Where it all began
To know where we're going, you should know where we've been. Primitive Irish, also called Archaic Irish in some texts, was the state of the language before Old Irish. It is known primarily through ogham (also <ogam>) inscriptions. (For reasons we'll get into later, <ogham> (and <ogam>) is /ɔɣam/ Old Irish and /oːmˠ/ in modern Irish; whatever you do, don't pronounce a [g]. You noticed the lack of explicit secondary articulation in my Old Irish transcription? Maith thú, as we say. It isn't a typo. We'll get to that.) The ogham inscriptions have given us a good grasp on Primitive Irish phonology. Ogham inscriptions are usually transliterated using captial letters. Following McManus' work, here's a round-up, in the traditional alphabetic order:
/b l w s n/ <B L F S N>
/j d t k kʷ/ <H D T C Q>
/m g gʷ ts† r/ <M G NG Z† R>
/a o u e i/ <A O U E I>
/ea oi ui io p ae/ <EA OI UI IO P AE>
†Regarding the letter straif: there is a lot of debate over the exact value of this letter. Much of that relates to issues surrounding Proto Celtic and Gaulish. The affricate /ts/, or allophonically /dz/ or maybe /st/, is often cited as a good candidate, alongside /sʷ/ and any other sibilant-dental-fortis interpretation of which you can imagine.
The last row was added later and was used to deal with, mostly Latin, loan words. The value of some of those signs changed over time, but we won't get into that right now. What you should note, however, it the moderate size of the consonant inventory. The whole language is rather Latinate looking at this stage of the game, eg: LIE LUGNAEDON MACCI MENUEH. This is about to change. Oh, how it is about to change.
Old Irish Orthography and Phonology: Introduction
Around 600 CE or so, the written language changed overnight. It's likely that a shift in the balance of power saw the written language catch up with the spoken language as Christian monks and scribes began to become prolific authors. This stage of the language constitutes Old Irish (also called: Goídelc, SenGoídelc, Sean-Ghaeilge, etc). Early Old Irish starts around 600 CE; Classical Old Irish, 700 CE; and, Middle Irish, 1100 CE. At this time, the language began to be written with the Latin alphabet. This posed some problems. Depending on your analysis, Latin itself has about 31 phonemes and 23 letters. As we all know far too well, English has roughly 44 phonemes and 26 letters. The Irish scribes were intent on making due with 18 letters for 66 phonemes. Why? How? Let's look at what happened.
Between Primitive Irish and Old Irish, a lot of changes happened. In particular, some rather extreme apocope caused the phonemicisation of palatalisation and lenition. It's fascinatingly mad. It's "let's throw out every even numbered vowel" mad. (If anyone would care for a summary, I'd be happy to draw up a sketch.) With all this, we go from the rather sedate 15 consonants of primitive Irish to an astounding 45 in Old Irish. Each consonant appears in palatalised, non-palatalised, lenited, and non-lenited forms.
A worthy aside, here. Modern Irish has palatalised forms standing against velarised forms. Scottish Gaelic contrasts palatalised and plain forms. Which is Old Irish doing? The answer is a little unclear. Thurneysen, who drew up the seminal grammar back at the beginning of the 20th century, proposed a three way (palatalised, plain, and velarised) contrast. This is no longer the prevailing view. Most of the literature notes the palatal and non-palatal contrast, interpreting non-palatal as plain, if the nature of non-palatal is mentioned at all. Make of it what you will.
Right, so 15 to 45 consonants, and don't forget your vowels. How are we going to manage this? A stroke of brilliance. Or insanity.
Old Irish: Consonants
Let's get down to business. BBCode is no good for making tables, so this is in paragraphs.
Modern conventions first:
Each consonant comes in a few varieties. It can be palatalised, noted with a superscript, ʲ, in IPA and an apostrophe or prime mark, ', in many Celtic studies contexts. The non-palatalised forms are not marked. Consonants can also be lenited. I'll use the standard IPA transcriptions, but there are some other ones you'll run across if you go searching elsewhere. Commonly, you may see the use of Greek letter for the lenited counter parts, eg a lenited /m/ as <µ>, which, of course, is pronounced something like /ṽ/, but might also be reckoned as /β/, which you really shouldn't confuse with <β> as the transcription for lenited <b>. The modern way isn't much better than the old one. You may also run into the use of capital letters. In this convention, a lenited n is <n> or <ν> (Greek nu) and a non-lenited (fortis) n is <N> or <n:> or <n>. Clear as mud. Alright, back to what the Old Irish scribes cooked up themselves. Here is a guide for lenited vs. not lenited. Palatalisation is being ignored for the moment.
<p c t>
- Word Initial: <p t k> is /p t k/, probably with non-phonemic aspiration before vowels
- Between or word-finally after vowels: /b d g/
- In word medial consonant clusters: usually /b d g/, but there are exceptions.
- After <r, l, n> word medially or word finally: either /p t k/ or /b d g/. Make it up as you go along. <derg> /dʲeRg/, but <derc> could be either /dʲeRg/ or /dʲeRk/. <anta> was /aNta/, but <antae> was /aNde/. Similarly, <altae> is either /aLte/ or /aLde/. Of course, all these words have different meanings.
