2.3.1 Stress placement and non-iterativity

For most dialects of Modern Irish (those in the Northern and Western groups), stress placement is straightforward: excluding a few lexical exceptions, stress falls on the first syllable of the word (Ó Sé Reference Ó Sé2008 among many others). In these dialects, stress is not conditioned by syllable weight:

Following Doherty (Reference Doherty, Mester and Robbins1991) and Green (Reference Green1997) I take the rigidly initial stress system of Western and Northern dialects to reflect quantity-insensitive trochaic footing at the left edge of the word (e.g. [(ˈkrɑ.ɡər)lɑːn ]).

There is no evidence of secondary stress in most varieties of Irish. One exception is Munster Irish, a full analysis of which lies outside the scope of this paper (but see Sections 2.4.2 and 2.7). The lack of secondary stress suggests that footing is non-iterative in most dialects of Irish – content words contain only a single foot. Since I know of no positive evidence for iterative foot structure in the language (apart from secondary stress in Munster Irish), I will assume non-exhaustive footing without further comment (see Ní Chiosáin Reference Ní Chiosáin, van der Hulst and Ritter1999, Bennett Reference Bennett2012 for evidence supporting this view).

2.3.2 Modeling Irish stress: OT analysis

The initial stress system of Northern and Western Irish can be modeled with four constraints: AllFeetLeft (afl), Trochee, Weight-to-stress (wsp), and Parse( ${\it\sigma}$ ) (Green Reference Green, Dobrin, Singer and McNair1996, Reference Green1997; see Prince & Smolensky Reference Prince and Smolensky1993/2004; McCarthy & Prince Reference McCarthy and Prince1993; McCarthy Reference McCarthy2003, Reference McCarthy2008a for standard constraint definitions). I assume that headedness( ${\it\omega}$ ), the constraint requiring every prosodic word ${\it\omega}$ to contain at least one foot (its head), is inviolable (i.e. it is part of Gen; Itô & Mester Reference Itô, Mester, Honma, Okazaki, Tabata and Tanaka1992/2003; Selkirk Reference Selkirk, Beckman, Walsh Dickey and Urbanczyk1995, etc.).

To capture the leftward orientation of stress, I take afl to be undominated. The ranking {afl, Trochee}  $\gg$ wsp (8b, c) ensures that stress placement will be quantity-insensitive, and afl  $\gg$ Parse( ${\it\sigma}$ ) (8d) derives non-iterative footing. Although low-ranked, Parse( ${\it\sigma}$ ) still eliminates candidates with monosyllabic feet (8e).

It should be noted that the winning candidate in (8), [(ˈkɨ.n^jiː)n^jiː ], contains an uneven (ˈLH) trochee with a heavy syllable in the weak position of the foot. In many quantity-sensitive languages, feet of this shape are actively avoided (Hammond Reference Hammond1986, Hayes Reference Hayes1995, etc.). In Section 2.5 I will argue that (ˈ ${\it\sigma}$ H) feet are indeed avoided in Irish plural allomorphy, despite the general acceptability of (ˈ ${\it\sigma}$ H) feet in the language.

While it seems uncontroversial that long vowels and diphthongs should count as heavy (i.e. bimoraic) in Irish, the question of coda weight is more fraught. Word minimality restrictions cannot be used as a diagnostic for coda weight because Irish freely allows monomoraic content words like rath [ˈrɑ ] ‘luck’ and te [ˈt^je ] ‘hot’ (Stockman Reference Stockman1974: 38, 65; Green Reference Green1997: 64; Ní Chiosáin Reference Ní Chiosáin, van der Hulst and Ritter1999: 572). In dialects with strict initial stress (7), stress placement simply provides no evidence as to the weight of coda consonants. However, in the quantity-sensitive stress system of Munster Irish (Section 2.4.2), codas are inert for stress assignment (‘only those syllables containing a long vowel or diphthong count as heavy’, Doherty Reference Doherty, Mester and Robbins1991: 19; also Ó Sé Reference Ó Sé1989, Reference Ó Sé2008). This suggests that codas are weightless in Munster Irish; I assume that the same is true for all dialects of the language, given the absence of evidence to the contrary. Nevertheless, in Sections 2.4 and 2.5 I argue that consonants can be moraic in Irish under very specific circumstances, and that consonant weight influences plural allomorph selection.Footnote ^[11]

2.4.1 Vowel reduction in Ulster and Achill Irish

Unstressed short vowels typically reduce to [ə] in Irish (or to [ɨ]/[ɪ] between palatalized consonants). Short vowel reduction is a fairly old feature of the language, and occurs in all dialects (Thurneysen Reference Thurneysen1946: 29; O’Rahilly Reference O’Rahilly1932: 110; Ó Sé Reference Ó Sé1989, and many others). However, in many Ulster (Northern) dialects of Irish, unstressed [ax] sequences do not undergo reduction to [əx] (Quiggin Reference Quiggin1906: 9; O’Rahilly Reference O’Rahilly1932: Chapter XIV; Ó Sé Reference Ó Sé1989: 167; Ó Dochartaigh Reference Ó Dochartaigh1987: Chapter 4; Ó Siadhail Reference Ó Siadhail1991: 33, etc.). The same pattern of (non-)reduction occurs in Achill Irish (9).

