opposition in the ending lacking in the first
pluraJ middle. And this fact leads us to consider the mor-
phological role of the -/a/ in the other sure case in which
it occurs, in the neuter plural. We have already seen that
the -/a/ here is secondary, not originally part of the end-
ing, and that it vvas added in Sanskrit to counteract the
tendency toward loss of final nasals and stops. And in
Greek the regular phonological development of *bherOnl
would have been 9{pov, the same form as the neuter sin-
gular. H ence the -/a/ in Greek (originally) senred on ly to
protect the -/t/ and was not itself the mark of t h e plural:
the long vowel marked plurality. Nor was *-/a/ the mark
of the first person plural middle: it served only to prevent
Gk. *cfXpoJl!<*bheromedh and Skt. *bhariimat<*bheromedh.
And the -/a/ of *ita also lacked morphological signifi-
cance. We can thus state that morphologically insignifi-
cant -/a/ passes to /i/.20
The -/a/ of *woid(h)a was grammatically relevant, and
was the only phonological device which served to distin-
guish *wmd(h)a from *woide, but also from possible con-
fusion with other middle forms in -i. We may thus rephrase
our ru le by extending it to cover all grammatically rele-
vant forms rather than specifically forms of the perfective
aspect, though it is on ly here that the rule has an oppor-
tunity to be applied. And perhaps, in order to make the
rule slightly more palatable to those \vho require phono-
logical conditioning, we may suppose that grammatically
relevant forms, if not exactly stressed, were at least not
unstressed either, and held a somewhat middle position
between stressed and unstressed. They were not stressed
to be sure, but were not unstressed. Hence on this inter-
pretation the rule for Indo- Iranian is: -/a/ passes (to[;)])
 lndo·EuropMn fa/ 53
to / i/ in unstressed open syllables, but remains elsewhere.
But I prefer to avoid such q uestionable phonology and to
assume that: */a/ in unstressed open syllables passes to
1-I / i/ save wh en grammatically relevant.2 1
This part of the argument, then, is over: "'/a/ between
consonants or between consonant and pause remains when
accented or when in a closed sy llab le , otherwise it passes
to I-I 0" initial ly, / i/ elsewhere. Perhaps, though, it would
be well at this point to provide gross total figures for the
different d evelopments of* / a/ in various positions. These
figures will show that there is, if not an absolute unanimity
in forms conforming to my rules, there is at least a statis-
tical bias in favor of assum ing only */a/ and not both
*/ a / and */ 4. In the tab le I give maximum figures for
each situation, and do not take into account the refine-
ments made above which involved dismissing a number of
apparent exceptions. In other words the totals include
every instance numbered or coun ted in the body of the
text. Only those forms relegated to the Appendix are not
counted. From the ta ble it wil l be clear that there are 91
positive instances and only 14 negative, a ratio of about
6: I.

Posilwe Negatwe

# _C- 21 12
-C_C- 34 16

- C _ C:;t
- C_ #
Total 58 21 12 12 2
91 14
54 Indo-European / a/
The rules formulated thus far han d le cases of* /a/ aft er
a consonant , but make no provision for* /a/ after a vowel.
\Ve have seen above that in nonfinal sylla bl es unaccented
• / a / a ppears as le ngth after /i/ or */u/ as in the cases
mentioned there (above p . 24), and in :2 2
inklwti 'moves up and down' < *i-ankh- 'mixes, stirs up'
(with I-1 root ankh-; M. 1.20, 95 ).
tjati 'sets in motion' <*i-ag- (M . 1.95).
irtsati 'wishe:; to inc rease', reduplicated desiderati ve to
r:dhat1 <*i-aldh- (.\1. 1.1 24).
pratipal) 'adverse opposite' <*proti + ap- 'water' (M .
11.36 1).

And it is quite clear that it appears as lengt h a lso in fina l

fX)Si tion, as in the cases men tioned on page 16, and
also in:
Skt. patni ' mistress, w ife', Gk. w6TPto: ' mistress'< *potnia
(:vt . 11.202-03, F. 11.586-587, Szcmerenyi 1964a:391 -
395).

Skt. dSrii(ni) ;tears', Gk. 00Kpvo. 'id.' <*(d)akrua.

The rule can thus be stated very sim ply and very generally
for the d evelopment of I E */a /. It remains / a / in t he
European languages, but develops in Indo-Iranian as
follows :

/ a/> /i/ ;c_ [~]

> /: / ;v _ [~]
> / a/ IRCJ
elsewhere, i.e. L___:_
 Indo-European / a/ ss
There is thus no need or call to contrast [a] and [J] in
lndo-European: they are allophones oft he same phoneme
/ a/.
And if [a] and [:.~]arc allophones of / a/ , then dearly
there is no reason to suppose that [d] became an a llophone
of / a/ only after having developed from an earlier laryn -
geal / H / . That conclusion rested on distributional state-
ments regarding /H / , namely tha t it appeared as [hj
initially, [:] after vowels (save sometimes in Hittite ), and
as [d] between consonants. If we want ( for reasons which
wi ll have to remain obscure to me ) to retain this analysis,
we shall now h ave to assume a rule :
/H/ > ]>] /C ~ c
]:]IY -
!h] / - V
Jal ! l~ cl
L- RcJ
a rule that will encounter difficul ties before a vowel, and
which is unm otivated by the comparat ive evidence.
Hence, though / h / may well have been pan of the IE
phonemic inventory (Wyatt 1964:149, Szemerenyi 1967:
89-90), it d id not appear between consonants as [dj, and
our guesses a s to the nature of prior stages of Indo-Euro-
pean shou ld h enceforth feature fewer h's. But there should
be more */ a / .
 Appendix I

Since it is customary in discussions ofiE phonology to make a

guess as to the nature of earlier stages of Indo-European, I shall
make one here , obviously not in the hope of settling the ques-
tion , but rather as a pledge to the sincerity of my proposals a nd
to forestall possible objections leveled at my proposals based on
presumed earlier stages of I ndo-Europcan. And I shall be able
to account further for the comparative rarity of •;a; in our
PTE reconstructions. In what follows I restrict myself to t he
sho rt vowel system, though long vowels and diphthongs could
ca~ily be accommodated, and in fact per force enter in later on .
In my discussion I shall employ the symbols PTE, PPIE,
PPPI E: PIE will stand for the language reconstructed on the
basis of cognate sets in the various IE languages; PPIE wil!
stand for the language (or de rigucur, the vocalic system)
which lies immediately behind PIE ; and PPPIE is the stage
anterior to PPIE, the earl iest system of I E vov.rels for which we
have any evidence at all . The num ber ofP's wil! be matched by
the number of asterisks when hypothetical fonns are cited: e.g.
**'"dg- > >
*"'dg- •ag-.
It seems that Indo-European always had the five ~hort vow-
els / i e a o u/, whatever else it may have had, and that these
vowels could appear both accented and unaccented; the ac-
cent seems to have had a strong stress component. The PPPIE
system contained.
accmtrd unaccmted

PPPIE

o6
 Indo-European f a/ 57
Later the unaccented mid-vowels disappeared through vowel
syncope, at least in open syllables and in words of more than
two syllables, and there resulted (in PPTE ) a tripartite opposi-
tion in unaccented syllables: the five-vowel system remained in
accented syllables.23
accwted unaccented

