1.The organization of the dictionary
This dictionary contains the lexical entries that can
be more or less reliably reconstructed for Proto-Celtic. It is intended to
contain Proto-Celtic words rather than roots, but in several cases, where
the word formation of cognates in the attested Celtic languages differs, a
rather speculative choice had to be made in order to decide on the Proto-Celtic
form. In some cases the OIr. form was projected to Proto-Celtic, but in many
instances the form with most parallels in other IE languages was postulated for
Proto-Celtic as well. Whenever the exact Proto-Celtic form is underspecified,
for one reason or another, this is clearly stated in the discussion following
In this dictionary, a Proto-Celtic
form is reconstructed whenever at least one of the following two conditions are
(1) Cognates are
attested in at least two primary branches of Celtic. By primary branches I
understand Goidelic (Irish, Scottish, and Manx), British (Welsh, Cornish, and
Breton), Continental Celtic (Gaulish and Lepontic), and Celtiberian. Whether
British was dialectally closer to Goidelic (the ‘Insular Celtic’ hypothesis) or
to Continental Celtic (the ‘P-Celtic’ hypothesis) was considered irrelevant in
deciding whether a given word was reconstructible for Proto-Celtic.
(2) Probable cognates of a word, attested in only one
branch of Celtic, exist in at least one other IE language.
PCelt. words are given as bare stems, e.g. the n-stem
*talamon ‘earth’ is adduced rather than the Nom. sg. *talamū. Where ablaut
patterns within paradigms of PCelt. nouns can be reconstructed, this was done
in the discussion of particular lemmas. If the etymologically related words
within Celtic do not agree in word-formation, the simpler form was usually
projected to Proto-Celtic. For example, PCelt. *barinā ‘rocky ground’ is reconstructed on the basis of
OIr. bairenn; it is assumed that the Brittonic forms (W brennigen, Bret.
and Co. brennik) represent derivatives thereof.
The meaning of Proto-Celtic
words is often rather difficult to reconstruct. Where meanings of cognates in
various Celtic languages do not agree, either all of the attested meanings were
projected to Proto-Celtic, or the meaning deemed most basic was reconstructed.
Whenever the meaning of a particular PCelt. word remained the same in one or
more of the attested languages, the meaning of the attested word was not
adduced in the fields containing these reflexes. For example, PCelt. *wiro-
‘man’ has reflexes with identical meanings in OIr. and MW, so the meanings of
OIr. fer and MW gwr were not adduced separately. The same
principle was followed in adducing the meanings of the PIE forms and their
reflexes: since the meaning of PIE *wiHro- ‘man’ was preserved in its reflexes (e.g.
Skt. vīrá-, Lith. výras, etc.), it was not adduced in the
field containing the attested forms in IE languages. The list of the attested
cognates of the Proto-Celtic lemmata is not meant to be exhaustive. For the
sake of conciseness, I usually adduced only cognates from two or three IE
branches, usually those that are most relevant for the PIE reconstruction, and
added a reference to Pokorny’s Dictionary (IEW), Lexikon der
indogermanischen Verben (LIV), and/or Encyclopedia of Indo-Eruopean
Culture (EIEC), where more detailed lists of cognates can be found.
Reflexes of the reconstructed
PCelt. forms were given from all of the attested Celtic languages. However,
since most of our knowledge about Gaulish, Celtiberian and Lepontic is derived
from onomastic analyses, cognates in these languages are sometimes adduced
although they are not established beyond reasonable doubt.
Every investigator of Celtic
etymology must make a principled choice: one can argue that, since Celtic is a
branch of Indo-European, it is a priori likely
that words in Celtic languages have Indo-European etymologies. If one accepts
this assumption, then finding any possible cognate in the IE lexicon is
preferable to not giving an etymology at all. On the other hand, one could
argue that we cannot possibly know the percentage of words that Celtic borrowed
from non-Indo-European languages, so that any Celtic etymon may be equally
likely to be inherited as it is to be borrowed from some unknown source. If
this is the case, then one needs more than possible cognates in other IE
languages in order to make an etymology plausible.
Let us take one example: OIr. ail
(phonologically [al']) ‘rock, cliff’ is a very short form, consisting of
only two segments. It could, in principle, represent a variety of Proto-Celtic
forms (*ali-, *fali-, *yali-), and these could go back to an even larger number
of possible PIE roots (*h2el-, *h2elH-, *ph2el-,
*pelH-, *(s)pel-, *ph2el-, *yel-, *yeh2l-, etc.). It is
obvious that, with such a short and isolated form, the possibility of finding
chance resemblances in other IE languages is considerable. Many linguists would
therefore consider any etymology of such a word hypothetical, and leave open
the possibility that it was borrowed from some non-IE language. On the other
hand, if one assumes that this word is much more likely to have been
inherited than borrowed from some unknown source, then finding a possible set
of cognates from the PIE root *pel- ‘rock’ (OHG feliza, etc.) is enough
to make a plausible etymology.
