Recursive Phonological Phrases in Conamara Irish* Emily Elfner University of Massachusetts, Amherst

1. Introduction Prosodic Hierarchy Theory (Selkirk 1978 et seq) defines the domain of application for certain phonological processes using a universally available set of prosodic categories (ι, ϕ, ω) that exist in a strict dominance relationship. Domain-sensitive phonological processes may target any one of these prosodic categories. In recent work, Ito & Mester (2006, 2010, to appear) have proposed an alternative picture of prosodic domains theory that capitalizes on the possibility of using recursion in prosodic structure. In addition to targeting distinct prosodic categories (ι, ϕ, ω), domain-sensitive processes may specifically refer to the maximal and minimal projections of recursion-based prosodic subcategories. Under this proposal, recursion-based subcategories may be used in place of relations between distinct intermediate categories, as in the distinction between Major and Minor Phrase proposed to account for phrasal phonological processes in Japanese (McCawley 1968 and others): (1) Maximal/minimal projections of recursive phonological phrases (ϕ) a. Major/Minor Phrase distinction

x… x

b. Maximal/minimal projections of recursive ϕ

ι

ι

MaP

ϕ

Maximal projection

MiP

ϕ

Minimal projection

ω

x… x

ω

In this paper, I discuss new data from Conamara Irish (CI)1 that examine the distribution of phraselevel tonal rises (see also Elfner to appear for an OT-based account of the same data). I argue that the data provide support for the idea that domain-sensitive processes may target recursion-based prosodic subcategories, as proposed by Ito and Mester. I argue that a rise falls on the leftmost ω in every recursive (non-minimal) ϕ, where recursive prosodic structure is built under pressure from syntaxprosody interface constraints that call for prosodic constituency to mirror syntactic constituency (following Match Theory, Selkirk 2009, to appear). The analysis proposes that, in addition to the *

Many thanks to my Irish consultants for generously donating their time and language: Peggy Cloherty, Michael Newell, Máire Uí Fhlaharta, Reiltin Ní Fhlaharta, and Yvonne Ní Fhlaharta. Thanks also to Lisa Selkirk, Jim McCloskey, John McCarthy, Armin Mester, Junko Ito, and audiences at UMass Amherst, UC Santa Cruz, NELS 40 (MIT), and SPINE4 (UMass Amherst) for discussion and feedback relating to this work, as well as to SSHRC (Doctoral Fellowship # 752-2006-1349) and the NSF (grant # BCS-0527509 to the University of Massachusetts) for financial support. 1 Conamara Irish refers to the dialect of Irish (Celtic) spoken in the Conamara region of County Galway, Ireland. The speakers who participated in this study are from the towns of Ros Muc (one speaker, now living in Boston, MA) and Carraroe/An Cheathrú Rua (three speakers). This dialect has many features in common with other dialects of Irish, and there is in addition some variation among speakers within this dialect. I assume that the patterns discussed here are features of the Conamara dialect.

maximal and minimal projections, phonological processes may target other natural classes of recursion-based subcategories, including the class of non-minimal projections: (2) Some natural classes of recursion-based prosodic subcategories a. Maximal/minimal projections

b. Non-minimal projections

ι ϕ

ι Maximal projection

ϕ Non-minimal projections

ϕ x… x

ϕ

ϕ Minimal projection

x… x

ω

ϕ ω

I show that a recursion-based subcategories account of the distribution of rises in CI correctly captures the distributional patterns of the H tones. Firstly, the theory correctly predicts that the presence of a rise correlates with the presence of recursive prosodic structure, as derived from syntactic structure. Secondly, by assuming that prosody is a distinct grammatical system, the theory predicts that purely prosodic constraints on prosodic binarity (Inkelas & Zec 1990) may over-rule syntax-prosody correspondence constraints, resulting in a mismatch between syntactic and prosodic structure. In CI, this results in the absence of rises in certain environments. Conversely, sentences that differ in syntactic structure may show identical prosodic structure, as derived under pressure from interface and binarity constraints, provided that they satisfy the constraints in the same way. I show that this prediction is borne out in CI through an examination of sentences with complex possessive structures in object position.