PPIE

At this point, the immediately prcablaut, and hence imme-

diately pre- IE period, /a i u/ could appea r both accented and
unaccented, / e o/ cou ld appear only accented. A~ a result of
a grammaticali zation of this accentual relation, accent came
to imply / e/ or / o/ , / a i u/ implied unaccented position, and
/e/ a nd /o/ were introduced into accented positions where
they previously had not been, notably in the present tense (im-
perfective aspect) of verbs and other categories later charac-
teri.:ed by the / e / -gradc. Only well-establ ished words, words
of high frequency such as **rig- 'set in motion' were able tore-
sist this tendency. T he reason t hat / e/ was felt to characterize
the present (im perfective) seems to be that there were in Indo-
European several words of imperfective meaning which con-
tained the vowel / c/ in their root, verbs like **hfuro- **bhers
**bhert .,.bherm(e) **bhert(e) **bhemt (c( L at.ferofersfatftrimus
ftrtejnunl), a verb which may have had an earlier paradigm
"**bhero- ***bheres * **bheret etc.
When society became sufficien tly complex that it was no
longer possible to utilize separate words (suppletive para-
digms) for semantically related but aspcctually diffCrcnt con-
cepts, the vowel / e/ was inserted between the firs t two conso-
nants of the root (if there were two initial consona nt<;) . That is
to say, old "paradigms" such as 'say' 91/,.U i.pW tl'lTOP in Greek
became ineffi cient and cumbersome and tended to be replace d
by more regular paradigms like Alyw Al~w k'X€~0:. There were
other such devices, but the insertion of / e/ was at this stage the
most prociuctive one. H ence to a root ** sghO- ' have, take'
(which may or may not have come from *** st.t:hO-) with the
58 /ndo-Europran / a /
paradigm **sghOm * *sghis **sghi(t ) **sghOm(t ) ** s_t[hit(f)
sgMn{t) there was formed a new paradigm *sighoa (with -/ a/
from the perfect ??) > *segho:, *sighesU) *sighet(t) *Jighom(t)
sigful(t) S~?hont(r) after verbs like * bhiro:. Because t his new im-
perfective paradigm had a vowel before the fina l consonant, a
vowel was introduced in the ending of most original imperfec-
tives so that **bhiro: **bhirs **bhirt became *bhiro: *bhires(t)
*bhlte/(1 ). Only rdics like Lat. Jrrofmftrt as opposed to Skt.
bhdrdmz hhdrau bhdmt1 and verbs of frequent occurrence like
'to be' held out against this trend.
If imperfective {etc.) forms were created to unaccented roots
containing /i/ or / u / such as **lzkw- in **!JkwrJ-- "I leave', they
took the shape *llik 11'()-, *lelri1k"'a rather than *fi/(>no- , *ldfkwa.24
Preswnably it wa.<; this insertion of an accented vowel before
another vowel (or semivowel) which converted t he old stress
system of accentuation into the late JE pitch-accent system ,
the system continued in Vedic Sanskrit and Greek. And this
new system of accentuation which op(Xlsed unaccented to ris-
ing and rising to falling (e.g . in *!ilk"'-) tended to preserve un-
accented vowels and did not allow them to be lost through
syncope. 2 :>
But if the verb contained a vowel other than /i/ or / u/ in
what sublcqucnt ly became (or remained) the imperfective as-
pect, the insertion of the vowel /e/ resulted in a long vowel,
if the root began with a single consonant or a stop consonam
preceded by / s/, as in *diu:- 'place', the original root of which
must have been **dht- with the imperfective paradigm:
PPPIE > PP IE > l'JE
(CV)dhem(i) (CV)dhCm i (CV)dhCcmi = dhe: mi
{CV)dhb(i) (CV)dhCsi (CV)dhfesi =
dhe :si
(CV)dhh(i) (CV)dhfti (CV)dhfeti = dhc: ti
(CV)dhcmC (CV)dh mf (C\')dhmf
(CV)dhetf (CV)dhtf (CV)dhte
(CV)dhCnt( i) (CV)dhCnti (CV)dhCnti
The PIE situation is exactly mirrored in the Skt. paradigm
which shows 1st sg. dddhiim1, 1st plur. dadhmd~1 beside the rc-
 /ndo-F.uropean /a / 59
modeled dddhiimab. Gk. had remodeled the 1st and 2d p lural
fonns after the 3d plural al ready before our earliest records.
*sista:m1derives from *sisuarm, 1st plu r. 5/.Stamf( to a root *5/a- ),
and this verb together with *dfdhe:m1 set up the IE pattern:
long vowel in singul a r, no vowel (or / a /) in plural which
changed *didnum (<*didroml): d1dmi 'give' to *dfdoom1
( =*dido :mJ'j; *dldmi(scen in Skt. ddd&m, dadmdl]). This same
relation u ltimately accoun ted for the change ofGk. *dtJknrwm
: deiknumm to 8dK.vfJp.t : 8dKVV}l<V.
In other cases /C/ may have replaced ; a;, as in the examples
mentioned above {p. 41 ). But when the root consisted of aeon-
sonant cluster plus vowel rather than of single consonant (sym-
bol TRV ), the / C/ o r / 6/ was frequently inserted between the
consonants, thus gi ving rise to dis yll a bic roots. Su ch roots were
then of the shape Tt/o RV- beside TRt/oV-. Thus from **lla -
'lift , bear' there was formed an allomorph *tila- (seen in
n;\a.p.Wv 'carrying-strap') beside t he other allomorph *ilea- >
*tla:- (in TirhiiKo: 'I have endured', Lat. !titus 'carried'). The
later disyllabic roots, then, were simply adaptations to the new
scheme of things of old roots endi ng in -/ a / .26
T he prt."Ceding sketch is obviously just that, a sketch, a nd is
fil.r from being a complete account of the origins of IE verbal
root structure : it is intended me rely to point the way and to
suggest possibilities for r econstru ction opened up by t he iden-
tification of / a/ and schwa . I do not pretend to have faced, let
alone answered , all the myriad p roblems m y (X>Sition entails.
\Vh at is more, my remarks are by the very nat ure of the case
somewhat fanciful, and a rc at least in part intended also to
forestall some criticisms which migh t be lcvckd by t hose who
feel that / d/ (or / H /) is demanded b y the requirements, if not
of comparative reconstruction, at least of IE root structure.
Nei th er /d/ nor / H / is required by t heoretical considerat ions,
and I hope to have shown that /a/ is.
 Appendix II
I n this appendix I gather t ogether all those casts which I have
found for which IE */ a / in positions other than initial has
been proposed by various scholars in the past. There are sev-
eral categories into which these words can be fit: I ) words
which may in fact show IE */a/ but which cannot be used to
establish */a/ because they are either a) nursery words or
b) onomatopoeic; 2) words which might be related and which
have been proposed as examples of* /a/, but which arc un-
likely because of some phonological difficulty or on semantic
ground~; 3) word s which, though proposed at som e time in the
past, cannot be rdatcd, or at least are most unlikely to be re-
lated. Clearly there will be some overlap betv•ecn 2) and 3),
and in gener al I include in 2) on ly those instances for wh ich a
protoform can be reconstructed. I number cases in I ) and 2)
*If[ to~how that they might be added to the instances of*/a/
given in t he body of the text. Words listed in 3) will be num-
bered t iff.: they cannot be added to the listofmorcor lesswcll-
cstablishcd cases of* / a/.

Ia) Nursery words (Pedersen 1900:83 )

* I. Skt. tatdl] 'father', Gk. -ram, Lat. tala, Russ. tdta
(M. 11. 471, W-H 11.650). Turkish has ala.
*2. Skt. nami'mothcr', Gk. J.>O:vva 'aunt ' (M. 11 .304 "redup-
lizierte Lallwort" , F. 11.1 3 1 " La\lwort")
1 b) Onomatopoeic formations
*3. Skt. balbaliikarotl 'begayer', Lat. balbw 'bCgue'; Skt.
barbara!] 'stammering', Gk. {3/xp{Japm 'foreign'

60
 /ndo-Eurupean /a/ 61
(M. Il.411-412, 420, F. 1. 219-220, W-H 1.94, Kury-
!owicz 1956 : 190)

•4. Skt. gaiijanaf;, 'despising, contemn ing', Gk. ya:yyo.ivnv-

rO p.t:Tt) y{Awrm 1ip00'1iO.i;av (H es.), 0£ cancrttan
'railer', OCS gpgnpli 'murmurer' (M. 1.315, F. 1.281 ,
Kurylowicz 1956 : 190) . f risk seems to accept connec-
tion of the G k. and Skt. forms, but Mayrhofer feels
that , since the Skt. word appears for the firs t time in
the M iddle Ages, it is not to be compared with t he Gk.
and O E forms . Kurylowicz fee ls that the words are
onomatopoeic in origin.

•s. Skt. carkarti 'makes mention of, praises', Gk. KO.pKalpw

'quake' (LSJ), i.KtipKo.tpov·-.}Ot!Jov m 'tl &1im'.\ovv (Hes.) ,
Lat. cannen 'song', ON hriidr 'Ruhm , Lob' (M . 1.377,
F. 1.789, W- H 1.1 69- 170). The Lat. form at least is
not to be compared here, for it derives fi-om •can-men
to canert 'sing'. Frisk , in comparing the Skt. form, treats
the word as originall y o nomawpeoic.

•6. Skt. kokiltil; 'the Indian cuckoo', Gk. KCH!Ka.Aia~, a kind

of bird, KaVa~, a seabird ( M. 1.268-269, F. 1.801 - 802,
Ku rylowicz 1956: 189). Both Mayrhofcr and Frisk
take these words as onomatopoeic, but Ku rylowicz,
comparing Lith. kaUkti 'heu rler' a nd Skt. kiiutt 'cries',
feels that all these words may derive from a long-
diphthongal form .

•7. Skt. krrHatz 'shrieks', Gk. KpatJY11 'crying, screaming',

Goth. hrul<.jan 'chanter (coq)' (M . 1. 281, F. 11.10-11 ,
Kury!owicz 1956 : 189). Both Frisk and Mayrhofer
~tress the onomatopoeic nature of these words, a fact
mentioned also by Kurylowicz, but he favors deriva-
tion from a long-diphthongal root because of Li th.
krokiU, krOkti 'dl.ler' and other forms which point to
•krii·.
62 lndo-J::uropran /a/
•s. Skt. kdkhat1 'laughs', Gk. KO:Xilfw ' laugh aloud', Lat.
cachmno 'laugh aloud ' (;\1. l.l 36, F. 1.804, W -H 1.1 26,
E-M 80, Kurylowicz 1956: 190). All authorities recog-
nize that these words a re sound-imitative, and only
Emout -M eillct consider them derived from an IE
word. Kurylow icz, however, objects that they cannot
be, si nce Gk. x cannot cotTcspond to Skt. / kh / save
after / s/.
"'9. Skt. lalafla ·'onomatopoetisch vom Laut eim:s Lallcrs"
(M. ), Gk. Ao.Mw 'talk, prattle', M:i\os 'talkative', Lat.
lallo 'sing la ll a' (M. III.92, F. Il. 77 , \V-H l.752- 753,
Kurylowicz 1956: \90 ).
"'10. Interjections such as those mentioned by Kurytowicz
(1956: 190): Skt. uvi, Ave. ~'!ij-'61, av61, Lat. vat, Gaul.
gwat, Goth. wm, Leu. vm .