Strings of the form /ax/ thus pattern with long vowels in resisting vowel centralization in unstressed positions. The implication is that /ax/ sequences are more phonologically prominent than other /aC/ or /Vx/ strings.

This interpretation of (9) is challenged by apparent instances of unstressed [a] in other phonetic contexts. It has been claimed that the long vowels are realized as short [a] when unstressed in Ulster Irish, e.g luchóg [ˈlu.haɡ ] ‘mouse’ (Ó Siadhail Reference Ó Siadhail1991: 80) and amhráin [ˈo.ɾan ] ‘song’ (Ní Chasaide Reference Ní Chasaide1995). Such examples might suggest that unstressed short [a] is actually permitted in these dialects. If so, these derived instances of unstressed [a] would undermine the claim that words like (9a–e) lack reduction because of the inherent prominence of [ax] strings.

While the Ulster $\rightarrow$  [a] merger is often portrayed as a categorical neutralization (e.g. Quiggin Reference Quiggin1906: 5–17), there is in fact phonetic evidence for a surface distinction between. Descriptive sources for West Ulster Irish often transcribe unstressed as long or half-long [a˙] (Wagner Reference Wagner1959: 88–91; Sommerfelt Reference Sommerfelt1922: 122–123; Wagner Reference Wagner1969; Ó Dochartaigh Reference Ó Dochartaigh1987: Chs. 2, 4).Footnote ^[12] For Achill Irish (9) the shortening of unstressed actually depends on phrasal position, giving rise to overt alternations like coileán  ${\sim}$ [ˈkɨ.l^jɑn ] ‘pup’ (Stockman Reference Stockman1974: 310). The sequence [ax], in contrast, is consistently transcribed with a short [a] (Stockman Reference Stockman1974: 307–308; Ó Dochartaigh Reference Ó Dochartaigh1987: Chapter 4).

The punchline is that the merger of unstressed and /a/ may be only partial, such that the underlying vowels are still distinguishable on the surface despite some degree of overlap in their phonetic realizations. This is especially true for unstressed , which encompasses not just [a(ː)] but also mid-back realizations like [o ɔ]. Assuming that these phonetic differences are indicative of a surface phonological distinction between the vowels in question (e.g. Benus & Gafos Reference Benus and Gafos2007), these findings are consistent with the view that West Ulster and Achill Irish have a phonological prohibition against unstressed short [a] except in the context of a following [x].

To summarize, in West Ulster and Achill Irish [ax] strings are exempt from an otherwise general process of short vowel reduction in unstressed syllables. This resistance to reduction suggests that [ax] sequences are prosodically more prominent than other [aC] or [Vx] sequences in these dialects (e.g. Green Reference Green, Dobrin, Singer and McNair1996, Reference Green1997). The facts are different in East Ulster Irish: here, long seem to fully reduce to short [a] when unstressed, and unstressed /ax/ further reduces to [əx] (Ó Dochartaigh Reference Ó Dochartaigh1987).Footnote ^[13] I postpone further discussion of East Ulster varieties until Section 2.7.1.

2.4.2 Munster Irish stress

Further evidence for the prosodic prominence of [ax] strings can be gleaned by examining the stress patterns of Munster Irish, a dialect group located in southern Ireland. Like Irish more generally, Munster Irish has default word-initial primary stress. However, Munster Irish differs from most dialects in that the stress system is quantity-sensitive (Table 3). The basic descriptive generalization is that primary stress falls on the leftmost syllable containing a long vowel or diphthong within an initial three-syllable window, otherwise on the first syllable.Footnote ^[14]

As mentioned in Section 2.3.2, coda consonants do not generally contribute to syllable weight in Munster Irish. There is, however, one exception: in the absence of long vowels, primary stress will fall on a syllable in second position.

Table 3

Quantity sensitivity in Munster Irish (Ó Siadhail Reference Ó Siadhail1991: 129–130; Doherty Reference Doherty, Mester and Robbins1991: 20–21).

The stress-attracting character of [ax] is supported by synchronic alternations like nead [ˈn^jad ] ${\sim}$ neadacha [n^jɪˈdaxə ] ‘nest(s)’ and cheannaigh [ˈx^janiɡ^j ] ‘(s)he bought’ ${\sim}$ ceannach [k^jəˈnax ] ‘buying’ (Ó Cuív Reference Ó Cuív1944: 77, 105; Ó Sé Reference Ó Sé2000: 89, 104).Footnote ^[15]

The Munster Irish stress system thus relies on the ternary weight distinction {  (Doherty Reference Doherty, Mester and Robbins1991, Green Reference Green, Dobrin, Singer and McNair1996, and references there). Only the combination of [a] and [x] draws stress rightward; other [aC] and [Vx] strings do not attract stress. The special prominence of [ax] sequences must therefore be due to the joint influence of [a] and [x] in contact.Footnote ^[16]

There is thus convergent evidence that [ax] strings are more phonologically prominent than corresponding [aC] or [Vx] strings in Irish. What remains to be explained is why [ax] shows this unusual constellation of properties. In the next section, I argue that the sequence [ax] is phonologically prominent in Irish because [x] bears an independent mora when following [a].