I am not sure which of these two
methodological principles one should adopt, but I thought it would not be fair
to the reader to be too critical with respect to possible, but uncertain
Indo-European etymologies of Celtic etymons. To do so would mean to limit
oneself to trivial and well-established etymologies, and my feeling is that
potential readers of this book do not expect it to contain just the information
that, e.g., OIr. athir is related to Lat. pater. Etymological
dictionaries are usually not best-sellers, but this does not mean that they
have to be boring. This means that many lemmata in this dictionary should be
understood as proposals to be evaluated, rather than as a collection of
well-established scientific facts.
However, it was my intention to
avoid too speculative etymologies, especially those that rely on alleged
reflexes of PCelt. words in only one, poorly attested Celtic language. For
example, the Gaul. month name ELEMBIU from the Coligny Calendar is
derived from the PIE word for ‘deer’ (PIE *h1eln-bho-
> Gr. élaphos). The Greek month name elaphobolēion,
derived from the same PIE word, is often adduced in support of this etymology.
However, I did not include it in my lexicon, since the meaning of ELEMBIU is
far from being assured, and there are no traces of this word in other Celtic
languages (but cf. PCelt. *elantī, a different formation arguably from the
same root). The form found in Coligny is actually compatible with many other
interpretations, and in order to relate it to PIE **h1eln-bho-
one would also have to explain the unexpected reflex of the syllabic nasal in
Gaulish (em instead of *am). I have also tried to avoid all ‘last
resort’ etymologies, which are often repeated in the handbooks simply because
there do not seem to be any better Indo-European etymologies of particular
words. A case in question is, e.g., OIr. dúil ‘creation’, which is
commonly derived from PIE *dhuh2li-, from the root *dhuh2-
‘smoke’ (Lat. fūmus, etc.). Now, although it is possible to imagine
a series of steps in semantic development that would lead from ‘smoke’ to
‘creation’, I find it difficult to believe this etymology: it seems to me that
accepting it would be a sign of desperation, rather than the result of a sound
consideration of probabilities.
In many similar cases, the fact
that some often adduced etymology is not included in the lexicon means that I
found it too incredible. On the other hand, I am sure that there are some good
Celtic etymologies that were left out simply because I was unaware that they
had been proposed.
2. The sources
In compiling the material for this lexicon, I have consulted all of the
existing etymological dictionaries of Celtic languages published after 1950. I
have not systematically used the older reference works, such as A. Holder's Alt-celtischer
Sprachschatz, or W. Stokes' Urkeltischer Sprachschatz, because the
material they contain has been well analyzed in later etymological
dictionaries. So far the largest collection of Celtic etymologies can be found
in Vendryès’ Léxique étymologique de l'irlandais ancien (LEIA); these
are generally reliable, but often inconclusive and seldom very imaginative.
Unfortunately, LEIA remains unfinished. Mac Bain's etymological dictionary of
Scottish Gaelic is completely outdated and unreliable. Etymological notes in Geiriadur
Prifysgol Cymru (GPC) are short, but often correct, and they remain the
most valuable etymological resource for Welsh. A. Falileyev’s dictionary of Old
Welsh is useful mostly for its rich philological documentation. Another
valuable etymological source is Xavier Delamarre’s Dictionnaire de la langue
gauloise, although it contains the etymologies only of those Celtic words
that are attested in Gaulish. Gaulish loanwords in French and other Romance
languages can be gathered from the relevant etymological dictionaries (e.g.
Gamillscheg and FEW for French), but they have also been the subject of several
articles, e.g. Bolelli 1941-2, Corominas 1976, Campanile 1983, Fleuriot 1991).
Latin words of Celtic origin have been treated quite exhaustively in a paper by
M. L. Porzio Gernia (1981). Words attested in Celtiberian inscriptions have
been gathered and subjected to a careful philological and etymological analysis
by Dagmar Wodtko in her Wörterbuch der keltiberischen Inschriften (MLH
V.1). For Breton, we have two etymological dictionaries. The dictionary by
Guyonvarc'h was conceived very ambitiously, but only a few fascicles were
published; the new dictionary by A. Deshayes (2003) is reasonably complete and
generally reliable, but does not offer detailed Proto-Celtic reconstructions
and any IE etymologies. Furthermore, for Old Breton, we have a very careful and
exhaustive work by Léon Fleuriot, Dictionnaire des gloses en vieux breton
(DGVB). Finally, for Cornish we have only one etymological dictionary by E.
Campanile, who analyzed the lexicon of the Old Cornish glosses. I have also
made good use of Stefan Schumacher's Die keltischen Primärverben, which
contains a lot of detailed etymological analyses of Celtic verbs with an
Apart from the mentioned sources, I consulted
the reference works on Indo-European etymology, most notably Pokorny’s
dictionary (IEW), EIEC, and LIV. Unfortunately, Nomina im indogermanischen
Lexikon (Wodtko et alii 2008) appeared too late for it to be used
systematically in the preparation of this dictionary. I also profited a lot
from the etymological databases prepared for the ‘New Pokorny’ project by my
Leiden colleagues, especially the Indo-Aryan database by A. Lubotsky, Latin and
Italic by M. de Vaan, Hittite by A. Kloekhorst, and Baltic and Slavic by R.