2) The second category consists of correspondences which,

t hough not excluded theoreticall y as was the case with 1), are
not acceptable as certain instances of / a / because of some
phonologica l or sem ant ic weakness in the eq uat ion .

(* I 1. Skt. bdlam 'power, strength, might', comparative a nd

superlative Mliyrln, btilt.Hhai.l, Lat. dihdH 'weak', OCS
bol!fb 'grOsser, besser' , Gk. fU>..T((JO~ , {U>.nOTm 'bener,
best' ( Pedersen 1900:82, ~- 11.416- 4 18, F. 1. 232,
W-H 1.326- 327). This word can be considered to
contain IE / a/ only if Phryg. /3a>.fw 'king', Oir. bnlc
'strong', Welsh balch are related ; and ifGk . f3l>.T(pos is
a repl acement fo r a n earlier* {1/Y.>..T(pos; and if the Skt.
word is not a Dravidian loan, as has been held by
some (Burrow 1946 : 19, denied by T hieme 1955:44 7).
The Lat . word is not related, since it d er ives from dibeo
(: 12. Skt. dhdn uan- ' bow', Ger. Tanne 'fir-tree' (M. 11.90-9 1,
Pedersen 1900:82, T hieme 1953:550- 551). Pedersen
is reluctant to adopt this comparison because the rela-
 lndo-J::uropran /a/ 63
tionship of the meanings is very distant. rhicme makes
the strongest case for the connection .
(*13. Skt. bhandkt1 'breaks', Lat. frango 'break' (M. II .469,
W-H 1. 541, E-~ 251-252, Pedersen 1900:78). This
comparison is va lid only ifSkt. glri-bhrdJ-, of uncertain
meaning (M . 11.527), is related. But even at that all
other evidence points to a root *bhtg-!*bfung- (Mayr-
hofer).
(* 14. Skt. brdhman- 'Form ung, Gestaltung', Lat. jfiimm
'priest' (.\-1. 11.452-456, W-H 1.512-513, E-:\.1 239,
Pedersen \900:82, Puhvel 1964: 1- 7). The Latin form
is more likely to be related to various Germanic words
such as Goth. bliitan 'verehren'
( *15. Skt. davab 'Brand ', doman- 'Brand, Qual' to dun6ll
'burns', Gk. 8a~w, _ ~2..~ 'kindle' (M. 11.49, F.
1. 342- 343, P. 179- 181).
('" 16. Skt. gdbhasllb 'arm, hand ; beam of light ', Lat. habto
'have, hold' (Giintert 1916:8). There is nothing to
recommend this equation.

( *!B. Skt. iarkarab 'Kiesel, Stein', Lat. calx 'small stone used
in gaming' ( L' . 305, F. 1.22, W-1-1 1.145, E-M 89,
Pedersen, 1900:78). The l-1.1. word is probably a
borrowing of Gk. xO:At~ 'small stone', and the Skt.
word may well be related to Gk. 1ip01aJ),1) 'seashore'.
(*19. Skt. karparab ' cup, ]XII', Gk . .W.)..11TJ 'pitcher' (M. 1.174,
F. 1.767-768, Pedersen 1900: 77). Frisk is probably
right in holding that the Gk. word is 'w ie so viele
Gef<issnamen ohne sichcre Erkllirung': he does not
mention the Sk t. form.
1 *20. Skt. kcimpau'trcmblcs', Gk. K0.fJ.1:"TW 'bend', Lat. campus
'plain, fie ld ', Lith. kam.paJ 'Winkel ' (M. 1.160, F. 1.775,
64 lndo-Europtan /a /
W-H I.1 48-- 149). All th ese forms can easi ly derive
from a *lwmp-, but the semantic connection is weak.

*21 . Skt. candrdJJ. 'shin ing, Moon', uiudali 'shines', Gk.

KirvOapo•rliv~pa{, Lat. candidw 'wh ite', Gaul. cann
'brilliant' (M. 1. 373, F. 1. 776, W-H I.! Sl- 152, E-M
92). The Skt. form points to an original *kend- because
of its palatalizatio n, but this may be an innovation .
All other forms point to *kand-.

*22. Skt. kandaraf; 'cave, glen', Gk. Kirv&Mot·Kot.AWJ.UX-ra,

fXtiJpa (H es. ). M ayrhofer (L 152 ) hesitates between
this connection and conn ection w ith kak.undn.ram of
Asiatic origin. Frisk does not discuss th e Gk. word.

*23. Skt. kapftlam 'cup, skull', Lat. caput, ' head', OE hafola
' Head' ( M. I.\ 55, W-H 1.163- 164, E-M 99, Pedersen
1900:77) . There does seem to have been an IE root
•kap- 'head', variously extended in the various lan-
guages. But it must h ave been accented, and the OE
form proves the Skt. accent secondary.

*24. Skt. kdprt(h) 'penis', Gk. Kthrpm 'boar', Lat. caper 'he-
goat' (M. Ll5 7, F. L783, W-H 1.1 57).

<,*25. Skt. kapa{l 'zwei hand voll ', Lat. capiQ ' take' ( M . I. 154,
W-H 1.159-160, E-M 97 , Pedersen 1900:77) . There
seems no reason to connect these words, a nd Mayr-
hofer suspects the Skt. form to be connected with
kavaW.IJ 'a mou thfu l, morsel', and of non-Aryan origin.

( *26. Skt. iapha 'hoof' , O H G htro/'h oof' (P. 530, U. 303,

Pedersen 1900 :82, Mayrhofer 1952:27). Mayrhofer
accepts the equation, b ut it is most unlikely because
the words point to different vocalic lengths in Indo-
European. Furthermore one must not ignore OCS
lwpytQ ' hoof' which must also be related, though point-
ing to a different type of IE k. Rather all these words
must be of onom awpoeic origin.
 lndo-Eurapran / a/ 65
f •27 . Skt. M.sti, fdsati 'cuts' , Gk. KWtw 'split' , Lat. mslro
'cast rate' (P. 586, U. 306- 307, F. 1.806, W-H 1.1 79-
180, E-M 104, Pedersen 1900:82). The La!. form is the
only one pointing to an original /a/.

' •28. Skt. ftitrulJ. 'enemy', Olr. cath 'lutte', PGmc. *hapu
(P. 534, .F. 1.93 1, Mayrhofer 1952: 27, Kurytowicz
1956: 191). IE *kat- is rendered unlikely on the one
hand by G k. ~~:07"0~ 'resentment' and OCS kotora
'Kampf'.

{* 29. Skt. ftiviral]. 'puissant', Ol r. caur ' h6ros', Welsh cawr

'geant' (P. 592-594, U . 306, F . II.54, Kurylow icz
1956 : 191 ). Skt. ifiral]. 'fort, vaillant , hCros' and Gk.
Kiipw~ ' having power or authority' combine to make
it seem likely that the Skt. fonn is secondary, for med
on the analogy of bhii.-: bhdvitum, and thus not directly
comparable with the Celtic fonns.

(•30. Skt. krdtu~1 'power, understanding', Gk. Kp(x.Tos

'strength', KpaTVs 'strong' (P. 531 , M . 1.276, F. 11.9-
10). All other Gk. fonns point to •je/.

(*31. Skt. lavif;, favitra- 'sickle' to IunAti 'cuts', Gk. ,\atoP

' ploughshare' (M. 111. 93, F. 11.73, Kurylowicz 1956:
191). The Skt. fonns a re attested late only, and arc
almost certainly late derivatives of funflt1~ and hence
not comparable with Gk. Aatov.

(*32. Skt. k.yatrdm 'might, power', Lat. satdles 'attendant'

(M. 1. 285, W -H Il.481, E-M 595, Pedersen 1900:82).
This comparison was discarded by Pedersen because
the semant ic difficulties were too great. Satellts is in
fact most likely an Etruscan loan-word.

*33. Skt. k$d)'ali 'rule~ , possesses', Ck. tmio).W".t 'get' (M.