2.4.3 Syllabification

Besides the stress-attracting properties of [ax] in Munster Irish, and its resistance to reduction in West Ulster and Achill Irish, Doherty (Reference Doherty, Mester and Robbins1991) claims that [ax] also behaves exceptionally with regard to syllabification: in a surface [axV] string, intervocalic [x] is parsed into the same syllable as the preceding [a].Footnote ^[17]

Strong evidence for backwards syllabification of [x] with [a] comes from the observation that [ax] sequences attract stress in Munster Irish even when followed by a vowel, as in oideachas [ə.ˈd^jax.əs ] ‘education’ (Hickey Reference Hickey2011: 312). Under the generally accepted view that stress is a property of syllables (Liberman & Prince Reference Liberman and Prince1977, Hayes Reference Hayes1995, etc.), it would be deeply surprising to find that hypothetical onset [x] in an [a.xV] string was responsible for attracting stress to a preceding heterosyllabic [a]. Given that onset consonants do not usually contribute to syllable weight – much less to the weight of the preceding syllable – such a pattern would be all the more striking. As hypothesized in Doherty (Reference Doherty, Mester and Robbins1991: 28n), these observations imply that [x], under some narrowly circumscribed conditions, counts as a moraic coda consonant in Irish (see also Noyer Reference Noyer1990). (I will return to the question of why [x] bears a mora specifically when preceded by [a].)

The additional fact that unstressed [ax] resists centralization in Ulster Irish (9) can be captured under the assumption that vowel reduction only targets monomoraic syllables in these dialects. If [ax] strings always form a bimoraic rhyme , it then correctly follows that they should be exempt from reduction, just like long vowels. This is so even when the [x] is intervocalic 

I conclude that there is credible evidence that intervocalic [x] in an [axV] string is moraic and tautosyllabic with the preceding [a]. All subsequent transcriptions will reflect this syllabification.

Doherty’s (Reference Doherty, Mester and Robbins1991) claim is not as radical as it perhaps seems: there is good empirical evidence for backward syllabification both in Irish and in other languages. Impressionistic and experimental studies of Irish are in rough agreement that intervocalic consonants are parsed as codas when following a stressed short vowel, [ˈCVC.V ] (see Green Reference Green1997; Dalton & Ní Chasaide Reference Dalton and Ní Chasaide2005, Reference Dalton, Ní Chasaide, Gussenhoven and Riad2007; Ní Chiosáin, Welby & Espesser Reference Ní Chiosáin, Welby and Espesser2012 on Irish; Clements Reference Clements, Bogers, van der Hulst and Mous1986; Ladefoged, Ladefoged, Turk, Hind & Skilton Reference Ladefoged, Ladefoged, Turk, Hind and Skilton1998 on Scottish Gaelic and Cohn & McCarthy Reference Cohn and McCarthy1998: Section 3; Bennett Reference Bennett2012: 221–222 for cases outside Celtic). I conclude that backwards syllabification, though typologically rare, is nonetheless attested and therefore plausible in the case of [ax.V] strings (especially when stressed [ˈax.V]). Bennett (Reference Bennett2012: 216–222) provides some additional arguments that backward syllabification might be especially favored for intervocalic [x] in Irish; those arguments are omitted here for reasons of space.

Still unexplained is why [a], but no other vowel, behaves as prosodically prominent when preceding [x]. There is good reason to believe that this is a non-accidental fact. In particular, I would suggest that Irish [x] is phonologically a glide-like counterpart of [a]. If this is correct, then [ax] sequences are quasi-diphthongal – a structural analysis that explains why [ax] strings pattern with true diphthongs for stress attraction in Munster Irish and vowel reduction in Ulster Irish.

I have in mind here a parallel between the behavior of [ax] rhymes and the distribution of post-vocalic [ɹ] in various varieties of English. In dialects of English with ‘intrusive r’, the approximant [ɹ] is inserted in hiatus environments whenever the first vowel is one of [a ɔ ə] (e.g. McCarthy Reference McCarthy1993 and references therein).

