Proto-Celtic language

These changes are shared by several other Indo-European branches.

• *e is colored by an adjacent laryngeal consonant:
  • eh₂, h₂e > ah₂, h₂a
  • eh₃, h₃e > oh₃, h₃o
• Palatovelars merge into the plain velars:
  • ḱ > k
  • ǵ > g
  • ǵʰ > gʰ
• Epenthetic *a is inserted after a syllabic sonorant if a laryngeal and another sonorant follow (R̥HR > RaHR)
• Laryngeals are lost:
  • before a following vowel (HV > V)
  • following a vowel in syllables before the accent (VHC´ > VC´)
  • following a vowel, resulting in compensatory lengthening, thus (VH > V̄)
  • between plosives in non-initial syllables (CHC > CC)
• Two adjacent dentals become two adjacent sibilants (TT > ss)

The following sound changes are shared with the Italic languages in particular, and are cited in support of the Italo-Celtic hypothesis.[7]

• Dybo's rule: long close vowels are shortened (or a laryngeal is lost) before resonant + stressed vowel.
  • īR´ / ? *iHR´ > iR´
  • ūR´ / ? *uHR´ > uR´
• Possibly, post-consonantal laryngeals are lost when before pre-tonic close vowels:
  • CHiC´ > CiC´
  • CHuC´ > CuC´
• Development of initial stress, following the previous two changes.
• Possibly, vocalization of laryngeals to *ī between a *CR cluster and consonantal *j (CRHjV > CRījV)
• Syllabic laryngeals become *a (CHC > CaC)
• Syllabic resonants before a voiced unaspirated stop become *Ra (R̥D > RaD)
• *m is assimilated or lost before a glide:
  • mj > nj
  • mw > w
• *p assimilates to *kʷ when another *kʷ follows later in the word (p…kʷ > kʷ…kʷ)
• sVs > ss, sTVs > Ts

One change shows non-exact parallels in Italic: vocalization of syllabic resonants next to laryngeals depending on the environment. Similar developments appear in Italic, but for the syllabic nasals *m̥, *n̥, the result is Proto-Italic *əm, *ən (> Latin em ~ im, en ~ in).

• Word-initially, HR̥C > aRC
• Before voiceless stops, CR̥HT > CRaT
• CR̥HV > CaRHV
• CR̥HC > CRāC

• Sequences of velar and *w merge into the labiovelars (it is uncertain if this preceded or followed the next change; that is, whether gw > b or gw > gʷ, but Schumacher 2004 argues on p. 372 that this change came first; moreover, it is also found in Proto-Italic, and thus arguably belongs to the previous section):
  • kw > kʷ
  • gw > gʷ
  • gʰw > gʷʰ
• gʷ > b
• Aspirated stops lose their aspiration and merge with the voiced stops (except that this counterfeeds the previous change, so *gʷʰ > *gʷ doesn't result in a merger):
  • bʰ > b
  • dʰ > d
  • gʰ > g
  • gʷʰ > gʷ
• *e before a resonant and *a (but not *ā) becomes *a as well (eRa > aRa): *ǵʰelH-ro > *gelaro > *galaro / *gérH-no > *gerano > *garano (Joseph's rule).
• Epenthetic *i is inserted after syllabic liquids when followed by a plosive:
  • l̥T > liT
  • r̥T > riT
• Epenthetic *a is inserted before the remaining syllabic resonants:
  • m̥ > am
  • n̥ > an
  • l̥ > al
  • r̥ > ar
• All remaining nonsyllabic laryngeals are lost.
• ē > ī
• ō > ū in final syllables
• Long vowels are shortened before a syllable-final resonant (V:RC > VRC); this also shortens long diphthongs. (Osthoff's law)

• Plosives become *x before a different plosive or *s (C₁C₂ > xC₂, Cs > xs)
• p > b before liquids (pL > bL)
• p > w before nasals (pN > wN)
• p > ɸ (except possibly after *s)
• ō > ā
• ew > ow
• uwa > owa

