Cluster Reduction : Deletion or Coalescence?

Abstract I show that a consequence of Correspondence Theory in Optimality Theory is that, for processes such as cluster reduction, if MAX outranks UNIFORMITY, candidates displaying coalescence are preferred to those displaying true deletion. It is thus incumbent on the analyst to identify the constraints that select appropriate coalescence candidates, even for apparent deletion cases. I show how Lamontagne and Rice’s (1995) account of the D-effect in Navajo must be modified to ensure the correct outcome in a language where both coalescence and apparent deletion are repairs to cluster constraint violations. If, for other reasons, it is necessary that UNIFORMITY outrank MAX, the admission of MAX(Feature) constraints becomes unavoidable. An analysis of certain cluster reduction phenomena in Ibiza Catalan shows how complex coda constraints, perceptual markedness constraints for clusters, Paradigm Uniformity, and featural faithfulness interact to derive a pattern of contextual variation. The paper includes a review of Correspondence Theory focusing on its effects in cluster reduction. Key words: cluster reduction, coalescence, Correspondence Theory, deletion, fusion, Optimality Theory, phonology; Catalan, Navajo. 1. Introduction Consonant cluster reduction, illustrated with an English example in (1), is one of several types of process by which the number of output segments deviates from the number of input segments. A parallel process involving vowels is «apocope», as in French l’état [leta] ‘the state’ /lə/ ‘the’ + /eta/ ‘state’ *[ləeta]. * I am very grateful to a CatJL reviewer for many suggestions which have helped to improve the text. Catalan Journal of Linguistics 4, 2005 57-82 Cluster Reduction: Deletion or Coalescence?* Max W. Wheeler University of Sussex. Department of Linguistics & English Language Falmer, Brighton BN1 9QN (United Kingdom) M.W[email protected] Table of Contents 1. Introduction 2. Correspondence Theory reviewed 3. Trying again with deletion and coalescence in Navajo 4. Cluster reduction in Catalan: a sample case 5. Concluding observations References Cat.Jour.Ling. 4 001-252 7/2/06 11:43 Página 57 (1) Base form Contextual cluster reduction hand [œhand] hands [œhanz]/hand+z/ handful [œhafυl]/hand+fυl/ If we find more segments in the output than in the input, we typically speak of epenthesis in the broad sense (covering all insertions), as in English drawing [œdɹɔɹiŋ] /œdɹɔ/ + /iŋ/, or Spanish está [esœta] ‘is.3SG.PR.IND’ /œsta/.1I use here the general terms «input» and «output», though, of course, deviation in the number of segments can be observed in the whole range of Optimality Theory correspondence relations such as Base-Reduplicant (2a), Base-Derivative (2b), or Word-Phrase (2c) illustrated again with examples of consonant-cluster reduction. (2) Base form Cluster reduction a. Prefixed reduplicant Ancient Greek πρσσω ππραχα [pɾássɔ] [pé-pɾaka] *[pɾé-pɾaka] ‘I make’ ‘I have made’ b. Prefixed root Latin ex eligo [eks] [e-lio] *[eks-lio] ‘out’ ‘I pick out’ c. Phrasal sandhi hand right hand side [hand] […han said] *[…hand said] Within Optimality Theory, it is Correspondence Theory which deals in general terms with deviations from faithfulness (McCarthy and Prince 1999, a revised and reduced version of McCarthy and Prince 1995; McCarthy 1995), and in Correspondence Theory several types of constraints have been formulated that penalize different varieties of unfaithfulness. In the case of processes involving deviation from faithfulness in the number of segments, phonologists have generally spoken of cluster reduction and apocope in terms of «deletion», or violation of MAX, and of epenthesis in terms of «insertion», or violation of DEP. It is the argument of this paper that exclusive reliance on MAX in dealing with cluster reduction works only if coalescence candidates are ignored, which, granted that GEN supplies such candidates, they should not be. For any grammar where, in some circumstances, coalescence candidates must win, it is necessary for the phonologist to show why, in a particular case, a deletion candidate is superior to an alternative coalescence candidate. What looks like deletion cannot be assumed to be simply the consequence of some markedness constraint outranking MAX. Candidates violating MAX have to be shown to be better than alternatives that lack the MAX violation. 