1. 287, F. Il. 30-33). Mayrhofer favors connection,
Frisk is opposed.
66 Indo-European /a/
*34. Skt. lambate 'sinkt' , Lat. liibor 'slide", !abo 'begin to sink'
(!\.1. 111.44, \\'-H 11.739, E-M 333-334, Pedersen
1900:78).
"'35. Ave. ma.s)'J 'grOsser', mas6 'GrOsse', Gk. p.o.Kp6~ ' long'
(F.ll .l64-!65, 224- 225, Pedersen 1900:82-83).
*36. Skt. mandtram 'dwelling, house', mandur& 'Pferdestall",
Gk. p.&vOpa 'enclosed space' (M. 11.582, F. 11.169,
Kurylowicz 1956: 192). All these scholars connect
these words, but all also assume borrowing and that
the word is not of IE origin.
*37. Skt. rnaiiJWJ 'beautiful', maiijuldJ; 'id.', Gk. p.D.yyo:vov
'philtre', OPruss. manga 'courtisane' (\1. 11.553, F.
11.1 55, \V-H 11.28- 29, Kury!owicz 1956 :191 ). !\.1ayr-
hofer fe els that the Skt. forn1 is not likely to be related
to European forms, and suggests Dravidian connec-
tions.
*38. Skt. midaQ 'fat, marrow', Gk. p.o:tr'is 'breast', OHG
mast 'Futter, Miistung' (P. 694-695, M. 11.683 - 684,
F. 11.1 83).
*39. Skt. sahlrlm 'Meer, Meerflut', Gk. (lAs 'salt', Lat. sal
'id.' (F. 1.78-79, W -H 11.465-466, Thiem e 1953:561
562). That there was a European word *sal- 'salt'
there can be no doubt. That *sal- was Indo-European
also depends on whether one accepts Thieme's identifi-
cation with this root ofSkt . sahlrim.
'"40. Skt. uccalat1 'springs upward', Gk. (lA.A.opAH 'spring',
Lat. salw 'leap' (!\1 . 1.99, F. 1.76, W-H 11.468, Pedersen
1900 :83). Mayrhofer favors connection with a root
*skd- seen in Genn. beschalen 'cover, horse (a mare)'

("'41. Skt. sasyrim 'blCs, grains', Gaul. sastam (transmitted as

aswm) 'seigle', \Velsh hazdd'hordcum' (P. 880, L:. 332,
W -H 1. 72, Kurytowicz 1956: 191). Agricultural tenns
of thi s sort, especially if restricted to geographically
distant areas of the IE domain, are far from secure.
 lndo-Europran /a! 67

*42 Skt. iNn~! 'q ui dcssi'x:he; sCchcrcsse', Gk. aJo~ 'dry'

(U. 31 7, F. I.189, Kul)·!owicz 1956: 191). These words
arc clearly re lated, but cannot be used as evidence for
/ a/ because as Frisk says the Skt. form, though for-
mally ident ical, is to be considered a late verbal noun
to iUyatz: it appears late, a nd d iffers from the Gk. form
in meaning.

3) The third category consists of words which, with a few

(X>SSible exceptions, can not be used a~ evidence of IE /a/,
though t hey have been in t he past.

tl. Skt. bhadrdf; ·fortunate, blessed ', Goth. batzza 'better'

(P. 106, M. II.467, Pedersen 1900:82 ). Though the
etymology of the Skt. word is not certain, it docs seem
(wi th :'vfay rhofer) to be connected with bhdndau 'is
praised', and hence stems from an earlier *bhiJ.drO·.

t 2. Skt. gMhyal; 'was m an gerne festhlilt ' to a root gadh-,

OFris. gadw 'vereinigen', Gem1. Catte 'husband ', OC:S
godl'J ' Zeit', Toch. kiitk- 'sich crfreuen' (P. 423-424, .\1.
1.320-321 , GUnter! 19 16: 8). These words are sore-
mote semantically and fonnally that they a re not
likely even to be rel ated. But if they arc, a root *ghedh-1
*ghodh- will account for them.

t3. Skt . g6.hvaraf; ' deep', Gk. f1flooo: 'wooded g len ' ( P. 465,
M. 1. 332, F. 1. 234, Pedersen 1900:77, Schwyzer 1932:
193 - 203, Szemer€nyi 1960: 2 11 - 2 16). The vowels do
not agree, and if the equation is to be ke pt , then Gk.
{1o:fJUs 'deep' must be brought in. But it, because of
f3affos 'depth' and {k.v&os 'depth', points unequivocall y
to an earlier *bmdh- !*bndh- .

t4. Skt . fakn0/1 'be a ble', Germ . behagen 'suit , please' (P.
522, U. 30 1, Pedersen 1900 :82). Aga in the semantic
link is weak, and if made, does not exclude a rcot
*kek-!*iok-.
68 h1do- Eurapean fa/
tS. Skt. kakUbh-, ka/;iui- 'peak, swnmit', Lat. cacUmen 'sum-
mit' (M. 1.135, W-H I.127 , E-M 81, Pedersen 1900:
77, Kurylowicz 1956:190). lbough this equation is
quite frequently made, it is impossible. As Kurytowicz
(1956: 192) points out, the words arc quite differently
formed: the lndic suffix is secondary, the Lat. primary.
And within lndic it is difficult to ignore the words
kiikUt 'palate' and krikul} 'cry of lamentation', both of
which may be related, if we can assume kakUdmant to
have passed through a semantic development: 'having
voice'> 'head'> 'summit'.
t6. Skt. khO.laiJ, 'threshing-floor', Arm . kal 'id.' (Pedersen
1900:82). But Mayrhofer (1.305 ) finds the Skt. form
' nicht sicher erkl3rt', and Thieme (1955:439) feels
that it is the 'vernacular counterfeit of the educated
form khara'.

t7. Skt. k.haliuam ' bit of a bridle', Gk. xo:Atv6s 'bridle'

(Giintert \916:8). But the Skt . word is clearly a loan
from Gk. (M. 1.306).
tS. Skt. kapUcchalam 'tuft of hair on the hind part of the
head', Lat. caput 'head' (M. 1.156, W-H l.163-164,
Kurylowicz 1956: 190). But the Skt. fonn is not to be
analyzed kaput-fala, but as ka-puccha-la with the pejora-
tive prefix kn- (Mayrhofer, Kurylowicz).
t9. Skt. kapil} 'monkey ', cf. Gk. IC'ij'ITos 'monkey' (Pedersen
1900:77 ). Though it is likely enough that the words
are related, it is further likely that Gk. Kfjflos and
Hebrew qiiph are also related (M. l.156, F. 1.836), and
that therefore the word is borrowed into both Greek
and Sanskrit and is not oflE date, a supjXlsition ren-
dered the more likely by the nature of the animal
designated.

tto. Skt. kapib 'olibanum', Gk. KO:'ITv6s 'smoke' (P. 596- 597,
Pedersen 1900:77). Mayrhofer (U56) feels that the
numerous by-forms in Sanskrit plus the fact that the
 lndo-Europran /a/ 69
word is not attested in literature is sufficient almost to
exclude tht: possibility of IE origin. Frisk (1.781~782)
does not mention the Skt. word.
t tl. Skt. kapanii 'caterpillar', Gk. Kiip.7T1j 'id.' ( Pedersen
1900:77, M. 1.154, F. 1. 774). The equation is possible,
but only on the assumption of an earlier "kemp- *kmp,
with the Gk. form remodeled after Kixp.mw.
ti2. Skt. fasman ' invocation', Lat. carmro 'song' (W-H
I.1 69 ~ 170, Pedersen \900:82). Pedersen discarded
this equation, proposed by others, because L1.t. -sm-
does not pass to -rm-. See above *5.
t\3. Skt. kvathati ' boils', Lat. ciisrus 'cheese' (P. 627, M.
1.283, W-H l.176- 177). The forms are too unlike
phonologically to be compared and to provide evi-
dence for an IE original.
t t4 . Skt. klwral; 'hard, rough', Gk. Klipxapo<; , Kapxa.\ws
'sawlike', 'rough' (M. 1.302, F. 1. 796, Kmytowicz
1956: \89). Kurytowicz holds that this equation is to
be discarded because Skt. kh - cannot correspond to
Gk.x.
t i S. NPcrs. lab 'lip', Lat. labrum 'id.' ( Pedersen 1900:83).
Pedersen also poinlS out that /e/-fonns of this word
occur, as in Germ. Lippe.
tt6. Skt. marUJ; 'wildernes.~', Lat. marr' sea' (M. 11.59 1~592,
W-H 11.38-39, E-M 387 , Pedersen 1900:82). Peder-
sen feels that a closer agreement in stem -form is de-
manded for words so semantically distant. E-~ deny
that there is a Skt. cogna te to Lat. mau.
ti7. Skt. pajrti~J 'solid', Gk. m)yi'Vp.t 'stick in, flx in' (M .
II .186, F. 11.525-526, Pedersen 1900:82). Frisk docs
not even mention the Skt. form, and MayrhofCr indi-
cates that the meaning of the Skt. word is not certain.
tiS. Skt. pastiyam 'dwelling', Ann. hast ·fixed', OJ\" Jastr,
OEfast ( M. H.242, Kurylowicz 1956: 19 1). Again the
70 lndo-Europtan /a/
meaning of the Skt. form is not completely certain ,
and if assured, th en, as Kurylowicz says, the seman tic
link between the Skt. and the European forms is weak.