3. From Proto-Indo-European to Proto-Celtic
It is assumed
here that PIE had the following phonemes:
e o ē ō
and *u were presumably just allophones of the semi-vowels *y and *w. The status
of PIE *a is controversial. Following the Leiden school, I believe that PIE had
no *a in the original, inherited lexicon (Lubotsky 1989), but this vowel occurs
in several words that are probable loanwords from unknown, non-IE sources. In
some cases, *a served as an epenthetic vowel separating difficult consonant
clusters, e.g. Lat. pateo, < *pt-eh1- (cf. PCelt.
*fatamā ‘palm of the hand, talon’).
labials *p (*b) *bh
dentals *t *d *dh
velars *k *g *gh
palatalized *ḱ *
labiovelars *kw *gw *gwh
resonants: *m *n *l *r
glides: *y *w
The phonemic status of the difference between pure
velars and palatalized velars in PIE is a disputed matter. It is quite probable
that the phonological opposition between them was restricted to just a few
environments. The syllabic resonants were just allophones of the non-syllabic
resonants, occurring in the syllable nucleus. Therefore, they are not
distinguished graphically from the non-syllabic resonants in the PIE
reconstructions. The exact phonetic realization of PIE stops is a matter of
controversy; the traditional ‘voiced stops’ may have been ejectives, perhaps in
Early PIE. The phonetic realization of the ‘laryngeals’ is unknown, so they are
marked with indexes (*h1, *h2, *h3).
Laryngeals may have been lost in some environments already in PIE, or
dialectally, not long after the dissolution of the proto-language, e.g. before
*y (Pinault’s rule), or after the sequence *oR (de Saussure’s rule). However,
the validity of these rules of laryngeal loss, as well as their exact
formulation, are controversial.
Here are the principal Celtic
sound changes ordered into an approximate relative chronology:
A) Dialectal IE changes:
1. *h1e > *e, *h2e > *a,
*h3e > *o
2. *eh1 > *ē, *eh2 >
*ā, *eh3 > *ō
3. *CHC > *CaC, cf. PIE *ph2tēr
> PCelt. *fatīr (OIr. athir).
4. *CstopHCstop > CstopCstop
in non-initial syllables, cf. PIE *dhugh2tēr
‘daughter’ > PCelt. *duxtīr (Gaul. duxtir). This development is
somewhat uncertain in the light of Celtib. tuateros ‘daughter’ (Gen.
5. TT > *-ss-, cf. PIE *krd-tu- > PCelt.
*krissu- ‘belt’; the same development is found in Italic and in Germanic.
6. *CRHC > *CRaHC (> *CRāC), cf. PIE *plh1no-
‘full’ > PCelt. *flāno- (OIr. lán), PIE *rHno- ‘grain’ > PCelt. *grāno- (OIr. grán).
Laryngeals were probably preserved after *Ra until the operation of Dybo’s law
(A7), and then lost, with the compensatory lengthening of *aH > *ā. The
change *CRHC > *CRāC occurred in Italic as well.
7. *VHC > VC in pretonic syllables (Dybo’s law, cf.
Dybo 1961): PIE *wiHró- ‘man’ > PCelt. *wiro- (OIr. fer). In all
non-problematic examples of Dybo’s law the laryngeal was lost after *i, *u, or
*a which is the result of the development of syllabic resonants before
It is assumed here that the laryngeals had already been lost after *e and *o,
which were lengthened (A2). Dybo’s law was posterior to the change of CRHC >
CRāC (A6) because of the development of *sfraxto- ‘eloquent’, *frati- ‘fern’, and *klad-o-
‘dig’. Something like Dybo’s law also operated in Italic, and, in some form, probably in Germanic as
well (cf. Lat. uir, OE wer < *wiHró-; maybe the vowel
shortening (or laryngeal loss) was restricted to the position before resonants
in Italic and Germanic). I assume that the operation of Dybo’s law in Celtic
was general (i. e. unrestricted by phonetic environment). The apparent exceptions
to the operation of Dybo’s law in Celtic are best treated as analogical
re-introductions of vowel length from the forms of the root where the length
was preserved regularly. Of course, since the position of the accent in PIE
cannot be established for many PIE etymons of PCelt. words, the operation of
Dybo’s law can often be just assumed, but not strictly proved.
8. #RHC- > RaC (cf. Beekes 1988). Although this
change is not universally accepted, it is found in the development of the
following etyma: *latyo- ‘day’, *natu- ‘poem’, *mati- ‘good’, *mak-o-
‘increase’, *mad-yo- ‘break’, *laxsaro- ‘shine’ (PIE *r could not occur
word-initially, so here R = m, n, and l). The same change occurred in Italic and
Balto-Slavic, and probably in other European IE branches. The development of
*#yHC- and *wHC- is uncertain, but cf. the lemmata *yalo- ‘clearing’ and
*waxto- ‘bad’ for the possibility that *H > *a in this position.