Several authors have pointed out that intrusive [ɹ] is the glide counterpart of the non-high back vowels [a ɔ ə] – precisely those vowels that license its appearance (Kahn Reference Kahn1976, Gick Reference Gick1999, Baković Reference Baković1999, Itô & Mester Reference Itô, Mester, Grijzenhout and Kabak2009). This observation has led many of those same authors to propose that intrusive [ɹ] is not epenthetic in a strict sense, but rather represents the breaking of a vowel in hiatus into a vowel–glide–vowel sequence. The choice of [ɹ] as the intrusive segment then follows from the fact that [ɹ] is roughly homorganic with [a ɔ ə], the vowels that provide its source.

The proposed analogy is this: the glide counterpart of [a] is [ɹ] in English, but [x] in Irish.Footnote ^[18] Evidence for this view comes from the featural composition of [a] and [x]. Under standard feature theories these two segments share at least some place features, most notably [dorsal] and [+back] (e.g. Sagey Reference Sagey1986). Furthermore, some dialects of Irish realize velarized /x/ as the uvular fricative [χ] (e.g. Sommerfelt Reference Sommerfelt1922: 72; Breatnach Reference Breatnach1947: 40–41; Ó Curnáin Reference Ó Curnáin2007: 171, 408–414; Bennett, McGuire, Ní Chiosáin & Padgett Reference Bennett, McGuire, Ní Chiosáin and Padgett2012, and fn. 6). It is therefore plausible that Irish /x/ shares the feature [-high] with /a/ as well. There is a real sense, then, in which [x] and [a] are phonologically homorganic.

For concreteness, I offer the following proposal. I assume that there is a weak bias against onset [x] in Irish, as suggested by the fact that word-initial [x] is always morphologically derived, never underlying (Ní Chiosáin Reference Ní Chiosáin, van der Hulst and Ritter1999; see Bennett Reference Bennett2012: 216–222 for details). This bias can be encoded as a violable markedness constraint, *Onset/[x] (Smith Reference Smith, Edwards, Midtlyng, Sprague and Stensrud2008, Flack Reference Flack2009). Second, I assume that *Diphthong penalizes any complex nucleus in which V and ${\it\beta}$ have differing dorsal specifications (Casali Reference Casali1996, Rosenthall Reference Rosenthall1997). Importantly,  nuclei do not violate this constraint: [a] and [x] are phonologically homorganic, sharing at least the features [dorsal,+back]. (As [high] and [low] are not contrastive for dorsal consonants in Irish, I remain agnostic regarding the specification of these features on [x]; see McCarthy Reference McCarthy and Keating1994; Dresher Reference Dresher2009; Backley Reference Backley2011: Section 3.6.)

In general, the markedness of onset [x] is not sufficient to trigger backward syllabification of intervocalic [x]: the drive to avoid codas and derived diphthongs outweighs the bias against onset [x] (14a). However, following [a], backward syllabification of [x] is permitted because [a] and [x] are homorganic: [x] is parsed as an offglide to [a], (14b). Like all complex nuclei in Irish, then counts as bimoraic. As *Onset/[x] is inactive for consonants other than [x], the default pattern of intervocalic syllabification will still be [V.CV] (14c).

Finally, to prevent other dorsal consonants from syllabifying with [a] (e.g. I assume first that obstruents are prohibited from syllable nuclei in Irish (*[-son]/Nuc is undominated; Prince & Smolensky Reference Prince and Smolensky1993/2004; Zec Reference Zec1994), and second that [x] is phonologically an approximant and thus exempt from this restriction (see Ó Dochartaigh Reference Ó Dochartaigh1987: Chapter 6 and appendices for possible supporting evidence within Irish, and Martínez-Celdrán & Regueira Reference Martínez-Celdrán and Regueira2008 on Western Romance).

One last word is in order regarding the exceptional prominence of [ax] rhymes. I have argued that the phonological prominence of /ax/ sequences can be attributed to the fact that such strings are quasi-diphthongal. This might lead us to expect that other homorganic [VC] sequences, such as [uv] and [ij], should behave similarly with respect to stress assignment, vowel reduction, etc. They do not. This contrast in prominence may be due to the relatively high sonority of low /a/. In phonetic terms, the low vowel /a/ has higher intrinsic duration and overall intensity than non-low vowels (e.g. House Reference House1961, Lehiste Reference Lehiste1970, Gordon Reference Gordon2006, Parker Reference Parker2002). In some languages (e.g. Gujarati) the high sonority of /a/ manifests itself phonologically through the preferential stressing of syllables that contain /a/ (Kenstowicz Reference Kenstowicz1997, de Lacy Reference de Lacy2002b, etc.). I suspect that something similar is occurring with /ax/ sequences in Irish. The phonetic salience of [a], along with the additional duration contributed by homorganic [x], might encourage speakers to treat [ax] strings as phonologically heavy (though not as heavy as long vowels and diphthongs; see Blankenhorn Reference Blankenhorn1981: 236–239; Ó Sé Reference Ó Sé2008: 87–88, and fn. 20).Footnote ^[19] Whether the special contribution of [a] is best captured by reference to moras, sonority, or phonetic duration is a debate that would take the present discussion too far afield; see Blevins (Reference Blevins2006), Gordon (Reference Gordon2006), de Lacy (Reference de Lacy and de Lacy2007a), Ryan (Reference Ryan2011, Reference Ryan2014), and work cited therein for details.