58 CatJL 4, 2005 Max W. Wheeler 1. «Epenthesis» in the narrow sense is restricted to string-medial insertions; see Appendix. Cat.Jour.Ling. 4 001-252 7/2/06 11:43 Página 58 Some examples of alternative candidates for cluster reduction in hands are presented in (3). To put it another way, cluster reduction illustrates not simply the ranking of Markedness above MAX; it must involve some explicit ranking of Markedness, MAX and UNIFORMITY, UNIFORMITY being the constraint that penalizes coalescence. This point appears to have been overlooked by phonologists who have treated cluster reduction in the light of Correspondence Theory, starting with Lamontagne and Rice (1995).2Cluster reduction in Catalan, the focus of section 4 of this paper, has been treated by Jiménez (1999), Dols (2000) and Pons (2004). All of these authors cite McCarthy and Prince (1995), and Jiménez in particular (225-240) has winning coalescence candidates in consonant cluster contexts, such as pots comprar [pɔËts.kom.œpɾaɾ] ‘you can buy’. In her extensive survey of consonant cluster reduction and epenthesis, Côté (2001: 163) too ignores the role of coalescence or breaking candidates (and of the constraints they violate): The markedness constraints against non-prevocalic consonants interact with faithfulness constraints to yield the attested patterns. Since I deal here only with epenthesis and deletion, I use the following two basic constraints [...] a. MAX Do not delete b. DEP Do not epenthesize. The problem involved in ignoring coalescence candidates provided by GEN can be illustrated in Lamontagne and Rice’s (1995) account of some consonantal cluster reduction phenomena in Navajo prefixal inflection known as the «D-effect».3 The D-effect involves both deletion (4a) and coalescence (4b) as repairs to potential NOCODA violations. Symbols such as [d], [] in Navajo transcriptions denote voiceless unaspirated stops, while [t], [k] denote voiceless aspirated stops. Note that Lamontagne and Rice’s NOCODA penalizes only internal codas, i.e. it is *C]σC. Cluster Reduction: Deletion or Coalescence? CatJL 4, 2005 59 2. McCarthy (1995: 50) does address the theoretical point, though in the context of discussion of umlaut (in Rotuman) rather than of cluster reduction. 3. I am grateful to Keren Rice for supplying me with a copy of this paper. Lamontagne and Rice’s account uses some preliminary formulations of correspondence constraints which I replace here with their now more familiar versions (after McCarthy and Prince 1999); the form of their argument is not affected by this modification. (3) Base form Contextual cluster reduction Inflected input Deletion Selected coalescence candidates hand [hand] hands /han1d2+z3/[han1z][han1,2z], [hanz2,3] [hand1,2z], [han1dz2,3], [handz1,2,3] Cat.Jour.Ling. 4 001-252 7/2/06 11:43 Página 59 (4) Navajo Cluster reduction a. /d/ + stop-initial stem /ʔi+ii+d+kááh/ →[ʔii.kááh] deletion ‘we make a sand painting’ b. /d/ + fricative-initial stem /na+ii+d+xaa/ →[nei.aa] coalescence ‘we look around’ In the structure illustrated in (4) codas are avoided (by NOCODA). Before a stem beginning with a stop the element /d/ is «deleted» (4a); in the case where a stem begins with a fricative (4b), the input sequence of /d/ + C is realized as a «coalesced» stop with the laryngeal stricture of /d/ and the place of articulation of the stem-initial consonant /x/, namely []. Lamontagne and Rice offer tableaux of the forms in (5) and (6) for the two processes respectively. (5) (6) On the basis of these tableaux Lamontagne and Rice conclude (1995: 218): “to summarize, deletion involves a [MAX] violation while coalescence involves a violation of another constraint on correspondence, namely [UNIFORMITY]. In Navajotype Athapaskan languages, the D-effect facts follow if [MAX] » [UNIFORMITY] (and both of these constraints are dominated by NOCODA).” Lamontagne and Rice add some discussion on why deletion occurs before stops, and why coalescence with stem-initial fricatives takes the form it does, which mentions aspects of differential featural faithfulness and featural alignment. What Lamontagne and Rice do not observe is that with the constraint ranking NOCODA » MAX » UNIFORMITY, their intended winning candidate (5b) would in fact lose to a fusion candidate such as [ʔii.k1,2ááh], or indeed *[ʔii.1,2ááh] which maintains the laryngeal stricture of its first correspondent stop segment just as the winner does in (6c). That is, once MAX dominates UNIFORMITY in the grammar, coalescence candidates will always be more harmonic than deletion ones, unless the coalescence candidates are themselves ruled out by some higher ranking constraint(s). 