The following consonants have been reconstructed for Proto-Celtic (PC):

Type	Bilabial	Alveolar	Palatal	Velar
				plain	labialized
Plosive		t		k	kʷ
Plosive	b	d		ɡ	ɡʷ
Nasal	m	n
Fricative	ɸ	s		x
Approximant		l	j		w
Trill		r

PC stops allophonically manifest similarly to those in English. Voiceless stop phonemes /t k/ were aspirated word-initially except when preceded by /s/, hence aspirate allophones [tʰ kʰ]. And unaspirated voiced stops /b d ɡ/ were devoiced to [p t k] word-initially.[10][11]

This allophony may be reconstructed to PC from the following evidence:[10][11]

• Modern Celtic languages like Welsh, Breton, and all modern Goidelic languages have such plosive aspiration and voice allophony already attested.
• Several old Celtic languages (such as Old Irish, Old Welsh, and Lepontic) used letters for voiceless stop phonemes to write both voiceless stop phonemes and their voiced counterparts, especially non-word-initially.
• The Celtiberian Luzaga's Bronze has the curious spelling of an accusative determiner sdam, where the d is clearly meant to spell [t]. This implies that Celtiberian /d/ had a voiceless allophone [t].

Proto-Indo-European (PIE) voiced aspirate stops *bʰ, *dʰ, *gʰ/ǵʰ, merge with *b, *d, *g/ǵ in PC. The voiced aspirate labiovelar *gʷʰ did not merge with *gʷ, though: plain *gʷ became PC *b, while aspirated *gʷʰ became *gʷ. Thus, PIE *gʷen- 'woman' became Old Irish and Old Welsh ben, but PIE *gʷʰn̥- 'to kill, wound' became Old Irish gonaid and Welsh gwanu.

PIE *p is lost in PC, apparently going through the stages *ɸ (possibly a stage *[pʰ])[10] and *h (perhaps seen in the name Hercynia if this is of Celtic origin) before being completely lost word-initially and between vowels. Next to consonants, PC *ɸ underwent different changes: the clusters *ɸs and *ɸt became *xs and *xt respectively already in PC. PIE *sp- became Old Irish s (lenited f-, exactly as for PIE *sw-) and Brythonic f; while Schrijver 1995, p. 348 argues there was an intermediate stage *sɸ- (in which *ɸ remained an independent phoneme until after Proto-Insular Celtic had diverged into Goidelic and Brythonic), McCone 1996, pp. 44–45 finds it more economical to believe that *sp- remained unchanged in PC, that is, the change *p to *ɸ did not happen when *s preceded. (Similarly, Grimm's law did not apply to *p, t, k after *s in Germanic, and the same exception occurred again in the High German consonant shift.)

Proto-Celtic	Old Irish	Welsh
*laɸs- > *laxs- 'shine'	las-aid	llach-ar
*seɸtam > *sextam 'seven'	secht	saith
*sɸeret- or *speret- 'heel'	seir	ffêr

In Gaulish and the Brittonic languages, the Proto-Indo-European *kʷ phoneme becomes a new *p sound. Thus, Gaulish petuar[ios], Welsh pedwar "four", but Old Irish cethair and Latin quattuor. Insofar as this new /p/ fills the gap in the phoneme inventory which was left by the disappearance of the equivalent stop in PIE, we may think of this as a chain shift.

The terms P-Celtic and Q-Celtic are useful for grouping Celtic languages based on the way they handle this one phoneme. But a simple division into P- / Q-Celtic may be untenable, as it does not do justice to the evidence of the ancient Continental Celtic languages. The many unusual shared innovations among the Insular Celtic languages are often also presented as evidence against a P- vs Q-Celtic division, but they may instead reflect a common substratum influence from the pre-Celtic languages of Britain and Ireland,, or simply continuing contact between the insular languages; in either case they would be irrelevant to the genetic classification of Celtic languages.