In this section I have argued that there are robust empirical grounds for assuming backwards syllabification of [x] in an [ax] string, such that [ax] is parsed as a bimoraic rhyme . The importance of this conclusion lies in the fact that the plural suffix /-axə/ contains an underlying, potentially bimoraic /ax/ sequence, as shown in the next section. This observation provides the key to understanding the pattern of Irish plural allomorphy described in Section 2.1.

2.4.4 -(e)acha is underlyingly /-axə/

The two phonological hallmarks of underlying /ax/ are (i) resistance to unstressed vowel reduction in West Ulster and Achill Irish and (ii) stress attraction in Munster Irish. Beginning with West Ulster and Achill Irish, we find that the plural suffix -(e)acha is indeed realized as unreduced [-axə]:

Similarly, the first vowel of -(e)acha attracts stress in Munster Irish (16); when stressed, this vowel retains its underlying /a/ quality.

I conclude that /-axə/ is the underlying form of the plural marker -(e)acha. This suffix therefore has the potential to surface as , with a heavy–light weight profile. This distinguishes -(e)acha /-axə/ from the allomorphically related suffix -(e)anna /-ənə/, which contains only reduced, monomoraic underlying vowels. In the following section I argue that the stress-sensitive distribution of -(e)acha and -(e)anna follows from the fact that -(e)acha contains a phonologically prominent /ax/ string. In Section 2.7.1 I return to some problems posed by the reduction of unstressed /-axə/ to [-əxə] in Munster, Connacht, and East Ulster.

2.6.1 Monosyllabic nouns and wsp_Ft

As long as priority is undominated, nothing compels the appearance of the dispreferred allomorph /-ənə/ with monosyllabic noun stems: -(e)acha will wrongly be preferred across the board.

This dilemma can be resolved by positing a variant of wsp relativized to foot-internal positions. Any constraint that evaluates foot-internal unstressed heavy syllables as being more ill-formed than unfooted heavy syllables will prefer a candidate like [(ˈklo.ɡə)nə ] (with the non-default allomorph -(e)anna) over a default form like . I call this constraint wsp_Ft  :

The intuition behind wsp_Ft  is that (ˈ ${\it\sigma}$ H) trochees are more ill-formed than (ˈ ${\it\sigma}$ L) trochees because (ˈ ${\it\sigma}$ H) trochees contain a prominent, heavy syllable in the weak branch of a foot – an extremely non-prominent position (e.g. Prince Reference Prince, Deaton, Noske and Ziolkowski1991; Hayes Reference Hayes1995; Dresher & van der Hulst Reference Dresher and van der Hulst1998; Kager Reference Kager1999: 151; Gouskova Reference Gouskova2003; de Lacy Reference de Lacy2004, Reference de Lacy and de Lacy2007a; McCarthy Reference McCarthy2008b, and citations therein). Bennett (Reference Bennett2012) argues that wsp_Ft  is independently active in the phonology of Conamara Irish, though that dialect group is largely outside the focus of this paper (Section 2.7.1).

High-ranked wsp_Ft  correctly favors -(e)anna over -(e)acha in post-tonic position. We now understand the prosodic motivations for -(e)anna ${\sim}$ -(e)acha allomorphy: allomorph selection avoids ill-formed metrical feet.Footnote ^[22]

The special prominence of [ax] strings therefore interacts with reasonable constraints on prosodic structure to determine the distribution of plural allomorphs in West Ulster and Achill Irish.

Since wsp_Ft  is freely violated in monomorphemic words of Irish, it must be the case that wsp_Ft  is dominated by the same metrical markedness constraints that dominate wsp (I assume throughout that high-ranked FtBin rules out candidates containing degenerate feet, e.g. *[(ˈɡɤ)bɑːʃ.t^jə ]).

Although wsp_Ft  is too low-ranked to condition surface stress placement, the effects of wsp_Ft  are nonetheless manifest in plural allomorph selection. As with other cases of optimizing pcsa, the selection of -(e)anna with monosyllabic stems thus amounts to the emergence of the unmarked (McCarthy & Prince Reference McCarthy, Prince and Gonzàlez1994; Mascaró Reference Mascaró, Durand and Laks1996, Reference Mascaró2007).

2.6.2 Parse( ${\it\sigma}$ ) and syllabic binarity

The ranking wsp_Ft   $\gg$ priority $\gg$ wsp accounts for most of the facts. However, one piece of data remains intransigent under this ranking: noun stems consisting of a heavy monosyllable are incorrectly predicted to surface with /-axə/:

Candidate (24b) is correctly eliminated by wsp_Ft . However, candidate (24c) sidesteps this violation of wsp_Ft  by leaving the offending unfooted,  This candidate then wrongly emerges as the optimal output form, as Priority favors the /-axə/ allomorph.

