¹ It is a pleasure to record my gratitude to Professors Kenneth Jackson and Joshua Whatmough and to Abbé Falc'hun of Rennes University, who were kind enough to read this paper and who have saved me from some errors of fact. Shortcomings that remain are of course my own.

The following are the consonantal phonemes that I assume for Old Irish and Welsh at the stage of analysis prior to setting up morphophonemes. The phonemes are here grouped in structural sets; the phonetic labels are merely a matter of convenience. Parallel data for Breton are given in detail in §3 below.

For Old Irish: buccal stops ,/p t k b d g m/; buccal spirants /f θ x v ð γ ; ‘emphatic’ liquids and nasals (unquestionably separate, as phonemes, from the next group; probably geminate and tense) ; ‘non-emphatic’ liquids and nasal (perhaps short and lax) /r 1 n/; sibilant /s/; aspirate (rare and restricted) /h/, /m/ may well have had free variants [m] ~ [m̄], and could be classed distributionally with the emphatics; it is hard to decide on the basis of the defective and erratic orthography, /f v/ are bilabial oral spirants; is a nasalized bilabial spirant. I further write the following components, one of which always accompanies every consonant: advanced and raised tongue position ('palatalization') /'/; lip-rounding and (probably) protrusion ; low tongue position, unmarked (Though the last component is just as distinctive as the others, I choose this arbitrarily as the 'base quality', canceled when other marks are added.)

For Welsh: obstruents /p tkbdgfθxvð/; nasals ,/m n ŋ/; voiced liquids /1 r/; voiceless liquids /λ /; semivowels /j w/; sibilants /s š/; aspirate ,/h/. / λ/ (conventionally written ll) is a voiceless unilateral fricative; /r/ is a post-aspirated r, usually voiceless. These two phonemes might be regarded as clusters /lh rh/, in which case they would be grouped distributionally with the nasals; but since their distribution in certain respects is different from that of the nasals and since their morphophonemic behavior is unique, it is more convenient to set them up as separate phonemes. Furthermore, a juncture phoneme must be posited; unless this is assumed, we should need three additional phonemes to take care of the contrast in medial position with voiced nasal + /h/. For dialects heavily invaded by English loans at least one further phoneme occurs: /č/.

² This rare morphophoneme (< IE *sw-, *sp-), though phonetically indistinguishable from /s/, must be separated from the latter by virtue of the above morphophonemic behavior.

³ But the initial of the finite forms of ober ‘do’ (< *gw-) > zero.

⁴ In some dialects /gwr-/ = [grw]; in that case, nonetheless, /-L gwr-/ = /vr/.

⁵ ř approximately as in Czech. All labial spirants in Irish are bilabial.

⁶ Professor Jackson calls to my attention that the unaccented a's of such forms were already no doubt reduced to phonetic shwa.

⁷ The deuterotonic forms of the verb keep us from writing the /g/ here as /Nk/. Synchronically, there are of course ambiguous cases, for example in isolated nouns.

⁸ For the first member, /LbəN/ would seem to do just as well; synchronically one cannot tell from the isolated form, but the oppositions /mi/, /ti/: /məví/, /tədí/: /-m/, /-θ/: /vəN/, /dəL/ (all 1st and 2nd singular pronominal forms) make the analysis with /m/ preferable; historically the consonant is *m.

⁹ /orS/ (morpheme alternant of /on/ ~ /o/, which alternate like the articles) spirantizes k and s only.

¹⁰ Among the Welsh vowel phonemes, I use: /i/ = [i] in all positions; /ə/ = [ə] in all positions (a rare phoneme); /y/ = [ï] in final syllables, [ə] elsewhere. The allophones of the last-mentioned overlap with other phonemes, but it is set up to avoid the necessity of stating countless morpheme alternants. Thus /gwly:b/ = [gwlï:b].

^10a Of the masc. pl. forms, those in the left-hand column are non-personal, those in the right-hand column are personal.

¹¹ By virtue of a general allophonic situation in Breton (see §3), before stressed V the n of the article or the initial n- or 1- of the noun is always phonetically geminate.

¹² To indicate OIr. consonant quality I use the simultaneous phonemes: ' for palatal, for rounded quality. 'Broad' or velar quality requires no mark. See fn. 1.

¹³ Here /'/ is regarded as a simultaneous palatalizing phoneme, thus reducing Miss Sjoestedt's stock by half.

¹⁴ The writer here records his great obligation to Abbé Falc'hun's learning; without his phonetic data one cannot hope, on the basis of the existing literature, to come close to the truth on Breton phonology. The articles referred to appeared in a Breton quarterly Sav, 1943–4 (Kemmaduriou kensonennou ar brezoneg: 26.8–21, 27.70–9, 28.58–63, 29.35–42, 30.56–66, 31.40–8); Abbé Falc'hun also makes reference to these phenomena in his monograph La langue bretonne et la linguistique moderne, ed. Librairie Celtique, esp. 43–4 (Paris, 1943).

^14a As regards phonetic quality these symbols are self-explanatory, except perhaps for the following: ' marks palatalization; φ is a voiced labiodental spirant, as opposed to voiceless f, and differing from v in being much more forcibly released; η is a voiced spirant varying freely between voiced [h] and a velar spirant; gemination is double length (although we shall soon see that it is simpler to regard this as normal length, and the corresponding single elements as half-length).

¹⁵ Sav 29.39.

¹⁶ Sav 26.16.

¹⁷ Sav 26.13–4.

¹⁸ Sav 2775, 30–59.

¹⁹ Sav 27.74.

²⁰ Sav 29.41.

²¹ Using the morphophonemes, it is then simpler tu write the phonetically geminate initials with single symbols, and to regard the non-geminate lenition products as half-length.

²² Sav 30.60. [δ] in this situation varies to [ç], the German ich-sound.

²³ Sav 29.39.

²⁴ Sav 30.60.

²⁵ Sav 29.42.

²⁶ Sav 30.60.

²⁷ See T. F. O'Rahilly, Scottish Gaelic studies 3.111 ff.

²⁸ Sav 30.60.