- When doubled <pp tt cc>: /p t k/. This was the most common way to indicate that the stop should be voiceless even though it was intervocalic. This does not represent a geminate. However, this rule isn't hard and fast. /mak/ may be spelt both <mac> and <macc>; /atəx/, <atach> or <attach>. Annoyingly, these sometimes also stand for their voiced counterparts: <becc> /bʲeg/ comes to mind (cf. ModIr beag).
- Word Initial: <b d g> is /b d g/
- Between and word finally after vowels: /v ð ɣ/.
- In word medial consonant clusters: usually /v ð ɣ/, but there are exceptions.
- After <r, l> word medially or word finally: usually /v ð ɣ/, but they could also be /b d g/.
- When doubled <bb dd gg>: always /b d g/. Again, these are not geminates, but, rather, ways to indicate that the sound isn't the lenited counterpart.
This one is easy. <ph th ch> is /f θ x/
<f> is almost always /f/. However, because mutations weren't always explicit, especially in older texts, it can also be /v/ or ∅. Eg: <féil> /fʲe:lʲ/ "feast", <a féil> /a vʲe:lʲ/ "their feast", <a féil> /a e:lʲ/ "his feast". How did that happen? Ask for a diachrony post.
<s> is usually /s/. Word medially, it is usually spelled <ss>. When not doubled, it sometimes stands for /h/ word medially and as the result of mutation. Sometimes, <x> is encountered; it stands for the cluster <chs> /xs/, not /ks/.
<r l n>
Ok, now we're in the muck. There are lenited and non-lenited varieties of these three. Various dialects of Irish still distinguish between them sometimes, but in different ways. The exact values of the non-lenited and lenited sonorants aren't really clear. Most modern scholarly texts use <R L N> for the fortis versions and <r l n> for the lenis ones. Rather than start speculating, I'm going to break down and do the same. This will be my one non-IPA concession.
- Word Initial: <r l n> is /R L N/
- In clusters: Before /t d s l r/ or after /s l r n/, <r l n> is /R L N/.
- Between vowels and word finally after vowels: <r l n> is /r l n/.
- When doubled <rr ll nn>: always /R L N/. These are not geminates.
- Leniting mutations cause <r l n> to be /r l n/ when they would otherwise be /R L N/. Of course, none of this is adequately indicated anywhere.
- Word Initial: <m> is /m/
- Between vowels and word finally after vowels: <m> is /ṽ/.
- When doubled <mm>: always /m/. Not a geminate. There are no geminates.
- Syntactical leniting conditions also cause <m> to be /ṽ/.
Well, at least the monophthongs are straightforward! The Old Irish vowel system is actually rather typical. There are five vowels, plus schwa, and phonemic length: /i i: e e: a a: o o: u u: ə/. It's likely that there were two varieties of long front mid-vowels, but we'll pass over that detail.
How are our diphthongs looking? It varies by time period, but, try: /ai̯ oi̯ ui̯ au̯ ou̯ a:u̯ o:u̯ eu̯ e:u̯ iu̯ i:u̯ ia̯ ua̯/ on for size (based on Stifter's grammar). Old Irish was very permissive of hiatus. Modern Irish has collapsed this, but it's retained in Gaelic. From Pokorny's grammar, I list the allowed hiatus combinations: /o.a o:.a o:.a: o.a: i.a i:.a i.a: i:.a: i.e i:.e i:.e: i.e: i.o i:.o i:.o: i.o:/ and "e: o: or u: (when from an u diphthong or old Celtic a, a:, o + u) + any vowel". Very permissive. Very very permissive. We're going to use five graphemes and an acute accent mark. We're also going to mark the palatal quality of a consonant based on the adjacent vowel. Ready?
Monographs of Monophthongs
<a e i o u> /a e i o u/
<á é í ó ú> /a: e: i: o: u:/
A consonant following a front vowel, /i i: e e:/, was considered as palatalised.
Digraphs of Monophthongs
The palatal quality of the consonant is indicated by the adjacent vowel. This poses a problem for words like /o:rʲ/ because you want to juxtapose /o:/ and /rʲ/, but need to indicate that the <r> is actually /rʲ/. The solution is to insert 'silent' vowels.
The most common of these was <i>. Consider: <ór> /o:r/ and <óir> /o:rʲ/, <ben> /bʲen/ and <bein> /bʲenʲ/. Also note the seemingly superfluous additions like <magen> /maɣʲən/ and <maigen> for /maɣʲən/ which begin to resemble the modern spelling rule caol le caol, leathan le leathan which calls for front vowels to flank palatalised consonants. This insertion of <i> results in the following digraphs for monophthongs and palatalised consonants: /aCʲ eCʲ oCʲ uCʲ a:Cʲ e:Cʲ o:Cʲ u:Cʲ/ <aiCʲ eiCʲ oiCʲ uiCʲ áiCʲ éiCʲ óiCʲ úiCʲ>. Remember, these aren't diphthongs. The story of <i> gets more complicated. It doesn't always indicate palatalisation, especially in monosyllables: <mind> is both /mʲiNd/ or /mʲiNʲdʲ/. Modern Irish has inserted an <o> to make things clearer: OIr <mind> is ModIr <mionn>. Of course.