9. Merger of PIE palatalized velars and pure velars,
cf. PIE *deḱm ‘ten’ > PCelt. *dekam (OIr. deich).
This development is shared by all Centum branches of Indo-European.
B) Early PCelt. changes:
1. *gw > *b, cf. PIE *gwow-
> PCelt. *bow- > ‘cow’ (OIr. bó).
2. Deaspiration of aspirated stops, cf. PIE *bher-o-
> PCelt. *ber-o- ‘carry’ (OIr. berid). This sound change was
obviously posterior to *gw > *b (B1), because PCelt. *gwh
> *gw, cf. PIE *gwher- > PCelt. *gwer-o-
‘heat up’ (MIr. geirid).
3. CLCstop > CLiCstop (where
L = r, l), cf. PIE *ḱrd- > PCelt.
*krid-yo- ‘heart’ (OIr. cride). It is probable that the same development
occurred before PIE *m (cf. *kwrmi- ‘worm’ > PCelt. *kwrimi-,
*h1lmo- ‘elm’ > PCelt. *limo- (s. v. *lēmo-, *limo-). This
change was anterior to the general change of CRC > CaRC (B5) which was otherwise
unrestricted by phonetic environment.
4. *eRa > *aRa (Joseph’s rule, cf. Joseph 1982),
cf. PIE *terh1tro- ‘auger’ > PCelt. *taratro- (OIr. tarathar,W
taradr). PIE *e did not become *a before *Rā (cf. PIE *gwenh2
>> *gweneh2 > *gwenā > PCelt.
*benā, OIr. ben), and the vowel *e was restored analogically before
*Ra in many instances, e.g. in the reduplicated syllables in the perfect
(PCelt. *me-mad- > OIr. memaid ‘broke’, 3sg. perf. of maidid ‘breaks’
< PCelt. *mad-yo-). This change preceded the decomposition of syllabic
nasals (B5) because of the development of PCelt. *elan(t)ī ‘doe,
hind’ < *h1eln(t)ih2 (rather than *alan(t)ī), but
after the vocalization of laryngeals between consonants, because of the
development of *taratro- < *terh1tro- above.
5. CRC > CaRC, cf. PIE *dnt- > PCelt. *danto-
‘tooth’ (OIr. dét, W dant), PIE *mrwo- > PCelt. *marwo- ‘dead’
(OIr. marb, W marw), PIE *bhrso- > PCelt. *barso-
> *barro- ‘point, top’ (OIr. barr). Note that syllabic liquids had
already developed to *ri, *li before stops and *m (B3).
6. Loss of laryngeals in non-syllabic position. This
change is later than the development of syllabic resonants (B5), because of,
e.g., PIE *ḱlHeto- > *kalHeto- > PCelt. *kaleto-
‘hard’, PIE *wlHo- > PCelt. *walo- ‘ruler, chief’, PIE *smh2eli-
> PCelt. *samali- ‘similitude’, PIE *snHi > PCelt. *sani- ‘without’.
7. *p...kw > *kw...*kw,
cf. PIE *penkwe > PCelt. *kwenkwe ‘five’
(OIr. cóic, MW pymp). This change predated the development of *kw
> *x before stops (C1) if PCelt. *kwerxt- ‘bush’ (W perth)
is from PIE *perkw- ‘oak’. It is assumed here that the similar
assimilation in Italic (cf. Lat. quercus) is a parallel development (for
arguments see below).
8. *ē > *ī, cf. PIE *Hrē- ‘king’ > PCelt. *rīg- (OIr. rí).
This change must predate PCelt. *p > *f > Ø (C4), because of PIE *h1epirom
> PCelt. *efirom (> *eyrom > *ērom, OIr. íar ‘after’, not
9. *ō > *ū in final syllables, cf. PIE *ḱwōn ‘dog’ > PCelt. *kwū(n)
(OIr. cú, W ci).
11. *V:RC > *VRC (Osthoff-type shortening before
resonants in closed syllables), cf., e.g., PIE *h2weh1nto-
‘wind’ > *wēnto- > *wīnto- > PCelt. *winto-, perhaps also
PIE *sih2m-do- > *sīndo- > PCelt. *sindo- ‘that’. This
change was obviously posterior to *ē > *ī (B7).
C) Late PCelt. changes:
1. *C1C2 > *xC2
(where C2 stands for any stop and *s), cf. PIE *septm ‘seven’ >
PCelt. *sextam (OIr. secht). This change is posterior to TT > *ss,
and also to CRCstop > CRiCstop (B3) because of PIE
*prptu- > PCelt. *frixtu- ‘form’, *mrgwto- > PCelt. *mrixto-
> PCelt. *mlixto- ‘milk’.
2. *pL > *bL (where L stands for any liquid), cf.