Q-Celtic languages may also have /p/ in loan words, though in early borrowings from Welsh into Primitive Irish, /kʷ/ was used by sound substitution due to a lack of a /p/ phoneme at the time:

• Latin Patricius "Saint Patrick"' > Welsh > Primitive Irish Qatrikias > Old Irish Cothrige, later Pádraig;
• Latin presbyter "priest" > early form of word seen in Old Welsh premter primter > Primitive Irish qrimitir > Old Irish cruimther.

Gaelic póg "kiss" was a later borrowing (from the second word of the Latin phrase osculum pacis "kiss of peace") at a stage where p was borrowed directly as p, without substituting c.

The PC vowel system is highly comparable to that reconstructed for PIE by Antoine Meillet. The following monophthongs are reconstructed:

Type	Front		Central		Back
	long	short	long	short	long	short
Close	iː	i			uː	u
Mid	eː	e				o
Open			aː	a

The following diphthongs have also been reconstructed:

Type	With -i	With -u
With a-	ai	au
With o-	oi	ou

The morphological (structure) of nouns and adjectives demonstrates no arresting alterations from the parent language. Proto-Celtic is believed to have had nouns in three genders, three numbers and five to eight cases. The genders were masculine, feminine and neuter; the numbers were singular, plural and dual. The number of cases is a subject of contention:[12] while Old Irish may have only five, the evidence from Continental Celtic is considered rather unambiguous despite appeals to archaic retentions or morphological leveling. These cases were nominative, vocative, accusative, dative, genitive, ablative, locative and instrumental.

Nouns fall into nine or so declensions, depending on stem. There are *o-stems, *ā-stems, *i-stems, *u-stems, dental stems, velar stems, nasal stems, *r-stems and *s-stems.

• makkʷos 'son' (masculine) (Old Irish mac ~ Welsh, Cornish and Breton mab)

• dūnom 'stronghold' (neuter)

E.g. *ɸlāmā 'hand' (feminine) (Old Irish lám; Welsh llaw, Cornish leuv, Old Breton lom)

E.g. *sūlis 'sight, view, eye' (feminine) (Brittonic sulis ~ Old Irish súil)

E.g. *mori 'body of water, sea' (neuter) (Gaulish Mori- ~ Old Irish muir ~ Welsh môr)

E.g. *bitus 'world, existence' (masculine) (Gaulish Bitu- ~ Old Irish bith ~ Welsh byd ~ Breton bed)

E.g. *beru "rotisserie spit" (neuter)

Before the *-s of the nominative singular, a velar consonant was fricated to *-x : *rīg- "king" > *rīxs. Likewise, final *-d devoiced to *-t-: *druwid- "druid" > *druwits.[13]

E.g. *rīxs "king" (masculine)

E.g. *druwits "druid" (masculine)

E.g. *karants "friend" (masculine)

Generally, nasal stems end in *-on-; this becomes *-ū in the nominative singular: *abon- "river" > *abū.

E.g. *abū "river" (feminine)

E.g. *anman "name" (neuter)

Generally,*s-stems contain an *-es-, which becomes *-os in the nominative singular: *teges- 'house' > *tegos.

E.g.*tegos "house" (neuter)

• r-stems are rare and principally confined to names of relatives. Typically they end in *-ter-, which becomes *-tīr in the nominative and *-tr- in all other cases aside from the accusative: *ɸater- 'father' > *ɸatīr, *ɸatros.