60 CatJL 4, 2005 Max W. Wheeler /ʔi+ii+d1+k2ááh/ NOCODA MAX a. ʔiid1.k2ááh *! ☞b. ʔii.k2ááh * /na+ii+d1+x2aa/NOCODA MAX UNIF a. neid1.x2aa*! b. nei.x2aa*! ☞c. nei.1,2aa* Cat.Jour.Ling. 4 001-252 7/2/06 11:43 Página 60 The structure of the remainder of the paper is as follows. In section 2 I review Correspondence Theory focusing especially on how correspondence constraints treat cluster reduction. In section 3 I show how Lamontagne and Rice’s account of Navajo can and must be elaborated to express the desired result. In section 4 I investigate a sample of consonant cluster reduction in Catalan, exploring further the contributions of deletion and coalescence, and their interaction with particular types of perceptual markedness and with morphological analogy. Section 5 introduces some broader consequences of the issues raised in the body of the paper. 2. Correspondence Theory reviewed In this section I review Correspondence Theory highlighting issues of multiple correspondence. I also draw attention to some other interactions between the types of constraint that compose Correspondence Theory. In the discussion which follows I refer to «input» and «output» generally, whatever the specific basis of correspondence. In the notation convention of McCarthy and Prince, S1denotes input in this general sense, while S2denotes output. The definitions of correspondence constraints (7)-(16) are those of McCarthy and Prince (1999: 293-296). (7) MAX [MAXIMALITY] Every element of S1has a correspondent in S2. Domain (ℜ) = S1. MAX penalizes segment deletion in any position. «Element» in the constraint definition conventionally means «segment», though moras have also been protected in this way. (In principle, if MAX is applicable to moras, one should expect it to be applicable to other elements of the prosodic hierarchy, syllable, foot, and so on.) The loss of features carried by a deleted segment is not specifically penalized by MAX. For this reason some phonologists make use of a MAX(Feature) constraint type, for individual features, so that the absence of a specific input feature in any correspondent in the output is penalized (for example, Lombardi 2001: 21, expanding suggestions made in McCarthy and Prince 1995: 71, and discussed slightly more fully in McCarthy 1995: 50-52). (8) DEP [DEPENDENCE] Every element of S2has a correspondent in S1. Range (ℜ) = S2. DEP penalizes insertion in any position, conventionally of segments, but in principle, by analogy with the interpretation of MAX, also of prosodic elements such as mora. DEP(Feature) seems not to be used, doubtless because the effect is more perspicuously achieved by markedness constraints, in Input-Output correspondence, at least. DEP(Feature) constraints are likely to have a significant role in Output-Output correspondence. Cluster Reduction: Deletion or Coalescence? CatJL 4, 2005 61 Cat.Jour.Ling. 4 001-252 7/2/06 11:43 Página 61 MAX and DEP are the most general constraints of a family whose other members, namely CONTIGUITY, ANCHOR, and ADJACENCY (see below), penalize deletion or insertion in specific segmental string patterns. Of these, ANCHOR and ADJACENCY have effects beyond penalizing deletion and insertion, whereas CONTIGUITY is simply a positionally restricted version of MAX/DEP. If there are MAX and DEP constraints for phonetic features, it follows that CONTIGUITY(Feature), ANCHOR(Feature), and ADJACENCY(Feature) will also be appropriate. (9) IDENT(F[eature]) Correspondent segments have identical values for the feature F. If xℜy and x is [γF], then y is [γF]. It is IDENT(F) that requires feature matching in correspondent