PIE *pi-prh3-se- > PCelt. *pibrase- ‘will bestow, will give’ >
*fibrase- (OIr. ebraid), PIE *dwey-plo- ‘double’ > PCelt.
*dwēblo- (OIr. díabul).
3. *pN > *wN (where N is any nasal), presumably
only after back vowels, cf. PIE *supno- > PCelt. *suwno- > *sowno-
‘sleep’ (OIr. súan, W hun).
4. *p > *f, cf. PIE *ph2tēr >
PCelt. *fatīr ‘father’ (OIr. athir).
5. *ō > ā, cf. PIE *deh3no-
> *dōno- > PCelt. *dāno- ‘gift’ (OIr. dán); this change
is obviously later than the change of *ō > ū in final syllables
(B8). Clear examples of this change in Celtiberian are lacking, but there are
6. *ey > *ē, cf. PIE *(H)reyd- > PCelt.
*rēd-o- ‘ride’ (OIr. réidid). This change was obviously later than
*ē > *ī (B7). There is some uncertainty whether this change also
occurred in Celtiberian.
7. *ew > *ow, cf. PIE *newyo- > PCelt. *nowyo-
‘new’ (OIr. núae, W newydd).
8. *uw > *ow/_C, cf. PIE *supno- > PCelt.
*suwno- > *sowno- ‘sleep’ (OIr. súan). This change probably did not
apply before *-i- in the next syllable, because of *dru-wid- > OIr. druí ‘druid’
(rather than **droí). It is unclear whether this change applied in
Proto-Celtic, or just in Goidelic and Brittonic (data from Gaulish and
Celtiberian are lacking).
D) Some other probable PCelt.
1. The liquid assimilations *rp > *rf > *rr (PIE
*serp- > PCelt. *serrā) and *lp > *lf > *ll (PIE *kulp- >
PCelt. *kul(f)o-), *-rs- > *-rr- (PIE *bhrso- > PCelt.
*barro-), *rst > *rt (PIE *trstu- ‘thirst’ > OIr. tart). All of
the attested languages show the results of assimilations, so it is simpler to
project those changes to Proto-Celtic. It is possible, however, that at least
some of the assimilations were parallel innovations of individual languages
after the break-up of Proto-Celtic.
2. The assimilation of *mw > *ww, cf.
*kom-wīro- > PCelt. *kowwīro- ‘true’ (W cywir).
3. The lengthening of the vowel before the cluster
*xsL, cf. *toḱ-slo- ‘axe’ > *tōxslo- > PCelt.
*tāxslo- (OIr. tál); it is possible that PCelt. *x was lost and
that the preceding vowel was subject to compensatory lengthening. However, *x
is conventionally retained in the PCelt. reconstuctions because the regularity
of this change is uncertain; cf. the lemmata *dīro-, *kīsrā,
*muxto‑, *sego- and *skāxslo- for possible instances of this change,
but also *tullo- for a possible counter-example.
4. The ‘liquid metathesis’ (*ar, *al > *ra, *la)
between labials and a cluster of two dentals: PIE *mlsto- > PCelt. *mlasto-
‘taste’, PIE *gwrsto- > PCelt. *brasso- ‘great’, PIE *h1wrsto-
> PCelt. *wrasto- ‘shower’, PIE *wlsno- > *walsno- > PCelt. *wlanno-
‘blood red’, etc. Like the preceding one, this change is proposed here for the
first time, so it will probably raise some controversy. Another possibility is
to assume an analogical zero-grade CCLaC of the roots with the full grade CLeCC
(after the regular pattern with the zero-grade CaLCC and the full grade
5. The loss of laryngeals after *ey (and *oy?) before
consonants (*VyHC > *VyC), cf. PCelt. *bēto- ‘food’ (W bwyd)
< PIE *gweyh3to-, PCelt. *dēno- ‘fast’ (OIr. dían)
< *deyHno-, PCelt. *fētu- ‘(grass-)land, territory’ (OIr. íath)
< PIE *peyHtu-, PCelt. *kwēno- ‘long’ (OIr. cían)
< *kweyHno-, PCelt. *rēno- ‘large expanse of water’ (OIr. rían)
< PIE *(H)reyHno-, PCelt. *wēro- ‘crooked’ (OIr. fíar, W gŵyr)
< PIE *weyh1ro-, etc. There are only two apparent
counter-examples in this dictionary: W rhaeadr ‘torrent’ and OIr. ríathor
(with disyllabic ía), do not represent PCelt. *reyatro-, but rather
*riyatro-; likewise, OIr. disyllabic bíad ‘food’ can be derived from
*biyato- or, less probably, *biwato- (rather than *beyato- < *gweyh3to-).
There are no truly reliable examples for the loss of laryngeals after *oy, but
cf. PCelt. *koylo- ‘thin’ (which can be from PIE *koyHlo-) and *oyno- ‘one’
(which some linguists would derive from *oyHno- < *h3eyHno-).