t19. Skt. rddati 'bites', Lat. riido 'scrape' (YI. lll. 39- 40,
W -H 11.41 5, E-M 563, Pedersen 1900:83). T hough
it is likely that the words are related , they cannot be
direct ly equated phonologically.
t20. Skt. ris.)'ati, ri,fal! 'be hurt, injured', Gk. /xxiw 'break,
smash' (.\1 . III .62 , B. 833, Kurylowicz 1956: 191, 192).
Mayrhofer does not even mention this equation, a nd
Kurylowicz says it is suspect because of the lack of pro-
thesis in Greek.
t2 1. Skt. sabardhrik 'Neume!k' , O HG saf'sap', or Gk. iicpap
'straightway, forthwith' (L". 328, F. 1.194, Pedersen
1900 :83). There is no reason to consider this equation.
t 22. Skt. sra;- 'chain, wreath', Lat.jragum 'strawberry'. This
comparison was discarded by Pedersen (1900 :82) .
t23. Skt. vagnUI} 'call, cry', Lat. viigirt 'cry, squall' (M .
Ill.l 23, F. 11.5\3, 1. 646- 647, W-H 11.725 - 726, E-M
711 ). The discrepancy in quantit y excludes this com -
parison, a comparison which would, if correct, be a
case of onomatopoeia anyway.
t24. Skt. vafyulaf:t 'calamus rotang', Lat . vagtiri 'stroll about '
(M . 111.1 28, W-H 11.726- 727, E- \1 711, K urylowicz
1956: 189).
t25. Skt. ap1-utitatz ' understand', Lat. vtitis 'seer' (:'vl . 111.1 32,
W-H 11.7 37 - 738, E-M 7 15 , Pedersen 1900: 77) . The
Lat. word is almost certainly relatL-'ci t o \A/estern Euro-
pean words li ke Olr.jtillh 'poet ', and not to apzvritat1

t26. Skt.)'litatl 'attach', Gk. tlJ1'iw 'seck' (:\1. III.S, F. 1. 6 13,

Pedersen 1900: 77). But t lJ7iW derives from a root *did-
seen in the reduplicated SitYJf.l.t 'seek Out'.
 /ndo-Europtan / a / 71
Inevitably there will be dispute about my classification, and
indeed abo ut a n umber of cases included in the text. Some will
feel that 1 have been too generous in admitting cognates,
others that I have been too ready to excl ude possible imtances.
I m yselffecl that of the cognate pairs included in 2) the fo llow-
ing have a fair chance of actually being related: *15 dnw-
'burn', *20 kamp- 'bend', *21 kand- 'shine', *24 lwpr- ' penis',
*30kratus, 'power', *42 sausos ' dry', though here the forms may
not be equatable directly on the phonologicallcvcl. The fo l-
lowing seem most unlikely : *1 2 dhaWJJ- 'fir', * 16ghabh- 'hold',
*25 *kap- 'take', *32 ksat- '?', these for primarily semantic
reasons; *1 3 bh( r)an- 'break', *26 kap- 'hoof' , these for phono-
logical reasons. It docs not seem possible that any of the cases
in 3) can be seriously considered.
Whether or not one is inclined to accept a ny oft he cases in
2) as reliable cognates, it is interesting that a lmost all conform
to my rules, so that they a t least cannot invalidate my hypothe-
sis. Only *23 kap-, *26 kap-, *32 bat-, *4 1 sasi- 'grain' fail to
conform, and we have seen t hat *26 and *32 are not seriously
to be considered . Only *23 kap - 'head' and *4 1 sas1- 'grain '
remain. I think it wou ld not be unfair to conclude that no cases
in 2) or 3) are of sufficient weight or importance to invalidate
my assumption that unaccented • / a />[;}].
 Notes

'For the history of the I E vocahc system, cf. PcdeJWn (1931 :24(J.-3 10),
Szc merO:nyi( l9 G4:2 - 6. 1967:67-69),Wyatt( I964:141-IH).andforthch.,tory
oflaryn~alanalyscsofthat systern. Polom e (l 96!">:9- 78)
2 1 herefollowdcSaussure (1922:168-- 169)and Kuryl:owicz (l9S6:201 - 208)
in regardingonlyGk. /a/ as theregu larreflexofiE *[ J ). Forafuller(but sti ll
madequate)discussio n oftheproblemseeb.o: lowfn. I J.
3 Tht,tabled~s notmclud., a llth eposs!blesources of /e/ and/o/ mthe
daughterlanguages,but issufficiemfortheargumentherepresemcd Forthe
rcrord, though, laryngcalists hold that /e/ can derive from both */e/ and
*/H 1e/, whlie / o l ean come from */o/, */Hse/, */ H 1u/, and • ; H3o/
"Burrow objected to[J] onothcrground\aswell( 1949:28-29 ). He found it
difficult toirnagine th atEuropean [a[and l · l[ i]. phonc ticallyd i,tantsoun ds,
shou ld both hav~ dcnved from [;.[;and that [ ~]> [a] ondependently m a ll
European languages. Uut 1f, as I shall propose,[;>] develops from IE */a/, both
the~e objcrtionsa re answen:d,a ndwe have onlyto e xplam (aswe• hall below)
flo.,·l -l/il <la/.Hisobjectionsdohaveforcc, though.whenapplied to laryn-
geal OOllS(Jnants, for it tshardto seo: how ; H ; could develop to I ii in ln d<>-
lraman, butto /a/ mdependcmlyeverywh{"rt clse.
~ \\' Cowgil l, to whom I am indebted for cntiosm of an earlier draft of th !S
paper, cnticism which has resulted m several major changes of position, objects
to thiS lme of argument. He feels that, aside from nouns with no derivative
form ationsentat li ngo -grade {hke*ikwru "hor..,")a ndverbs like* tJ · 'tobc'which
makenof>"rfecttense forrnsorcausauves,therearenocasa ofnona!X'phonic
0 / e/ m l nd<>-European. And that further, Martinet { 1953 : 233- 267) has m fact

inferred prec1sely from the fact that nonapophonic [oJ is rare in Indo-European
save tnitiall y,thatiniual[o]istobcanalyzedas•/H.'!<'/ -,though cleariy•/H 1o/
and • ;J-130/ a re a lso possibiliues. With Cowgil l's statement about */e/ I sha ll
have toagree,w>ththeprovisoonlythatlthin k there m ust haveb<:encasesof
nonafXJphonic 0 /e / at one time in the h1.1tory of Indo-European Any root
onginally contaming nonapophoni c 0 / e/, when rt appeared in a mo'l'hological
category wh>eh reqmred */o/. was drawn into the *lei- */o/ apophonic
rdation,and hcnce ipsofactocame tocontainafXJphonic 0 / e/. About 0 / o/
things a re n ot so simple, and I cannot dtrcc!ly aruwn Cowg,IJ's objection. But

72
 Indo- European /a/ 73
to anticipa.t~ my rules co n""rning I E */a/, a nd the speculation m Append1x I,
I give (provi•ionally) t he following rull" wh1ch it seems to me wd l handk IF. • / o/
•;ol>lo/when accentcd

/ : / un=ntedaftcr{~j
~elsewhere
Ths ruk applies only to nonapophonic 0 / o/, because clearly apophonic */ o/
dot.; not pa«• to~ when unaccented It operata~ to give Skt. <14• 'C)'(''. Gk. li01n
'eyt'S(doal)', butSkt. IA.)ate' sees'(< • HJk"'<-;d. .\1. 1.9.'>)
~ In fact the idea goes back farther than that, b<U phrased in un;u;o:ptable
focm in terms of a thrce·vowd (Ia i u/) , ablaut-gov~rm:d oyste m. Thus
Schleicher (1866: 18) give<~ as the vocalreih~ for a
Schwachuog gr undvocal 1.steigerung 2. •teigerung
i,u,i,U
He restricted t hi sty pe ofweakeningtoposition before r Examples 1861i:21ff
7 HUbschmann ( 188:.:J)feltthat•ag6shouldgive Skt. •y.iml
~ Buck ( 1896 ;285-286 ) objected tothe numberofaorist presents r"<j uir..d by
Becbtel 'sbypotbesis,ForBocbtclwouldhavetoassume that : •ago < •ag8< •agJ,
and al"' to the-/,. participles bhak~ from Ma;- {below 11.4), matllih from mad·
(below II 25), and also aft (below 1. 16), all of which words should by Bechtd 's
rulc:ol show *Iii in the initial syllable in Sanskrit. And Brugmann (1897:173)
(Xli111< to Jthillh = 0:11ita•~ ru proof that 1 ~ 1 could appear under t ht accl" nt , an
argume nt which we shall ha"" to con.<idM· below. Buck's fin;t objecnon, though
legitimate and telling against Rechtd 's formulation of the rule, will prove to
have no bearing on mine. His second dOCll at finot glance seem difficult, b ut th{'
vowel appears in a closed ~)'liable in hhnk.tO.IJ and matl<i(>., a pOI'ution wh1c.h pre-
ventedthepassab<eof / a / to / i/ ;andbothformshavebcsidethcmpl"eS('nts"'llh
th ~ •ame gradc:ofthe root,andthisgrade mayhavebeenanal<>g~c.allyre;to,-,d
to the -/6. participle Th~y would, th;;n, OC no mo,-, unusual than ,pa.fl<ib rnpai
'see'{< 0 (J)pd;W..) orGk. r.(mr)\- tomoow ' cook' ( < "ptk"'I6J ), Skr pal<~ About
ajl ..eshall&ttbr:low.
Y"1b.,..,formsar..taken primarily from Pokorny (777-778 ). Ann amo~ ··~wen '
and Mac-edonian ix{Jpo- zusammcnziehend", Gk. ix{J;:,.pV·hpiymxw ( ll es.) ar..
omitted a• being too uncertain
lOCf. above note 3 for 0 /o/ >/ :/. Again t he proviiKl that the "owd heac-
cent!rl must be added, for IE "yO.< 'who (rei )' appean; as Gk. lif and Skt. """-
not Skt. •0
II \Ve could omit one step in this de\T. lopment, perhaps thus increasmg Its
plausibi1it)·, by assummgthat•n,,. pa><SedJ~rect ly to * amQ· :-;"o otherwords nl
G~k would exclude such a devdopment, but the family of <i""p Itself seems to
stand in th;; way For if we assume •nro- > •..,,,., we dc pnve ourseh·" ' of an y
possibility of explaining the i rregular Homeric scansion of ix....Spori1m ( <-~ - i! . =
•O:pori)m?)
 74 Indo -European / a/
t2Butofcours<'ifoncw1She.< to,..,stnetthcapplicationofth~ rule topo61tion
bt:fore semivowds, ~1.1ch ts certainly ~oh],. Doing so would at one blown-move
allthcexceptionstothcrukg•veninthctcxt,andthen<"wrulet·ouldbcstated