What about at the end of a word? Well, there you have your choice. The caol le caol, leathan le leathan flanking rule hasn't taken hold, so a word final <i> or <e> doesn't palatalise it's preceding consonant. Happily, an <a> is sometimes inserted to remind us of this. You don't pronounce the <a>. That would be silly. <dalte> <daltae> are both /daLte/. To indicate a final back vowel after a palatalised consonant, <iu ea eo> was used for word final /u a o/.
The Reduced Vowel (ə)
Like in modern Irish, non-stressed vowels reduced to ə. The orthographic choice depended on which of the surrounding consonants you wanted to palatalise:
<CaiC CiC> /CəCʲ/
These guidelines were ignored if etymology was considered to be important. They were also ignored around labial sounds. Instead of reducing to /ə/, /u/ sometimes stayed in a series of complications into which I won't delve at this juncture.
Here are the standardised conventions. Which one is used is ... sort of random:
/oi̯/ <oí óe>
/ai̯/ <aí áe>
/a:u̯/ <áu áo>
/e:u̯/ <éo éu>
Everything else is in hiatus. Everything. As in Latin, many scholarly transcriptions mark this with a trema.
I was Promised <h>'s
Even the Irish monks couldn't stand their own orthographic system. They started explicitly marking the lenited consonants, rather than just relying on word position or "everyone knows". This was done with a dot over the letter, also called a ponc séimhithe or a punctum delens if you prefer. It looks like this: ḟ ṡ. In Old Irish, when it's used at all, it's only found on the <f> and the <s>. The lenited counterparts of <p t c> were written with the <h> following them: <ph th ch> /f θ x/. (Recall that the lenited forms of <b d g> weren't expressly indicated.) This combination dot and <h> system resulted in the generation of two whole systems. One exclusively used the ponc séimhithe, and the other, the <h>. The short story is that, because of imported typewriters, the <h> version won. Thus, modern Irish (and Gaelic) use Ch to mark the lenited counterpart of a consonant C. Well, most of the time. The vowel-flanking rule is still in effect for indicating palatalisation.
Why <h> at all? Why not <q>? A good question without a good answer. Personally, I think it might have something to do with an intuition regarding aspiration and spirantisation.
Putting it all together
Here's your final phonology:
/N Nʲ n nʲ/
/R Rʲ r rʲ/
/L Lʲ l lʲ/
/p pʲ f fʲ/
/t tʲ θ θʲ/
/k kʲ x xʲ/
/b bʲ v vʲ/
/d dʲ ð ðʲ/
/g gʲ ɣ ɣʲ/
/s sʲ h hʲ/
/a e i o u ə/
/a: e: i: o: u:/
/ai̯ oi̯ ui̯ au̯ ou̯ a:u̯ o:u̯ eu̯ e:u̯ iu̯ i:u̯ ia̯ ua̯/
<Boí rí amrae for Laignib, Mac Dathó a ainm. Boí cú occo. Imdíched in cú Laigniu uili. Ailbe ainm in chon, ocus ba lán Ériu dia airdircus in chon.>
/boi̯ rʲi: aṽre for Laɣʲnʲəvʲ, mak daθo: a aNʲmʲ. boi̯ ku: ogo. imʲdi:xʲəð in ku: Laɣʲnʲu hulʲi. alʲvʲe aNʲmʲ iN xon, ogus ba La:n e:rʲu dʲiä arʲðʲəRʲkʲus in xon/
Edited for some typos.
Also, another note on /h/:
/h/ Isn't Real (and neither are word initial geminates)
/h/ isn't real? This seems to have been the prevailing view of the time. It isn't without merit. The citation forms of native words don't start with /h/ and the sound isn't well tolerated word medially. Recall that <h> was being used to mark a lenited consonant, not make an [h] sound. In line with their "everyone knows" policy, the result of h-prothesis was unmarked. H-prothesis is a grammatical sound change, similar to the lenition (séimhiú) and eclipsis (urú) mutations of the consonants. As with all the initial sound changes for which the Celtic languages are well know, it arose from the influence preceding proclitic. In general, it's triggered by words which end in a vowel, but don't otherwise cause lenition or eclipsis. In the example above, there is h-prothesis in the phrase <Laigniu uili> /Laɣʲnʲu hulʲi/, "all Leinster", even though it isn't indicated in the orthography. To confuse matters, the lack of <h> where there is an [h] occurs happens in the same environments where you might see what look like word initial geminates. For example, this kind of change is triggered by the possessive pronoun <a> /a/, "her", or the preposition <a> /a/, "out from": <ech>, /ex/, "a horse", and <a ech> /a hex/ "her horse", but also <medón> /meðo:n/ "a middle", and <a mmedón> /a meðo:n/ "from a middle".