E) Some doubtful changes:
1. *(C)RHCdentalC > *(C)RaCdentalC,
cf. PIE (?) *prh3-sneh2 ‘gift, share’ > PCelt.
*frasnā (OIr. rann); however, PCelt. *frasnā, just like Lat. pars,
can be derived from a root without laryngeal (generalized after the present
stem *pr-neh3- / *pr-nh3- (> PCelt. *far-na-), and
explained by ‘liquid metathesis’. Alternatively, the short *a in the Celtic
reflexes of PIE roots of the form *CeRHC can always be the result of Dybo’s
law, and it is, of course, more economical to assume fewer sound changes.
2. #RCvoicedC > #RaCvoicedC
(Schrijver’s rule, cf. Lat. magnus < *mnos); however, all of the alleged examples involve the
root *meh2- ‘great’ (cf. PCelt. *maglo-,
*magyo-, and *magos). None of those etymologies is beyond reproach.
3. *h2rCstop- > *arCstop-
and *h3rCstop- > *arCstop- (but *h1rCstop-
> *riCstop-, cf. *rig-o- ‘go’ < *h1rgh-o-). However, examples of
this change are few in number and quite controversial. PCelt. *orgyā
‘testicle’ can be derived from *h1orhi-
rather than *h1rhi- or *h3rhi-
assumed by some linguists; PCelt. *arto- ‘bear’ is from PIE *h2rtḱo-, but it probably went through the stage *h2rþk’o-
(and *#rþC- > *#arþC- may be assumed just like *#rsC > *#arsC). PCelt.
*arganto- ‘silver’ is a
problem, since the reconstruction *h2rnto- seems somewhat more probable than *h2ernto-. However, the word for ‘silver’ may have had an ablauting paradigm
in PIE (Gr. árgyros can be both from *h2er- and *h2r-, and Skt. árjuna- ‘shining’ is clearly from *h2er-).
4. The Problem of Italo-Celtic
shares a number of sound changes with Celtic, I remain unconvinced of the
‘Italo-Celtic hypothesis’. Very few phonological and morphological changes are
actually exclusive Italo-Celtic isoglosses, and, more importantly, one cannot
really establish a relative chronology of those isoglosses, as one can in the
case of, e.g., Balto-Slavic. However, there is little doubt that Italic and
Celtic developed from a group of closely related Western Indo-European
dialects. For a recent discussion of the Italo-Celtic hypothesis see Kortlandt
The following phonological
innovations of Italic and Celtic are shared:
1. The development of PIE syllabic resonants followed
by laryngeals, PIE *CRHC > *CrāC, cf. PIE *rHno- ‘grain’ > Lat. grānum, PCelt.
*grāno-. Note, however, that in PCelt. the development was actually from
*CRHC to *CRaHC, and then to *CRāC with loss of the laryngeal and
compensatory lengthening of *a (see above, changes A6-A7 in the relative
chronology). It is uncertain whether the same two-step development occurred in
2. The assimilation *p...kw > *kw...kw.
However, this change appears to be late in Celtic. It failed to occur in OIr. deac,
deëc ‘10’, which is often derived from *dwey-penkw-om
‘two-fives’, and when this compound was formed (in Proto-Celtic) *p was not in
the beginning of the word, and so it regularly changed to *f > Ø rather than
assimilating with *kw (see Watkins 1966: 145, but also the lemma
*dekan below for problems with this etymology). In any case, such an
assimilation is phonetically trivial (cf. the reverse assimilation in PIE *penkwe
‘5’ > Go. fimf).
3. The shortening of vowels in pretonic position
(Dybo’s law mentioned above); however, this change may not be limited to Italic
and Celtic, because it appears to affect Germanic as well, at least in some
examples, cf. OE wer ‘man’ < *wiHró- (Skt. vīrá-, Lith. výras,
Lat. uir, OIr. fer).
Italo-Celtic isoglosses are not more convincing:
1. The genitive ending *-ī is neither pan-Celtic
(it is lacking in Celtiberian) nor pan-Italic (it is lacking in Sabellic), and
it is not exclusively Italo-Celtic (it occurs in Messapic and probably in
Tocharian). Actually, it is an old petrified adjectival form (see
Matasović 2004) and, as such, does not represent a common innovation in
Italic and Celtic.
2. The generalization of the *so- stem of the PIE
demonstrative pronoun *so-/*to- is a parallel development, since there are
clear traces that PCelt. still had the pronominal stem *to- (see PCelt. *tod).
Moreover, it is unclear whether Celtiberian shared the generalization of the
3. The introduction of the Gen. ending *-strom in the
1st and 2nd person plural pronouns is not very significant, since the stems of
the 2nd person plural pronouns are different in Celtic and Italic, and the
forms that actually have this ending are attested only in OIr. (nathar)
and Latin (nostrum), so we cannot be sure if they were ever pan-Celtic
4. The spread of the Dat./Abl. pl. ending *-bhos
is uncertain, since in Gaulish we only have -bo, and Irish generalized
the ending *-bhis from the Instrumental.