E.g. *ɸatīr 'father' (masculine)

E.g. *mātīr 'mother' (feminine)

The following personal pronouns in Celtic can be reconstructed as follows:[14]^{: 220–221}[15]^: 281

The following third-person pronouns in Proto-Celtic may also be reconstructed.[16]^: 62[14]^: 220

From comparison between early Old Irish and Gaulish forms it seems that Continental and Insular Celtic verbs developed differently and so the study of Irish and Welsh may have unduly weighted past opinion of Proto-Celtic verb morphology. It can be inferred from Gaulish and Celtiberian as well as Insular Celtic that the Proto-Celtic verb had at least three moods:

• indicative — seen in e.g. 1st sg. Gaulish delgu "I hold", Old Irish tongu "I swear"
• imperative — seen in e.g. 3rd sg. Celtiberian usabituz, Gaulish appisetu
• subjunctive — seen in e.g. 3rd sg. Gaulish buetid "may he be", Celtiberian asekati

and four tenses:

• present — seen in e.g. Gaulish uediíu-mi "I pray", Celtiberian zizonti "they sow"
• preterite — seen in e.g. 3rd sg. Gaulish sioxti, Lepontic KariTe
• imperfect — perhaps in Celtiberian kombalkez, atibion
• future — seen in e.g. 3rd sg. Gaulish bissiet, Old Irish bieid "he shall be"

A probable optative mood also features in Gaulish (tixsintor) and an infinitive (with a characteristic ending -unei) in Celtiberian.[17][18]

Verbs were formed by adding suffixes to a verbal stem. The stem might be thematic or athematic, an open or a closed syllable.

The primary endings in Proto-Celtic were as follows. They were used to form the present, future, and subjunctive conjugations.[14]

There were two or three major preterite formations in Proto-Celtic, plus another moribund type.

• The s-preterite
• The reduplicated suffixless preterite (originating from the PIE reduplicated stative)
• The t-preterite
• The root aorist

The s-, t-, and root aorist preterites take Indo-European secondary endings, while the reduplicated suffix preterite took stative endings. These endings are:[19]^: 62–67

The Old Irish t-preterite was traditionally assumed to be a divergent evolution from the s-preterite, but that derivation was challenged by Jay Jasanoff, who alleges that they were instead imperfects of Narten presents. Either derivation requires Narten ablaut anyway, leading to a stem vowel i in the singular and e in the plural. The stem vowel in the t-preterite was leveled to *e if the next consonant was either velar or *m, and *i in front of *r or *l.[20]

One major formation of the future in Celtic, the s-future. It is a descendant of the Proto-Indo-European (h₁)se-desirative, with i-reduplication in many verbs. The Old Irish a- and s-future come from here.[21]

Another future formation, attested only in Gaulish, is the -sye-desiderative.

Most verbs took one subjunctive suffix in Proto-Celtic, -(a)s-, followed by the thematic primary endings. It was a descendant of the subjunctive of an Indo-European sigmatic thematic formation *-seti. The -ase- variant originated in roots that ended in a laryngeal in Proto-Indo-European; when the *-se- suffix was attached right after a laryngeal, the laryngeal regularly vocalized into *-a-. It would then analogically spread to other Celtic strong verb roots ending in sonorants in addition to the weak verbs, even if the root did not originally end in a laryngeal.[21]

There were also three verbs that did not use -(a)se-, instead straight-out taking thematised primary endings. Two of these verbs are *bwiyeti "to be, exist" (subjunctive *bweti) and *klinutor "to hear" (subjunctive *klowetor).[22]

Primary subjunctive formations in Proto-Celtic generally use the e-grade of the verb root, even if the present stem uses the zero-grade.

Imperative endings in Proto-Celtic were as follows:[14]^{: 147–148}

The second-person singular imperative was generally endingless in the active; no ending was generally added to athematic verbs. On thematic -e/o- verbs, the imperative ended in thematic vowel *-e. However, there is also another second-person singular active imperative ending, -si, which was attached to the verb root athematically even with thematic strong verbs.[23]

The thematic deponent second-person singular imperative ending was *-eso. The -the in Old Irish is secondary.[24][14]^: 140

Scholarly reconstructions [6][25][26][27] may be summarised in tabular format.

The copula *esti was irregular. It had both athematic and thematic conjugations in the present tense. Schrijver supposes that its athematic present was used clause-initially and the thematic conjugation was used when that was not the case.[28]