2. The distribution of rises in Conamara Irish Basic word-order in CI, like other varieties of Irish, is VSOX in finite clauses, where X is an adjunct or indirect object; adjectives follow the noun: (3) a.Tabharfaidh mo mháthair leabhar don leabharlann give.FUT my mother book to.the library ‘My mother will give a book to the library.’ b. Rinne a rúnaí nua dearmad ollmhór maidin inné. made his secretary new mistake huge morning yesterday ‘His new secretary made a huge mistake yesterday morning.’ Speakers of CI show a pattern of L-H rises, where rises align with some, but not all, lexical words.2 For example, in an all-new VSO sentence, an H tone falls on the verb (díolfaidh ‘sell.fut’) and the leftmost lexical word of the subject (máthair ‘mother’). For this speaker, the L tone aligns with the initial (stressed) syllable, with the H tone peak being realized on the second syllable:

2

All data cited in this paper are based on original fieldwork. Consultants were asked to read sets of sentences in which the target sentence was preceded and followed by context sentences. None of the words in the sentences were focussed.

(4) Pitch track for VSO sentence

2009_08_04_013YF

Note that there is a final H-L fall on the final word of the sentence. While beyond the scope of the present paper, this fall appears to be a right-edge boundary tone that may be associated with the right edge of the sentence or intonational phrase. Additional data indicate that the rises do not simply mark the left-most word of certain syntactic constituents (i.e. verb and subject). For example, if an X element is added to form a VSOX sentence, a rise now falls on the leftmost word of the object (leabhar ‘book’) as well as on the verb and the subject as in (4): (5) Pitch track for VSOX sentence3

2009_07_29_007YF

Lexical stress is generally initial in CI.4 At present, the data do not distinguish between an analysis where the rises are boundary tones marking the left edge of a prosodic constituent (L-H%) and an analysis where the rises are pitch accents that align with the stressed syllable of the leftmost word in a prosodic constituent (L*+H). For the purposes of this paper, I will refer to them as rises (L-H) and remain agnostic with respect to whether they are boundary tones or pitch accents, as this will not affect the current proposal. See also Dalton & Ní Chasaide (2005a, 2005b, 2007) for a study of tonal alignment in Irish dialects including Conamara Irish. These studies similarly found that the H-tone peak appears to align with the beginning of the post-tonic syllable, though the phonological analysis presented there differs somewhat from the current analysis. Future research will help to shed light on this issue.

3

This example also shows a pause between S and O, as well as a rise on the rightmost edge of the subject. Lengthy pauses are also common in natural speech (Bennett 2008) and a rise often appears before pauses, suggesting that it is a boundary tone and not the same L-H rise that marks the left edge of recursive ϕs. In my data, pauses and pre-pausal rises appear to be found only in VSOX sentences, where OX is at least three words. 4 Note that in Irish orthography, acute accents as in máthair indicate the presence of a long vowel and not the location of stress.

3. Match Theory and Recursive Prosodic Subcategories Prosodic Hierarchy Theory (Selkirk 1978 et seq.) asserts the existence of a finite set of universally available prosodic levels that exist in a hierarchical relationship: (6) Prosodic Hierarchy ι Intonational Phrase ϕ Phonological Phrase ω Prosodic Word These levels define prosodic constituency, and form the basis for prosodic domains within which domain-sensitive phonological processes may apply. Match Theory (Selkirk 2009, to appear) is a correspondence-based approach to the syntax-prosody interface where syntactic constituents correspond to prosodic constituents (as opposed to edge-based or alignment-based approaches e.g. Selkirk 1986, 1995; Truckenbrodt 1995, 1999): (7) Match Theory MATCH CLAUSE: Syntactic clause à MATCH PHRASE: Syntactic phrase (XP) à MATCH WORD: Syntactic word à

Intonational phrase (ι) Phonological Phrase (ϕ) Prosodic Word (ω)

These constraints, when fully satisfied, create a prosodic constituent structure that is recursive and closely mirrors syntactic structure. For example, if we take the constraint Match Phrase on its simplest terms, this constraint predicts that a syntactic phrase of any type (DP, VP, TP, etc.) should correspond to a ϕ. For an Irish VSOX sentence, Match Phrase predicts that the abstract syntactic structure in (8)a (based on McCloskey 2001, 2009) should correspond to the recursive prosodic structure in (8)b, provided that each of S, O, and X are XPs (DPs or PPs):5 (8) a.Syntactic Representation

b. Recursive Prosodic Representation

ΣP

ΣP

ϕ

Vi TP

V

S

V’ ti

X

VP

ϕ

ϕ S

VP

TP

ϕ

ϕ O

ϕ X

DP/PP

O

Recursion-based prosodic subcategories theory (Ito and Mester 2006, 2010, to appear) distinguishes between the maximal and minimal projections of recursive prosodic structures, such that phonological constraints can specifically target these projections:

5

This syntactic structure and the syntax-to-prosody mapping is simplified in several respects. For example, I ignore the presence of phrasal projections like vP if they do not contain any phonologically overt material or if they dominate the same set of elements as a lower phrase, as in a structure like XP[YP[ZP[lex]]]. I also assume that functional projections headed by weak functional elements (such as P or D) do not project their own ϕ because they are not prosodically heavy enough.

(9) κMax: κ not dominated by κ κMin: κ not dominating κ I propose an extension to this proposal: in prosodic structures with multiple levels of recursivity, phonological constraints may also distinguish non-minimal prosodic constituents from minimal prosodic constituents: (10) κNon-min: κ dominating κ The distribution of rises in CI represents a process of this type: (11) Distribution of rises in CI A rise tone falls on the leftmost ω in a non-minimal ϕ. The contrast between the VSO and VSOX sentences above falls out from this account. In (12)a, the VSO sentence contains two ϕNon-min, and two rises, while in (12)b, the VSOX sentence contains an extra ϕNon-min and a corresponding third rise: (12) Proposed prosodic representations for VSO and VSOX sentences a. VSO sentence: rises on V and S ΣP

ϕNon-min ω V L-H

ϕMin

ω ω ω N Adj N L-H V

S

ΣP

ϕNon-min

TP

ϕNon-min ϕMin

b. VSOX sentence: rises on V, S, and O

ω VP/DP or PP

ω Adj

O

TP

ϕNon-min

V L-H

ϕMin

VP

ϕNon-min

ϕMin ω ω ϕMin N Adj L-H ω ω ω ω N Adj N (Adj) L-H V

S

O

DP orPP

X

4. Interaction with prosodic constraints The recursion-based subcategories/Match Theory analysis assumes that a level of prosodic constituency mediates between syntactic structure and domain-sensitive phonological processes such as CI H tone insertion. Accordingly, the theory predicts that prosodic well-formedness constraints, as on prosodic binarity (e.g. Inkelas & Zec 1990), may intervene to create a mismatch between syntax-toprosody mapping. Additional data indicate that rises disappear when the weight of S or O is reduced from two ω (N+Adj) to one (N only). For example, in a VSOX sentence as in (13), where the subject contains a single lexical word (máthair ‘mother’), the rise on the subject is not present. In contrast with the fully binary VSOX sentence in (5) above, rises fall on the verb and the object but not the subject:

(13) Pitch track for VSOX sentence with a one-word subject: no rise on subject

2009_08_04_025YF

The absence of the rise on S suggests that the prosodic representation of (13) contains only two levels of recursive ϕNon-min, rather than three as in the fully binary VSOX sentence in (5). I propose to group the subject with the verb to form a binary ϕ. This results in a prosodic structure that contains only two levels of recursivity, as in (14)a. However, a representation as in (14)b that fully satisfies Match Phrase incorrectly predicts that the subject should be marked with a rise, as it is the leftmost ω in the ϕNon-min that corresponds to the TP: (14) Possible prosodic representations for (13), a VSOX sentence with single-ω subject a. Proposed prosodic representation ΣP

ϕNon-min

ω ω V N L-H

V

S

ϕMin

ϕMin

ω ω ω N Adj N L-H O

* VP

ϕNon-min

ϕMin

b. Match Phrase representation (incorrect) ΣP

ϕNon-min

ω V L-H

ω Adj

TP

ϕNon-min ω N L-H

ϕNon-min ϕMin

VP ϕMin

ω ω ω ω N Adj N Adj L-H

X V

S

O

X

The contrast between the two structures exemplifies a mismatch between syntactic and prosodic structure: the proposed prosodic structure in (14)a shows less recursive structure as compared to (14)b. This suggests that there is a prosodic constraint that prefers binary ϕs, and that this constraint can overrule Match Phrase. A full analysis is beyond the scope of this paper, but see Elfner (to appear) for an OT analysis of the same data.