segments; however, IDENT(F) is not violated in segments that lack a correspondent. So features of deleted segments are lost without penalty by IDENT(F), and insertion can introduce features not present in the input without violation of IDENT(F). In cases of coalescence or breaking (= splitting), IDENT(F) is typically violated, for some feature or features, except where coalescence and breaking consist of degemination and gemination respectively. McCarthy and Prince (1995: 71), initiating a discussion of MAX(F) and DEP(F), ponder whether IDENT(F) may actually be replaceable by constraints of the MAX and DEP types. (10) I-CONTIGUITY (“No skipping”) The portion of S1standing in correspondence forms a contiguous string. Domain (ℜ) is a single contiguous string in S1. I-CONTIGUITY penalizes syncope. In cases of syncope (see Appendix) a segment internal to S1lacks a correspondent in the output. Thus in a1b2c3→a′1c′3the portion of S1that stands in correspondence consists of a1and c3, which are not contiguous. Deletion at an edge is not penalized; thus in apocope, for example a1b2c3→a′1b′2, the portion of S1that stands in correspondence is the contiguous a1b2. Notice that I-CONTIGUITY does not require that what is contiguous in the input be contiguous in the output: I-CONTIGUITY does not penalize (internal) epenthesis —that is the role of O-CONTIGUITY— nor does it penalize coalescence. Thus the realization a1b2c3→a′1c′2,3, where c′2,3 in the output corresponds to the sequence of segments b2c3in the input, does not involve an I-CONTIGUITY violation, even though a′1is contiguous with c′3in the output, while their correspondents in the input are separated by b2. (11) O-CONTIGUITY (“No intrusion”) The portion of S2 standing in correspondence forms a contiguous string. Range (ℜ) is a single contiguous string in S2. O-CONTIGUITY penalizes epenthesis in the strict sense, that is, non-edge insertion of segments. Thus abc →a′xb′c′incurs an O-CONTIGUITY violation, while 62 CatJL 4, 2005 Max W. Wheeler Cat.Jour.Ling. 4 001-252 7/2/06 11:43 Página 62 abc →a′b′c′x does not. Like I-CONTIGUITY, O-CONTIGUITY does not penalize coalescence or breaking. The constraint sometimes simply named CONTIGUITY is an abbreviation for the constraint conjunction I-CONTIGUITY & O-CONTIGUITY, or refers to either or both of I-CONTIGUITY and O-CONTIGUITY, as may be relevant. CONTIGUITY does not inherently penalize metathesis provided that the corresponding portions of S1and S2 form contiguous strings, as is the case in abc →b′a′c′. However, it is not entirely clear how one is intended to identify the “portions of S1/S2standing in correspondence”. The portions standing in correspondence are usually taken to be whole morphemes (Kager 1999: 251), but not morpheme strings. Strictly, then, the CONTIGUITY constraints, like ADJACENCY (see below) need to have specified a morphological or prosodic domain. Thus, a sequence of two morphemes such as English hands /h1a2n3d4+z5/, realized [h1a2n3z5], is not interpreted as involving an ICONTIGUITY violation, but rather (beyond the general MAX violation) as involving a violation of RIGHT-ANCHOR at the edge of a Stem morpheme. (12) [RIGHT, LEFT] ANCHOR (S1, S2) Any element at the designated periphery of S1has a correspondent at the designated periphery of S2. Let Edge(X, {L, R}) = the element standing at the Edge = L, R of X. RIGHT-ANCHOR. If x = Edge(S1, R) and y = Edge(S2, R) then xℜy. LEFT-ANCHOR. Likewise, mutatis mutandis. Conceptually, ANCHOR constraints reflect the stronger faithfulness requirements of constituent edges; in this respect, they are part of a Positional Faithfulness approach (Beckman 1998). In the definition of ANCHOR, X stands for a prosodic category, like Foot, Syllable, or Phonological Word, or for a morphological category, such as Root, Stem, or Affix. McCarthy and Prince (1999: 295) intend that ANCHOR constraints should subsume Generalized Alignment (McCarthy and Prince 1993). The same point is made by McCarthy (2003: 89) who introduces a D-ANCHOR constraint specifically regulating the concatenation of morphemes that I do not consider further here.4 (13) LINEARITY (“No metathesis”) S1is consistent with the precedence structure of S2and vice versa. Let x, y ∈S1and x′, y′∈S2. If