1-1 ini tial unaccen ted /a/ > fi __ [~l /a/ d~wherc· ayi. a;fih ('-1 6), amici~
(l i \),arg hal;(l.l3),an d mril;( l.i 9)w•ll nowsurvivea ss uchbr:cau""notbdo...,
semivowel I do not favor this new ruk for two reasons . Fint, 11 introduces a
<--.:nam compkxny mto an otherw~ s traightfon.'llrd rule , since now surround -
•ngphoneme.aswellasaccenta...,r<-lcvant Thisisprobablynotmuchofa con-
.io:kra tion, but more Important, rtal) (to ani/z and ciram) s=m• to indicat e that
unaccented /a/ regularly disappeared mitially bt:fore /T/ as well. Tlusfact
mdicates that the d dininnn of semivowel would have t o hi: extl'nd" d to mclude
/r/ if the new fonn ulat wn 1! to hi: ret ained, but would still exclude / rn/ ThL'l is
nntimpossible, hut evenatthatauihandalghtilJ,arestillexceptions,andnarrow-
mgtheapplicationoftherule<mlytoaccoumfortwoexcepuonsdoesnot..,.,m
worth it 1 favor the broadest possibleforrnulationoftherulc, but would not
argueUXJst renuowly aga.tn~t the narrower one
Perhaps at tlus point li)'Uh, ~-~ (L23), and ii.vih (L 24) can be reintrodu ced and
r~habditated with IE * /a / - The Av~. genit ive of<ly!l 'U~ n.•dauer' i•yao.i, "lth-
out 101Ual vowel h m ay well be that this a blaut relauon pomts to a n origmal
1-1 par"d1gm *ay6us, which became •ay,, •yO..., as the r"~'ult of the I=
ofinnml •/a /. At adatcpostcnortothlSStage,but•till w1thinthe
1-l f>"riod,the a"""ntwasremovedto theendmg inthenonunauve aswell .and
the vowel lengthmed (vrdhhled) inordcrtoavo,dthelooo;of•/a/-, now un-
accented. ·me long /a:/ - was then generahzed in Sansknt to all fonns and
, derivative'! of t his word, Ave. alone prcservmg th e origi nal ~cni tive. \Vc must
then unagme a sim1 lar . though unexpla med, ablaut relation in Skt iidh. This
"prophylactic kngthcn ing"which . whcninvolving apophonic/o/, t ravelsunder
the namcof Brugmann's Law, can apply, then, to • !a/ as well a< to • /o/
13 Conneetion ~n-·reen a Skt . word and LaL ~<Uca can still be maiotained if
we assume that Lat. !.,.em replaces ~arher Lat . •,iica, with geminatiOn replacing
length (cf. lil!•a - ilUrm), and that Skt. viihW. contains the sam(' word suffixed
with - lti. We shall see OClowthat internal IE */ a/ passes to 1-1
aca:nted H ence on this assumption the IE form will have
lOffil preserved m Lat in but extended by -tci in Sanskm Skt
JXI"'l>bly then be a differ('nt remodehn~ of t he original form
HJ{nt laia- (Friedrich 1952:188)1SSOmenrncsbrough t intoconS>derat!on
here, and would of eom~ suppon my argument by ~howing that there was an
Ea,1em IE word meanmg 'hare' (<•.w_,,_ ')not connected w1t.h Western •{as -
Rut the •neaningofthe H lll word is not "'-"Curdy established, and hence iaia
mnnot ~used FnNlrich lat er (1937: I B) r<"<X>rds l.arochc'scompa rioon of this
-.urd w1th the ideogram 1JDL~.KL'R RA in the mcanmg 'antelope
1 ~ 1 admi t here only t h""" cases which •how a correspo ndence Eur /a /

Skt. /i/,andthcrcforcdo not include:

 lndo~Eurapran /a/ 75
Lat datus =
= Gk . &-TO• Skt d•tah 'g.ven '
l..at.fadus = Gk {tmX = Skt. h11ah "pi~"

because the Gk.. form docs not contam /a/.! want, mother word.s , to avmd m
rhe textthequestwnofthe "t nple reflex ofs~hwa m Greek" (Lejeum• l9 55:164)
Titerc are m fact only two poss ibk explananom for th11 tnple rdlex, nell her one
reall y affectmg my arg ument at thts pomt. Either Greek P'"-''""""" the anginal
vucalirntand Latin and Sanskrit experienced a mergerofunaccented/e / and
/o/ with unaccented / a / when in di.yllabl"" between stop consonants, or Greek
has mnovated rn bringing the timber of t he vowel of the strong grade over into
the weak grade : on this as.sumpuon the original IE forms were *dat6• and 'dhatO'
J fee l th atinfact G r..-.:k ha!innovated,butinsodoonghasrccapturcdthcorigi-
na l form. That is to •ay I fed that the de,·elopme nt of'placed" was. *'[rlhet""] >
'[ dh;,tos] = / dhatos/ > Skt hllu/J.. Lat. fa<tu.r ( Fo r th~ notatoun ami presup-
positions~ Appendtx I) T hu' I get the best of both worlds, but for the ume
being at least prefer to exclude cases of this son from conSJderatJUn of I E */ a/
16 Brosman's e xplanauon of the ongin of -1 m H m. may of coune not be cor-

rect .; ;;was added to /r 1 and / n/ '"Indo-Eu ropean mother categories as

wel l. I n Hitt . Jtse lf -/i/ was added to · / r / in the m tdd le endmg ·lan, asoppo:o;ed
to Lat. ·lur And m Vedte Sanskrit the neuter smgular of no11ns m · It " · laTI
(Lanman 1880:421-423 ). There JS alw the altcrnauon m the loca!J w"ngular
in VediC betw~-cn r.i;a•l and r~;am. It may be that the Hnt · • tn t he ne uter pl wal
has the same ong.n. T he tmportant tlung, though, is that. even tf thi$ · It! be
transported as an allumorph of the ne uter plural back mto Indo-European, n
is restricted to,. and n·stems, and ll snll rcmaJns the case that - /a/ was th e
vowel added to~l-stcms.
11 Jl.enveni.te (1%2: 111 - 11 2) deni es connection of Ht \1 mtHo· wnh the I E

forms m m•g· because the -U·, accordmg to Sturtevant\ ruk, pomts rather to
IE* /k/ thant o *if(l , and because t hemeamng 'viel,sc hr'istoodistalllfrom
"gru.<;s' He com pares the H itt . word wtth Toch. maA "vJel' But ncJ thcr of these
objecuonsisenoughtodesrroythe connN1lon( C\1ayrhofer l964 194 I%), for
the semantic difficulty ts slight indeed. and there are oth~r cxcrptJUns to Stone·
vant'<rulesuchas"qqa"ego'
lSTwuothe rforms haveoL-.:as tonal lybcenconslderedtucon tamu naccemed
1- 1
a) Sin kUha, Ave. hd.i, OCS kM. Osc. puj(cf Lat. al< ·<l<b<) <
•J:udha ( P 647,
M 1.249, W -H II 739, E -Y1 716)
h) Skt <iml>a'mother (voc.)' . Gk. voc.atives hke P~l'¢a "matden" (Wac kernagel
1896:6, Peden~en J900:78 - 79,Guntcrt 1916.10- ll)

a) The ·ham the Skt. form wo uld seem to be the same · hll as m dui, and hence
76 In do-European / a/
form to bean, one mav wdl conclud~ that the IE form was *kPidhr, not kUdha
(soP.). b) Pede rsen introduced limb.. m support of bis ~untemion that IE final
•;a!> fa/ inlndo-lranian.Andthough itislikclythatSkt.voc/e /ofa-stcms
derives from earl ier •;aJ/ (= /a/+ /i/), ciml>acannot he used as evidena:for
voc -/ a/ equatable with Gk. voc. -/a/ . it is q uite dearly a Lallwort, as Gtinten