The task, then, is to find a constraint that eliminates for containing a monosyllabic foot. A clear contender is Parse( ${\it\sigma}$ ), which will favor disyllabic footing under all circumstances (FtBin is another option, but under most formulations it does not distinguish [(H)H] feet from [(HH)] feet). Parse( ${\it\sigma}$ ) is relatively low-ranked in Northern and Western Irish: words contain just one left-aligned foot, and so many syllables are left unparsed. However, low-ranked Parse( ${\it\sigma}$ ) can still exert pressure to make that single foot disyllabic (Section 2.3.2). To eliminate candidates that avoid wsp_Ft  violations by leaving  unfooted, Parse( ${\it\sigma}$ ) must outrank priority:

The formerly problematic (25c) is eliminated by Parse( ${\it\sigma}$ ), as it fails to foot the post-tonic syllable. The crucial observation is that losing candidates either violate wsp_Ft , or violate Parse( ${\it\sigma}$ ) to a greater extent than the winner. Effectively, this prevents /-axə/ from attaching to monosyllabic stems:  plurals will always be ill-formed with respect to some aspect of foot parsing.

2.6.3 Noun stems with exceptional stress

Recall from Section 2.1 that -(e)anna(í) also attaches to polysyllabic nouns with exceptional final stress, such as [də.ˈɡ^jr^jiː- ənə] ‘degree(s)’ (Hickey Reference Hickey1985b; the example is from a Western dialect). I assume that final stress in most varieties of Irish corresponds to a prespecified, but non-initial trochee (but see Section 2.7 on Munster Irish).Footnote ^[23]

The fact that such nouns pluralize with /-ənə/ rather than /-axə/ follows directly: since post-tonic syllables are parsed as the weak member of a trochee whenever possible, /-axə/ will be dispreferred in post-tonic position whether or not the stressed syllable is word-initial.

2.6.4 Accounting for lexical exceptions

Some monosyllabic nouns exceptionally pluralize with /-axə/, despite ending in a stressed syllable.

I assume that different noun stems in Irish may be associated with different cophonologies, expressed as lexically specific rankings of priority and wsp_Ft  (e.g. Anttila Reference Anttila2002, Inkelas & Zoll Reference Inkelas and Zoll2007; see Bonet et al. Reference Bonet, Lloret and Mascaró2007, Pater Reference Pater and Parker2010 for alternative approaches). Exceptional monosyllabic nouns belong to a cophonology in which priority dominates wsp_Ft .

priority being undominated, this ranking will always select the default allomorph /-axə/ for these noun stems, even at the cost of violating wsp and wsp_Ft  (29a).Footnote ^[24]

One more lexical generalization needs to be captured. Nouns taking /-ənɪ/ always end in a stressed syllable, without exception (Section 2.1). This asymmetry falls out immediately from the ordering of allomorphs enforced by priority:

As long as priority consistently dominates wsp, no ranking of these constraints will force /-ənɪ/ to attach to an unstressed syllable. The general point is that an account of plural allomorphy in terms of priority can derive all and only the attested lexical exceptions to the basic distributions of -(e)anna and -(e)acha.

A reviewer worries that -(e)acha occurs with too many monosyllabic stems in Irish for such forms to be considered ‘exceptional’. Four points are worth emphasizing here. First, all detailed treatments of the Irish plural system agree that monosyllabic nouns taking -(e)acha belong to a minority pattern, and an unstable one at that (Section 2). Second, there is good evidence that morpho-phonological regularities (such as the avoidance of -(e)acha with monosyllabic stems) may be learned as productive, grammatically controlled patterns even in the face of numerous exceptions (e.g. Zuraw Reference Zuraw2010, Gouskova & Becker Reference Gouskova and Becker2013). Third, speakers appear to learn restrictions on inflectional allomorphy which are not straightforwardly predictable from lexical frequencies, but which may nonetheless reflect phonological well-formedness conditions (e.g. Becker, Nevins & Levine Reference Becker, Nevins and Levine2012; see also Marcus, Brinkmann, Clahsen, Wiese & Pinker Reference Marcus, Brinkmann, Clahsen, Wiese and Pinker1995, Pinker & Ullman Reference Pinker and Ullman2002). Finally, all analyses of these data must account for the fact that -(e)acha freely occurs with polysyllabic stems, and -(e)anna does not. This observation holds regardless of how often -(e)acha might also occur with monosyllabic stems. I conclude that even if exceptional monosyllabic forms with -(e)acha were especially frequent, that itself would not be a counter-argument against the OT analysis developed here.