5. The superlative ending *-smmo- is indeed a shared
exclusive isogloss, but in itself it is not enough to prove the existence of an
5. The Sub-classification of Celtic
The exact genetic subclassification of the Celtic languages is still an
unsettled matter. Two approaches dominate the current discussions:
The traditional view,
defended, among others, by Schmidt (1977), Koch (1992), and de Bernardo Stempel
(2006) who classify Brittonic together with Gaulish (and Lepontic, which is
probably just an early offshoot of Gaulish) into Gallo-Brittonic, while the
Goidelic languages remain as a separate branch of Celtic (see Fig. 1.1. below).
An alternative theory,
defended by e.g. McCone (1996) and supported by Schrijver (1995) and Schumacher
(2004), who see Brittonic and Goidelic as a separate Insular Celtic branch,
while Gaulish and Lepontic are viewed as the Continental Celtic branch.
Celtiberian, as is becoming increasingly clear, is almost certainly an
independent branch on the Celtic genealogical tree, one that became separated
from the others very early (see Fig. 1.2. below):
The problem of
sub-classification within Celtic is relevant to this work only inasmuch as it
affects the reliability of Proto-Celtic reconstructions. If there was an
Insular Celtic branch within Celtic, then forms reconstructed on the basis of
just Old Irish and Middle Welsh, the two best attested early Celtic idioms,
need not go back to common Proto-Celtic, but may instead represent
Proto-Insular Celtic. Likewise, if we assume the existence of a Gallo-Brittonic
branch, then we should be careful in attributing reconstructions based on
evidence from the Brittonic languages and Gaulish to Proto-Celtic.
As I have argued elsewhere
(Matasović 2008), I tend to view Insular Celtic more as an areal than as a
genetic grouping. This does not, however, imply that I believe in
Gallo-Brittonic as a valid genetic grouping, either: in the matters of genetic
sub-classification within Celtic I think it is wise to remain agnostic, until
more is known about Gaulish, Lepontic, and Celtiberian. In any case, since the
argument about Insular Celtic vs. Gallo-Brittonic tends to revolve more around
the morphological than around the phonological isoglosses, it is unlikely that
the eventual resolution of the debate will substantially affect the
Proto-Celtic reconstructions proposed here.
6. On the reconstruction of Proto-Celtic
several unresolved issues in the reconstruction of the Proto-Celtic
phonological system. I have generally tried to follow the consensus opinion,
where there is any, but in some cases difficult choices had to be made. It is
assumed here that Proto-Celtic had the following phonemes:
a) short b)
i u ī ū
e o ē
ay aw āy āw
I take the monophthongization of PIE *ey > *ē
to be a Proto-Celtic change, although it is not absolutely certain that this
change occurred in the prehistory of Celtiberian (cf. MLH V.1: XVII). I also
believe that the change *ew > *ow is Proto-Celtic, and that instances of
alleged eu in Gaulish (e.g. in Neviodunum, a toponym in Slovenia)
are just spelling variants of a diphthong that did not exist in Latin at the
time of the adaptation of the Roman alphabet to Gaulish (cf. McCone 1996).
I assume that *kw merged with *kw in
Proto-Celtic, so I reconstruct PCelt. *ekwo- ‘horse’ (OIr. ech,
etc.) from PIE *h1eḱwo-
(Lat. equus etc.). Apparent exceptions, such as W ci ‘dog’ <
PIE *ḱwōn can be explained by assuming early
delabialization of *kw in certain environments (e.g. before PCelt.
*ū as in the preceding example: *ḱwōn
> PCelt. *kwūn > *kūn > W ci). Similarly,
the reflexes of *gw(h) and *g(h)w are indistinguishable
in Celtic, cf. PCelt. *tangwāt- ‘tongue’ < *dnhw- (OIr. tengae, W tafod).
f s [x] (an
allophone of *k before stops and *s]
[z] (an allophone of *s
before voiced consonants)
I do not assume that there was a PCelt. phoneme *ts (from PIE clusters
with two dentals, and/or from PIE *-st-). I believe that PIE *st was preserved
in PCelt. (as it is in Celtiberian), and that PIE *TT yielded *ss already in
PCelt. (see Schrijver 1995). The fricative *f is the regular reflex of PIE *p.
It may have been a bilabial voiceless fricative [φ] phonetically, rather
than a labiodental fricative [f]. I also assume that the assimilations of *rs
> *rr and *ls > *ll are Proto-Celtic (see McCone 1996); however, I adopt
the ‘etymological’ spelling for the clusters *-sr-, *-sl-, *-sn-, *-sm-, and *ly,
as if they were intact in Proto-Celtic reconstructions, although they could
have changed to *-rr-, *-ll-, *-nn- and *-mm-, respectively, already in PCelt.
m n l r
I assume that
the change of PIE syllabic *m, *n > *am, *an is pan-Celtic. The fronting of
*am, *an > *em, *en in Goidelic is a later development that occurred only in
some environments (see McCone 1996 for details). I also assume that word-final
*-m was preserved, as it is in Celtiberian, and occasionally in Gaulish.