I9Jn suppon ofnonphonological conditioning d. K iparsky ( 1967: 128): 'Non-

phoneticenvororunents insoundchange,andthecorollaryofphoneticdoublets
m the C"-"" of ambiguity, deserv"' more ancmiun than they usually get. ·10th._.
,.xtentthat theyaredeah wtth at all in rhen eogrammarian-structuralisttradi-
tion ofhistoncalling mstics, 11 IS in te!TT1s ofcon.,.,ptssuch ""'"restressing"or
"analugy"whichar~ inadequatetothepurposc,asshouldbeapparentfrumthoo
abo"" e xampl es'. Professor Kiparsky kindl y refers me to a massively documented
ca'le in wluch the sam~ phoneme de.,.,[oped in two ways (Li ndgren !%:!)_In
Middle lligh German final ·• diSappeared in c e rtain ca= eady on, while it was
r =rvcdforsometimetnothe rs:e.g.inthrnoun-,rernaonedasamarl<of
g«nder and number distinct ions, b ut dts.appeared as a ca'ie marl<~r (Lmdgo-en
! 9.'>3 -211 - 2~5) Since there is no phoncmK di st mction, the only c"Omhtwni11g
mustha.,., beenfunctional,i .e.morphological.Healsorefe,.,metothr caseof
final -11 in Estono an. In Esto noan Balto-Finrnc final -n i~ normall y lost, e.g. gen
sg. •t.m>OOn 'base' (Finn . kanna~) > Est. ka11na. But in tb~ first p~.TSOn singular-~
hasremained *kanilll n'[ carry'(Fin n.kallllan) > F.st.Aannnn . ThcfollowLngothcr
=••" from t h e da&IJCall anguages which m ay he instances o f the same phenom-
Citon occur to m e. I) •- /ewa/ develofl' in two ways in Attic G rttk d~pending
on whe ther itistheendltlgofthen e ut erp lural ofa no.,norofanadjtttivc
*/.lstrwa/ 'cities' > lio'Tl), but */ hcdewa / 's.....,.,t ' > i)lil<~. 2) Though the
spintusas peris regularlyp reservedinLocrian (andothcrforrnsof)G,-,.,k,itJS
lost in tOC ddinite anid<' ( Buck 1955: 5~54). 3 ) IE • / t y / develops in two ways
once rtainforonsofGreo:l<dependmgonwhe therth<'*/y / formedpartofadcriv-
ative suffix based on Gk. rnat<'rial, or wh.-ther the form was unmotivated from
the Gk. point of vi e w (Lej,.une 1955:87 - 88, Buck 1%5:69-71). 4) I E */ o: /
<k\'elops in Umbrian to / u / in the ablati"" singular of o-stems and in impcra
loves f<*lo :d / ), but to / o/ in tlx: ge niti•-e plural of o-s tcms (<* l o:rn/) and
<'lsewbe..., ( Poultne y 1959:36--37). Doubtle"'! other explanations c.an be eon
ocivcd of to a~'Wunt for these divergent developrncnts-aa-.:nt orr I ) and 2),
cbronologocal layers m 3), phonetic rnvoronm.,nt m 1}---, but grammatical
interfe,-,nceSttmsat pre s-:ntthebestone.
W]bc -/ a/ in thes<' forms, then, may well have had tb~ same funcnon as the
-/a/ inGk. interjtttio ns (olrro,f'TTO) a ndnamesof lettersofthcalplra\x:t
(lii\<f><,t, f3f]m) for r/ 1 pst.' and al'f'~, bit respectively (Schwyzer 193 1) _ In all th=-
c.ases th e addotion of -/a/ was the alt ernath-e to phonotactic lOb-S of final

21Sti ll anotherpossibtlttyofan explan at ionls thatatthetimcthatu naccmtcd

 lndo~European /a/ 77
-/a/ was passing to[::> ], the endings of the perkn wtrt accented and henc~ un -
affected, In this event the perlec t paradtgm would hav~ been •• wtd<i ••wti>IM
•• wtdi, whieh later, with the introduction of /o/ became PIE •wOtda •weustlw.
*wOtdt. On this pos~~ibilit y a nd the throreucal framework underlying it, see below
Appendtxll,andalso,withadifferentthroreticalbase,Sehmitt- Brandt(l96i.
124- 125)
22Cascsinvolvi ngvowelsothe r than/a/incl ude
/o/ Skt linikam •face, front' < •tm + ok"'· . Gk. ivwmj 'face" (~1 . \. 34)
Skt. abhikt 'immediately before' < •aMn + ok'"- {M 1.42)
Skt. pr<itil.o.m 'su rface' < •prott + ok"'-, Gk. .:p6ow~ro~ 'face' (M H. 361)
/e/ Skt.apicyah'secrt"t,hidde n'< •rpt+mA"'·(M.I.39).
2J Some such devdopmcnt together with thiS resulting sy!lem is customanly
a!ll!l.lmed. The difference between my view an d older vievo--s is tha t I do not assume
that unaccented long ,-owds appear as scl1wa, an assumption sha~d by most
laryngealists and the more tradit tonal Indo-Europo::anists. Hut 1\ is clear that
unaccented long vowels are not reduced (d. Gk. ~~;,, 'sw~·. Skt. >OOdW) 'id.'),
and it would be mos t surprising (theo~tically) tf they were Again one may
compare Latin, a language in wh ieh internal short vowels were short ened or
raised (as ttl ammus < •antmw or •anamw), but in whtch long vowels were t.Hl·
affected(cl.amtirw)
zt I thus agTCI' with Schmitt-Brandt {1966, 1%7:8-3 1), and also Wtth Skt
grammarian,, in positing d u: so-called red uced grade of the root as th~ basK
fonn ofthe root. at least in tho&e roots whichcontamed a semivowel I canno t,
however, subscrtbe to Schmiu-B rand t"s specu lauons concemmg sti ll earhcr
stage~ of indo-E uropean
2~ 1n thts way I hope to forestall the plausible obj ection to my theoryrrusc d

by Cowgill. He finds it difficult to accept that unaccented / a /> [ ~]> 1-1 Iii.
whi le/e/remat ns/ e/whetheraccent edor not.h._. fe<:lst hatif/a/ > ;i/, then
/e/ should b... affected aswd l The answer 1S to be sought in relatJvcchronology
Atthetime that /ai> J;,], /e/> ll,butafter th._.ac""m shift (bothinsyll abl._.
accented a nd in type of accent) thcrt" was no funhcr sy ncope of shon vowels
26Mypositioninvolvcsmeincertainpredicuom,andthoo;ghlltstheca.se
that their nonfulfi llment wtll not invalidate the theory, their fulfillm ent will
provideameas ureofsupportforit. L'smg /l/to representboth/i/and/u/.
/e/ to represent both;._.; and/o/,weshouldfindthefollowingrestncuonson
the ocru rrence of stem shapes. ;e; should occur anywhe~, while IV (unac-
ccnted/e/ ) should ocruronlyi nt hosestcmswhiehalsocomain/0 siglaOCi-
orCiO-,forevery ;e; isafonner /O,excepti n thecaseofanalogtcally intro-
duced ;e; as in thematic present endmgs. GrCi- must d enve from CIC- , wh!l e
CiO- d en vesfrom CCi- the accented vCM·el is the mo~ recently mtrodo;cedone
If/ should not occur in ablautingparadigms,sinc... ;I; wasalway:sreplacedby
/6/ ~or should /ft; occ ur,sav~inthe firsts y ll ableofthetmperfect iveofa root
78 fndQ- Europfan / a/

which had at one time app<"".ared as CaCi-. as 111 the thema\lc (second ) aorist
Otherwise onginal / a / appeared as either CiCO. - < CCJ - or a• ClaC. (>Ca: C-
or CQ:C-, depending on wheth<:r It::/ was a front vowdor a back vowel) <CtiC-;
or as CCia-, 1f from COi-. T hat th.,., predictions m fact are frequ..,ntly reahzed
i licorroOOrativeofmyv•~.but•trnustinfairn=beadmittcdthattheolder
thrones of ablaut make manr of tM sam., predictions
 References

BARTHOLO'>t A ~, CHRJS nAN

1895 Vorgesdu cht~ d~ r 1rani'i<:hen Sprachen. Grundriss der Imm.>eh<:n
Philolog•c Hrsg. W Getger & E Kuhn Vol I l-1~ 1 Stra."burg

BECHTEL, f'Rl1Z
1892 Ore H auptproblcmt· dcr indogcrmaruschen La ut lehr~ s~-it &hlerdtcr
Gotting.,-n

llENVIC'<JSrE, f:MJL>:
1962 H ittite et l ndo-Europe.,n f:tudes ccmparauvcs. Pans.