2.7.1 Opacity and dialect variation

In Section 2.4 I argued that post-vocalic [x] counts as moraic just in case the preceding vowel is [a]. This claim was leveraged to account for stress-sensitive allomorphic variation between the plural suffixes /-ənə/ and /-axə/. In West Ulster and Achill Irish, where unstressed /ax/ is realized as surface [ax], the prosodic motivations driving allomorph selection are perfectly transparent. However, other varieties of Irish introduce a complicating factor. In most dialects spoken in Connacht (the West) and Munster (the South), underlying /ax/ surfaces as reduced [əx] rather than [ax] when unstressed (O’Rahilly Reference O’Rahilly1932: 109–110 among many others). This reflects a more general process, found in all dialects, which reduces unstressed short vowels to [ə] (or to [ɪ]/[ɨ], depending on consonantal context). In some surface forms containing -(e)acha, then, there is no [ax] sequence, and thus no motivation for assigning a mora to the intervocalic [x] in [-əxə].

This pattern of vowel reduction is problematic for the analysis of plural allomorphy proposed here. If reduced [-ə.xə] is preferable to its unreduced counterpart on general phonological grounds, nothing prevents [-ə.xə] – with a non-moraic [x] – from occurring with monosyllabic nouns (38). (For convenience I assume that the constraint driving the reduction of unstressed short vowels is McCarthy’s Reference McCarthy2008b , though nothing depends on this.)

In this way, the simultaneity of vowel reduction and affix selection incorrectly circumvents the use of the non-default allomorph /-ənə/. This is a clear case of morpho-phonological opacity: vowel reduction conceals the underlying quality distinctions that condition plural allomorphy. It follows that plural allomorphy in Munster Irish cannot be ‘output optimizing’ in the strictest sense, since the prosodic motivations behind allomorph selection are masked by the application of an independent process of reduction. This observation is consistent with the finding of Paster (Reference Paster2006: 143) that, cross-linguistically, prosodically conditioned suppletive allomorphy is ‘sensitive to input elements, not surface elements’ (see also Wolf Reference Wolf2008, Anderson Reference Anderson, Maiden, Goldbach, Smith and Hinzelin2011, Bonet & Harbour Reference Bonet and Harbour2012).

There are a number of ways to address this opacity problem. One solution is to locate vowel reduction in the post-lexical component of Irish phonology. On this view vowel reduction is intrinsically ordered after allomorph selection, given that word formation necessarily occurs in the earlier lexical component of the grammar. These assumptions lead to serial derivations like (39), in which post-lexical vowel reduction obscures the conditions governing plural allomorph selection in the lexical stratum.

There is some supporting evidence for a post-lexical treatment of vowel reduction in Irish. In Munster Irish, nouns like cipín  ‘stick’, which have an [LH] weight profile, normally bear stress on the second syllable. In sentential contexts, however, stress may retract to the initial syllable of the word. The conditions governing phrasal stress retraction are complex and not fully understood; what matters here is that when words with an [LˈH] profile are followed by a word bearing initial stress, stress typically retracts on the first word, e.g. cipín dearg  ‘small red stick’ (Ó Siadhail Reference Ó Siadhail1991: 31–32; see also Ó Cuív Reference Ó Cuív1944: 67; Ó Sé Reference Ó Sé1989, Reference Ó Sé2000: 52–54; Ó Buachalla Reference Ó Buachalla2003: 2).

The crucial observation about this pattern of stress shift concerns the quality of the newly stressed vowel. Vowel reduction eliminates all underlying place features from unstressed short vowels; as such, it destroys the otherwise unpredictable information about contrastive vowel quality that is stored in underlying representations like corcán . (While the backness of short vowels is predictable from the following consonant in Irish, height is not; Ó Siadhail Reference Ó Siadhail1991: 36–37, Ní Chiosáin Reference Ní Chiosáin1991: 140, etc.) Post-lexical stress shift, on the other hand, allows the underlying quality of such short vowels to surface unchanged, e.g. corcán mór  $\rightarrow$  

Further support for this analysis of the vowel alternations in (42) comes from the observation that certain words resist post-lexical stress shift, e.g. bothán tuí  ‘hayshed’. Ó Sé (Reference Ó Sé2000: 53) accounts for this contrast by assuming an underlying /ə/ in the initial syllable of words resisting stress shift; since schwa is not generally stressable in Munster Irish, the underlying /ə/ blocks phrasal stress retraction (see also Green Reference Green, Dobrin, Singer and McNair1996). The distinction between words like bothán (which block stress shift) and words like corcán (which do not) provides strong evidence that phrasal stress retraction recovers the underlying quality of the initial vowel, rather than simply raising or fronting/backing [ə] under phrasal stress.

Now, if vowel reduction were a lexical process, it would have to precede phrase-level stress shift, which is clearly post-lexical. However, this cannot be correct: lexical vowel reduction would neutralize underlying short vowels to [ə]/[ɪ], making it impossible to recover the different vowel qualities seen under post-lexical stress shift (43). Vowel reduction must therefore follow phrase-level stress assignment.