7.The Celtic languages
For the purpose
of this lexicon we adopt the following periodization of the attested Celtic
1. Lepontic (attested from the 7th, or early 6th
century B.C. until ca. the 1st century B.C.). In all likelihood, Lepontic is
just an early offshoot of Gaulish. The evidence for Lepontic as a separate
branch of Celtic heavily relies on the archaeological data, especially on the
early individualization of the (Lepontic) Golasseca Culture (see Uhlich 1999:
2. Gaulish, attested onomastically since the 6th
century B.C., but with a sizeable corpus of inscriptions only from the 3rd
century B.C. (inscriptions in Greek alphabet). Inscriptions in the Roman
alphabet are attested later, chiefly after the Roman conquest of Gaul (2nd half
of the 1st century B.C.). It is unclear when Gaulish died out, but it was
probably spoken until the 6th or 7th century A.D., at least in some isolated
pockets in Gaul. Although Gaulish is attested for at least a millennium, no
attempt has been made to distinguish between early and late Gaulish in this
lexicon. However, the source of Gaulish words (except for names) is always
attested from the 3rd or early 2nd century B.C. until ca. 1st century A.D. The
earliest inscriptions are in Iberian syllabary, but from the 1st century B.C. a
considerable number of inscriptions are in Roman alphabet.
represented by Old Irish and Middle Irish, attested since 4th century A.D.
(Ogam inscriptions). We distinguish the following phases of Goidelic:
Ogam (4th - 6th centuries)
Old Irish (7th - 9th centuries)
Middle Irish (10th - 11th
Modern Irish (12th century -
Usually only the earliest attested form of the word is
adduced in the lexicon. This is regularly the Old Irish or the Middle Irish
form found in DIL. It should be noted that Old Irish and Middle Irish forms are
often not easily distinguished. In principle, all words attested in the glosses
and other texts from the Thesaurus Palaeohibernicus are Old Irish, but
Old Irish forms can often be found in later manuscripts as well. So, even if a
word is only attested in texts, the manuscripts of which were preserved in the
Middle Irish period (e.g. in the sagas of the Ulster cycle, or in the Leinster
eulogistic poetry), we can often be sure that the same word existed in Old
Irish. Therefore, in some cases where I adduced a word as OIr., although it
does not occur in the proper OIr. texts, the reader will have to trust my
philological judgement, or check the sources for himself.
Scottish Gaelic and Manx forms
were not adduced in this lexicon (with a handful of exceptions), since they
yield very little additional information about the reconstructed PCelt. words
and their origin.
represented by Welsh, Breton, and Cornish. Dialectal diversity within Brittonic
is far greater than within Goidelic, so reflexes of PCelt. words from all three
Brittonic languages were adduced, whenever attested. We distinguish the
following phases of Brittonic:
Old Welsh (7th - 10th centuries)
Middle Welsh (11th - 14th
Modern Welsh (15th century -
Old Breton (9th - 11th
Middle Breton (12th - 16th
Modern Breton (17th century -
Old Cornish (9th -12th
Cornish (13th - 18th centuries)
Middle Breton, and Middle Cornish forms are adduced by default. If a word was
attested in Old Cornish, Old Breton, or Old Welsh, it is adduced separately in
the Cornish, Breton, and Welsh fields, respectively. Modern Welsh forms, as
cited in GPC, are adduced only when they are different from MW forms, and often
the difference lies only in spelling.
8. Structure of the entries
Each entry in
this dictionary consists of several fields. The first field contains the
reconstructed Proto-Celtic word and its meaning, as well as the information
about the word-class to which it belonged. The following fields contain
reflexes of the reconstructed etymon in the primary branches of Celtic,
together with some basic grammatical information about them: as a rule the
gender of Old Irish and Middle Welsh nouns, the inflectional class of the Old
Irish nouns and adjectives, and the attested stems of Old Irish verbs.
In the next field the PIE root of the Proto-Celtic etymon is given,
together with a reference to the page in IEW where that root is discussed.
After that, there follows a discussion of the proposed etymology and the
alternative proposals found in the literature. I have tried to make the
derivation of the attested forms from PIE and Proto-Celtic as explicit as
possible, without concealing any of the uncertainties or unresolved problems.
This field also contains the discussion of other possible cognates of the
reconstructed Proto-Celtic etymon. In many cases the etymology proposed here is
not the only possible one, but all of the etymologies in this dictionary are
meant to be consistent with the Celtic sound laws accepted in this Introduction
(see above). I have tried to be maximally clear in making the distinction between
mentioning various possible etymological proposals and claiming that a
particular etymology is true.
The last field contains the references. The list of references is
not meant to be exhaustive, but rather to include the most relevant books and
articles, published during the last fifty years, in which etymological
discussion of the etyma in question can be found. They are ordered in such a
manner that the more general reference works precede publications dedicated to
a particular word, or specific problems of phonological development relevant to
the lemma in question.