BrRwt, RoBF.RT
19'>6 Revtewof M . I I-XXV, l-128,mlF62:19 J-200

BROSMA.~, P 11uL W
1962 :-ieuter Plural rn -r among Hiuite COOilOilant Stems .J AOS 82 : 63-65

1904 Ku rze vergleichcnde Grammatik der indogermanisc hcn Sprache n

llerlin
1897 Grundriss der vcrgleJchcnden Grarnmatik de r mdo''""'""',j '""
Sprachen Vol I pt . I Eml eitung und La utlehn: Strassbur~
19 11 OpaL Vol 11 pt. 2 Lehrevonden Wortformen und1hremC~brauch
2dedStrassbu rg

B~.;cK,C.-.RL 0.-.Ru"G
lll96 Some Ge nl':ral P robl~rns of Ablaut Amencan Journal of Ph1 lology
17:267 - 21l8
1 95~> T he G reek D1al ~= Chtcago
80 Indo-European /a/
BLRROW, T
Loanwords m Samkrit Transacuons of t he Pln lologlcal Sooe ry 1- 30
194'l ·'Schwa" m Sanskrit Transacuonsofrh.- Ph ilological Society 22 -6 1
1955 The Sanskrit Language Lundun

CA~I>O""· GEoRr.F.
1961 Greek klimnO and ttimnii. Language 36:502-507.

l'kl £DR!CII, jOllA.~S£8

t9J2 Hethaisches WOrt~rhuch. Heidelberg
1957 Hethnioch"" Worterbuch I Ergiintungshcft Hcidd bcrg.

FKrSK, 1-l )"LMAR

1934 Zur lndoiranrsciK.-n und Griechischen l\'omiualbildung GOt.,borgs
K u ngl. Vetenskap5- och Vitt erhets-Samhiilles l-l andlingar, VA 4 nr. 4
COte borg
1938 ln dogennanica ( "'Goteborgs H Ogs kolas Arssknft 44 rH. 1). GOtcborg

G o.:,.;TER r, l·l uMM'S

191 6 l ndogerm a ni :;.cheAbl autprohleme. Strassburg.

HrRT,I-I ERM ANN

1900 Derindogermanische Ablaut. Stra111bnrg.

1928 lndugcrmanische Grammatik. Vol. IV: Doppd nng, Z usammenset·
zung,Ve:rbum H eidelberg.

HVBscHMAI<N, ll m"RlcH
188:"1 Do> mdogt>mamrchr Vol.n/;plnn. Strassburg
1!100 Review of Hirt 1900, 1n If An~.eiger 11. 24 - 56.

KlPAI\ SIW , P AU L
1967 A Phonol ogical Rule of Greek. G lotta11: 109-1 31

KKAHE,HANS
1962 lndogerm anische SprachwL=nschaft. Vol. I. Berlin

K u RvLOwJcz, jER Z>"

l'l'l6 L'a pophoni e en indo-curop6 ::n. Wroc/aw.
19.'>8 L'accrmuation des langues indo-e urop&nnes. 2d ed. Wroclaw
J9.~8a New DL<;eov.:ries in I ndo-European Stodies. a Le Hittite. In Pro-
ceedin gs of the V lll Internationa l Congr<::SS of L inguists. Oslo
1%2 Probkm e der indogermanischcn Lautlehre. ll I-'achtagu ng fi.tr mdo-
germanische und allgemnn~ Sprachwisse nsc!oaft. J Knobloch, ed
{lnnsbrucker Be~trii.ge zu r Kulturwi=nschaft 15) lnnsbruck.
 Indo-European /a/ 81
LANMAN, CHA KLF.S R
1880 A Statistical Account of!'\oun -Jnllection in th~ Veda JAO S 10 : 32:,
60 L

LEJEUNE, MICHEL
19S5 Traitfd~phonftiquegrecque. 2ded. Paris.

l.rNUGRf.N,KAJ
19:i3 Die Apokope d~ MHD ·tin teinen vcrschieden cn Funktirmen ( = An-
nates AcademiaeScicnuarum Fennicae B 78:2). Hdsmk1

MARTINET,ANDR£
1953 :\"on-Apophonic 0-Vocalism in Indo- European \Vord 9: 253- 267
Reprinted in French trans. in f.conomi~ d e!< cha ngt'm<"rm phonenqu<"•
212 - 231 Bcrne,i9JJ

r.1AYRHOFF.K, M ANFR~D
1952 Das Gunuralproblem und das indogermani&ehe Wort fur '" !lase'"
Studicn zur indog~rmamschen Grundsprache 27-32 ll r.;g v W
Brandcnstcm (Arbeit~n aus dem lnstinn fiir a llg u vgl. Sprachw•o;.se n-
schaft 4). Graz
1964 R<"vie'o'>·oflknvenillte 1962,in DicSprachc-10 174-197

PAL\tER, L EO"ARD R
196 1 T he Laun Langttagt'. TIU rd lmpr css•on London

PEOER!il'N, HOI <JF.Il

Wieviel Laute .r; ;abesim l ndogermanischen. KZ 36.71--110
1905 Die nasalprilsenua undder sla ,ischc-abent.KZJ8:297 -4 21
1907 Die idg -semttische H ypothcse unddie idg Lau t1ehre IF22: 34 1-1f{>
1909 Vcr:;:leiche ndeGrat mnatikd erkdusch cu Sprachcn. \'ol.l Ein leitung
und Lautkhre Gotnng:~n
1926 La d nq uiO:me d k: linai.son latine (: fkt Kongdigc Danskc- Viden-
skaOCrnes Selskah Hmonsk-filologiske M edddd ser XI 5) f'.Of>"tl ·
ha~en
Lingmsuc Sc•ence in the :'\!meteenth Century Trans .J V\' ~pargo
Cam bridge.

PowMt , EooAR G. C
1965 The Laryngeal Theory So Far. A Critical Bibltographrcal Survq·
EVIdence for Laryngeals 9-78. W. Wmter , ed The Hague

PouL n< fv . .JAMF.' W

1959 fbe Bronze Tables of l guvtum ( =American Phtlologrcal Associat ion
M ono~raphs 18) Ba ltimore
82 Indo-European /a/
PL'HVFJ..,jAAN
1964 A MyC<"naean-Vedic utularcomcid.,nce_ KZ 79:1 - 7

RENOii, l.oms
1964 V&lique •adh·,AhO.d·et l<i.! · lndo- lramcal63-167 Milangespr~scntCs
i>.GeorgMorgtnsticrne.Wiesbadcn

SAL'l>St;RE, F ERDI"ANDD£
JU79 MCmoinc sur lc syst&ne primitif des voydles dans les langur--S indo-
europ&:nnes. Le ipztg. Cited by pagenu;nber in Saussure, 1922.
1922 Recueildespublicationsscicntifiqucs. Gcneva.

ScHLElCIIER,AUOl.."ST
1866 Comf>"ndium der vergleich=den grammatik der indogermanischen
Sprachen.2dcd.Wcimar

ScHMIDT,jOHASNI'.S
1889 Die Pluralb•ldungen der indogermanischen Neutra Weimar

$cHMl'IT- BilANDT, ROIIER·r

1966 Probkme de. indogennani,fficn Vokalismus Kratylos 11 : 16&--174
1967 Dte Entwicklung de8 indogermanischen Vokals~tems. Heidelber~

ScHWY~ER, EouAJto
1932 :\':eugriech. {Jlooo. (ChJO'< ), altgriech. {li)ooa und v~rwandtes . Rh~in­
ti!ChesMuseum88:193 - 203
Griechi.che lm~rJektioncn und griecllische Buchstabennam~n auf -a
KZ S8 :170- 203
Gricchtsch~ Grammatik. Vol. I Mum ch

STRil.lTBEkG, \V
19 15 Ferdinand d<' Saussurc lndogermanioch""jahrbuch 2:203-213.

Osw ... l. D J. L.
Sz~<MERE.~· vl ,
1952 n,,. Et)·mologyofG~rman Atkl. Word8 :42 SO.
1960 "l"h<: Ongin ofGn.-ek {la{)V<; and flM1llf"D~. GIQ{ta 38:211 - 216
1964 Structuralism and Substratum ~ indo-European~ and Aryans m th~
Ancient Near E.a!.t. Lm""" 13:1-29.
l961a Syncope m Greek a nd Indo-European and the :'oiaturc of Indo-Euro-
pean Acu:nt ( = lnstJtuto univcrsitario oncmale di :-:apoli, Quaderni
dellasezione li nguisticadegli ann ali 3). :'oiapks
1967 The ;>;"ew Look of Indo-European - Reconstruction and Typology.
Phonetica 17 : 6.'1 - 99
 Indo-European /a / 83
THJ£M.i:, PAL'L

J'))J Dw l le tm at d"r indo~ermamschen Gemeinsprachc ( = Abhandlun !o"'"

d~r Ge~Stes· und Sonalwr=nschafthchcn Klas>K', Akade rme der
Wll!Sen.chaftenunddtrLllcratur. 1 \:)J_~-.f>\3 MarJU
19)') R evww ofBllrrow 1956. Langua;;<' J 1 :428- 4J6

W ACKI\IlNA0FL, j AKOB
l!l88 M"~dlen zur gneduschen Grammatik· \4 _ho.oTo< KZ 29· 144- 1Jl
(=Kleine Schnften 647-654) Grittmgm
Altindischc Grammal!k. Vol I: l.autkhre Goningen
Op C!l Vol I I.\ Eio leitung zur Y...'ort lehre :"iommalkompu mron
Gottmgen
1954 Op m Vol 11.2 Die :-.lominal>ilffixe (by Albert L),brunner) Got-
ung._.n

\\-'v ,.,TT, WtLLIAMf. , JR.

1964 Struc!Ural Lrnguistics and the Laryngeal Theory Langua;::c 40.138--
1;2