5. A look at complex DPs: Possessive objects Additional support for the recursion-based analysis of the distribution of rises in CI comes from an investigation of VSO sentences with complex possessive constructions in object position. These show a slightly different syntactic structure as compared to VSOX sentences, but the two sentence types behave identically with respect to rise distribution.

In Irish, the possessed object is followed by the possessor in the genitive case:6 (15) a. fear an tí man the.GEN house.GEN ‘the man of the house’ b. hata gorm an fhir mhóir hat blue the.GEN man.GEN big.GEN ‘the blue hat of the big man/the big man’s blue hat’ The syntactic structure of a binary possessive construction as in (15)b is shown in (16)a (based on McCloskey 2006). Blind application of Match Phrase predicts the recursive prosodic structure in (16)b, where the N and the Adj in the possessor are phrased into separate ϕNon-min: (16) The structure of possessives a. Syntactic structure

b. Predicted prosodic structure (Match Phrase)

DP

FP

ϕNon-min

D ø

FP

ω N L-H

DegP

N

Deg Adj DP NP t N Adj (GEN)

DegP

ϕNon-min ω Adj L-H

O

ϕMin ω ω N Adj possessor

However, given the observed interaction with the prosodic constraint on binarity, we predict that the prosodic structure in (16)b will be dispreferred to a structure as in (17)a, which phrases the N and Adj into a single ϕ. The structure in (17)b shows that when the binary possessive object is embedded as the object in a VSO sentence, we predict rises to fall on V, S, and O: (17) Prosodic representation of possessive and possessive object in a VSO sentence a. Prosodic representation: Possessive FP

ϕNon-min ϕMin

ϕMin ω N L-H O

ω Adj

ω N

b. VSO with binary possessive object

ω Adj

ω V L-H

TP

ϕNon-min ϕMin

ϕNon-min

FP (DP)

ω ω ϕMin ϕMin N Adj L-H ω ω ω ω N Adj N Adj L-H

possessor

V 6

ΣP

ϕNon-min

S

O

possessor

A determiner precedes the genitive possessor, but not the possessed object (see e.g. McCloskey 2006). Irish has a definite determiner an ‘the’, as well as possessive pronouns (e.g. mo ‘my’), but no overt indefinite determiner.

This prediction is borne out: in a VSO sentence with a binary possessive object, rises fall on each of V, S, and O: (18) Pitch track for VSO sentence with binary possessive object

2009_06_19_009MN

This pattern is identical to that of a fully binary VSOX sentences as in (5). Compare to the pitch track for a VSO sentence with a simple possessive object. This behaves like a regular VSO sentence with a heavy (binary) object, with two rises on V and S, but no rise on O (compare with (4)): (19) Pitch track for a VSO sentence with a simple possessive object (two nouns)

2009_08_04_024YF

As before, the distribution of rises in CI can be shown to target non-minimal ϕs. The degree of recursivity in prosodic structure is predictable from the interaction between Match Phrase, which is sensitive to syntactic constituency, and the prosodic constraint on binarity.

6. Conclusion The distribution of rises in CI can be summarized as follows: (20) Distribution of rises in CI A rise tone falls on the leftmost ω in a non-minimal ϕ. The present analysis differs from other possible approaches to the construction of ϕs in a number of different ways. Firstly, under the present analysis, prosodic structure is assumed to be recursive, under the additional assumption that phonological constraints can target either prosodic categories (ι, ϕ, ω) or recursion-based subcategories, including the class of non-minimal projections. Secondly, recursive prosodic structure is motivated by an interface constraint Match Phrase, which results in a prosodic

representation that closely mirrors syntactic structure in terms of constituency. This means that not only can prosodic structure be recursive, but it is preferentially so, as this will provide a better match with the syntax. However, prosodic constraints, as on binarity, can cause the recursive prosodic structure to deviate from the structure provided by the syntax. This analysis of the distribution of rises in CI correctly accounts for a number of patterns that would be puzzling under alternative accounts that either adhere to Strict Layering (Selkirk 1981, Nespor & Vogel 1986, Pierrehumbert & Beckman 1988) by banning recursivity. This paper has also provided support for a theory of prosodic domains that allows phonological constraints to target natural classes of recursive prosodic subcategories, as proposed by Ito & Mester (2006, 2010, to appear), including the proposed class of non-minimal ϕs.