If vowel reduction follows post-lexical stress shift, then reduction must be post-lexical too. On this view of things, plural allomorph selection is both phonologically transparent and output optimizing at the lexical level, even if post-lexical vowel reduction leads to surface opacity down the road. See Kim (Reference Kim2010), Anderson (Reference Anderson, Maiden, Goldbach, Smith and Hinzelin2011) for similar issues in Huave and Surmiran, and Paster (Reference Paster2006), Wolf (Reference Wolf2008) for general discussion. (See Wolf Reference Wolf2008, Reference Wolfto appear for a non-stratal model of opaquely conditioned allomorph selection couched in a different variant of derivational OT.)

We can now return to some issues raised above regarding dialect variation in the surface form of the plural suffix -(e)acha. This suffix is regularly realized as unreduced [-axə] in West Ulster and Achill Irish (Section 2.4.1). For these dialects, there is no opacity problem: the selection of -(e)anna with monosyllabic noun stems is always transparently output optimizing (Section 2.6). In Munster Irish, the interaction between plural allomorphy and vowel reduction is indeed opaque, but arguably transparent and output optimizing at the lexical level.

Things are not so clear-cut for Connacht Irish, the third of the major dialect groups. To the best of my knowledge, the plural suffix -(e)acha always surfaces as reduced [-əxə] in these varieties (or as -(e)achaí ; see footnote 10 for references). Consequently, in Connacht Irish there is little empirical evidence that the plural suffix -(e)acha is underlyingly /-axə/ rather than /-əxə/. Similar issues arise for East Ulster Irish: here, -(e)acha is typically realized as [-əxə] or [-əhə] (Section 2.4.1 and footnote 31). For these dialects too the underlying form of -(e)acha must begin with /ə/, not /a/.

It should be clear that the preceding facts rule out the possibility that allomorphic alternations between -(e)acha and -(e)anna are also optimizing in Connacht and East Ulster Irish. If -(e)acha is underlyingly /-əxə/ (with a non-moraic [x]), then the stress-conditioned distribution of -(e)acha and -(e)anna must be phonologically arbitrary. I am perfectly willing to accept this conclusion (see also Section 3 below). Many regularities in Irish plural affixation are phonologically and morphologically arbitrary (Section 2). Furthermore, it does not follow that alternations between /-axə/ and /-ənə/ are phonologically arbitrary in all dialects. Indeed, all of the complicating factors at play here – the reduction of unstressed /ax/ to [əx] in Connacht and East Ulster; unstressed vowel shortening in Ulster; non-initial stress in Munster; and the [x]  ${>}$  [h] change in East Ulster – are innovative features of the dialects in question (O’Rahilly Reference O’Rahilly1932; Ó Dochartaigh Reference Ó Dochartaigh1987; Ó Sé Reference Ó Sé1989, Reference Ó Sé2008, and many others). On the other hand, the distribution of -(e)anna and -(e)acha must be a relatively old feature of the language, as it cuts cleanly across all major dialect groups. In the Western and East Ulster dialect groups, where all trace of underlying /ax/ has disappeared from the plural suffix -(e)acha, the non-optimizing character of this case of plural allomorphy is simply an innovative change.Footnote ^[30]

To summarize:

2.7.2 Consequences

In the preceding sections, I have argued that a subpattern of Irish plural allomorphy involving the suffixes -(e)anna and -(e)acha can be analyzed as emerging from the interaction of fairly uncontroversial constraints on metrical structure. Taken together, the constraints wsp_Ft  and Parse( ${\it\sigma}$ ) heavily penalize [ˈ ${\it\sigma}$ H] sequences. Irish plural allomorphy avoids such ill-formed structures by creating [ˈ ${\it\sigma}$ L] sequences whenever possible. However, Irish does generally tolerate [ˈLH] and [ˈHH] sequences, as in (45).

Parse( ${\it\sigma}$ ) and wsp_Ft  are mostly dormant in the language at large: the pressures that they exert are generally too weak to materially affect prosodic structure. Irish plural allomorphy thus constitutes an interesting case of the emergence of the unmarked in the metrical domain (Kager Reference Kager, Buszard-Welcher, Lee and Weigel1992, Reference Kager and Mead1993; Mascaró Reference Mascaró, Durand and Laks1996, Reference Mascaró2007).

A crucial piece of this analysis is the assumption that -(e)anna and -(e)acha are allomorphs of a single underlying morpheme: when the phonology of Irish has a choice between two allomorphs, it selects the allomorph that leads to an optimal prosodic structure (at the lexical level). In contrast, other plural suffixes, which have only a single surface form, often give rise to ill-formed [ˈ ${\it\sigma}$ H] sequences, e.g. the [ˈHH] profile of brógaí  ‘shoes’ (Stockman Reference Stockman1974: 317).

The analysis developed here thus successfully integrates a corner of Irish plural morphology into the broader morpho-phonology of the language. In the following sections I argue that an explicit connection between plural allomorphy and metrical well-formedness must be part of any explanatory account of the prosodically conditioned distribution of -(e)anna and -(e)acha.