References Bennett, Ryan. 2008. Donegal Irish and the syntax/prosody interface. Ms., UC Santa Cruz. Presented at PRIG (Prosody Interest Group), UCSC. Dalton, Martha and Ailbhe Ní Chasaide. 2005a. Tonal alignment in Irish dialects. Language and Speech 48:441464. Dalton, Martha and Ailbhe Ní Chasaide. 2005b. Peak timing in two dialects of Connaught Irish. In Proceedings of Interspeech: The 9th European Conference on Speech Communication and Technology, Lisbon. Dalton, Martha and Ailbhe Ní Chasaide. 2007. Melodic alignment and micro-dialect variation in Connemara Irish. In C. Gussenhoven and T. Riad (eds.) Tones and Tunes: Experimental Studies in Word and Sentence Prosody. Berlin: Mouton de Gruyter, 293-316. Elfner, Emily. To appear. Recursivity in prosodic phrasing: Evidence from Conamara Irish. To appear in Seda Kan, Claire Moore-Cantwell, and Robert Staubs (eds.) Proceedings of the 40th Annual Meeting of the NorthEast Linguistic Society. Amherst, MA: GLSA publications. Inkelas, Sharon and Draga Zec (eds.). 1990. The Syntax/Phonology Connection. CLSI. Ito, Junko and Armin Mester. 2006. Prosodic adjunction in Japanese compounds. Formal Approaches to Japanese Linguistics: Proceedings of FAJL 4. MITWPL 55: 97-112. Ito, Junko and Armin Mester. 2010. The onset of the prosodic word. In Steve Parker (ed.) Phonological Argumentation: Essays on Evidence and Motivation. London: Equinox. Ito, Junko and Armin Mester. To appear. Recursive prosodic phrasing in Japanese. To appear in Toni Borowsky et al. (eds.) Prosody Matters: Essays in honor of Elisabeth Selkirk, London: Equinox. McCawley, J. D. 1968. The Phonological Component of a Grammar of Japanese. The Hague, the Netherlands: Mouton. McCloskey, James. 2001. On the distribution of subject properties in Irish. In William D. Davies and Stanley Dubinsky (eds.) Objects and Other Subjects. Dordrecht: Kluwer, 157-192. McCloskey, James. 2006. The morphosyntax of agreement in Irish. Paper presented at Yale University, September 12. McCloskey, James. 2009. The syntax of clauses in Irish. Lectures presented at Formal Approaches to Celtic Linguistics, University of Arizona, March 23-27. Nespor, Marina and Irene Vogel. 1986. Prosodic phonology. Dordrecht: Foris. Pierrehumbert, Janet and Mary Beckman. 1988. Japanese Tone Structure. Cambridge, MA: MIT Press. Selkirk, Elisabeth. 1978. On prosodic structure and its relation to syntactic structure. In T. Fretheim (ed.) Nordic Prosody 2. Trondheim: TAPIR, 128–149. Selkirk, Elisabeth O. 1981. On the nature of phonological representation. In J. Anderson, J. Laver and T. Meyers (eds.) The Cognitive Representation of Speech, Amsterdam: North Holland. Selkirk, Elisabeth. 1986. On derived domains in sentence phonology. Phonology 3:371-405. Selkirk, Elisabeth. 1995. The prosodic structure of function words. In J. Beckman et al. (eds.) Papers in Optimality Theory. Amherst, MA: GLSA. Selkirk, Elisabeth. 2009. On clause and intonational phrase in Japanese: The syntactic grounding of prosodic constituent structure. Gengo Kenkyu 136:35-73. Selkirk, Elisabeth. To appear. The syntax-phonology interface. To appear in Goldsmith, Riggle, & Yu (eds.) The Handbook of Phonological Theory, 2nd edition. Truckenbrodt, Hubert. 1995. Phonological phrases: Their relation to syntax, focus, and prominence. Doctoral dissertation, MIT, Cambridge, Mass. Truckenbrodt, Hubert. 1999. On the relation between syntactic phrases and phonological phrases. LI 30:219-255.