xℜx′and yℜy′ then x < y iff ¬(y′< x′) Cluster Reduction: Deletion or Coalescence? CatJL 4, 2005 63 4. The definition is as follows (McCarthy 2003: 90): D-ANCHOR (CI, CO, E) If x = Edge(CI, E) and y = Edge(CO, ˚), then xℜx′and x′is immediately adjacent to y. “Any element at the designated edge of CIhas a correspondent that is adjacent to an element at the opposite edge of CO.” Cat.Jour.Ling. 4 001-252 7/2/06 11:43 Página 63 The LINEARITY constraint penalizes all metathesis of corresponding segments, though not coalescence or breaking. That is to say, for example, if a precedes b in the input, it does not matter if a′coalesces with b′in the output (so a′ceases to precede b′); it is only when precedence is reversed so b′precedes a′in the output that a penalty is incurred. None of the constraints so far listed (7)-(13) penalizes multiple correspondence, that is, where one segment in the output corresponds to more than one segment in the input (coalescence), or vice versa (breaking). UNIFORMITY and INTEGRITY are the constraints that address such cases. (14) UNIFORMITY (“No coalescence”) No element of S2has multiple correspondents in S1. For x, y ∈S1and z ∈S2, if xℜz and yℜz then x = y. UNIFORMITY penalizes segmental coalescence, also referred to as fusion (for example, in Pater 1999). Except in the case where coalescence consists of degemination (e.g. [k1k2] →[k1,2]), coalescence will entail some violation of IDENT(F), since different adjacent segments must differ in some feature. However, the coalescence of two consonants as an affricate, with a segment-internal «contour» [[–cont][+cont]], such as [t1s2] →[ts1,2], need not incur an IDENT(F) violation. The same consideration applies to a vowel bearing a contour tone, such as [[H][L]] for a falling tone, or to a short diphthong such as [ea] [[–low][+low]]. Exactly how Correspondence Theory deals with contour segments remains to be worked out. What, for example, is the constraint that would penalize unfaithful ordering of contour feature values? By what constraint are both [st′] and [ts′] not equally good coalesced correspondents of input [ts] (or, indeed, of input [st])? It may be that a constraint LINEARITY(F) is required, to penalize reversal in the order of features separately from the segments they appear in.5Is UNIFORMITY evaluated categorically or gradiently? In practice it is evaluated categorically —and this is what follows from the literal interpretation of the definition; so coalescence of three segments into one (a1b2c3→x′1,2,3) is not more penalized than coalescence of two segments (a1b2c3→x′1,2c3). (15) INTEGRITY (“No breaking”) No element of S1has multiple correspondents in S2. For x ∈S1and w, z ∈S2, if xℜw and xℜz, then w = z. INTEGRITY is the matching constraint to UNIFORMITY, penalizing what McCarthy and Prince (1999) label «breaking» (a term applied to the diphthongization of vowels in the history of Old English, for example), though «splitting» might be a conceptually more neutral term —or indeed «scission», if we seek to match the Greco64 CatJL 4, 2005 Max W. Wheeler 5. An «anti-affricate» like [st] might be excluded by a high-ranked markedness constraint *[[+cont][–cont]]. But markedness could not select the correct ordering of contour features of vowel quality or of tone. Cat.Jour.Ling. 4 001-252 7/2/06 11:43 Página 64 Latin derivation of the remainder of the body of terms for both correspondence deviations (see Appendix) and correspondence constraints. The addition of another pair of correspondence constraints with the general label ADJACENCY is proposed by Carpenter (2002). ADJACENCY constraints are similar to CONTIGUITY constraints in that they penalize syncope or epenthesis. However, ADJACENCY also blocks coalescence. Within a specified domain, such as a syllable, ADJACENCY permits some cases of metathesis (a1b2→b′2a′1), including metathesis around a pivot (a1b2c3→c′3b′2a′1). (16) I-ADJACENCY(DOMAIN) (Carpenter 2002) If x is adjacent to y in the input, and x and y ∈Domain, then x′must be adjacent to y′in the output. Let x, y ∈S1and x′, y′∈S2. If xℜx′and yℜy′, and x is adjacent to y then x′is adjacent to y′. (17) O-ADJACENCY(DOMAIN) (Carpenter 2002) If x is adjacent to y in the output, and x and y ∈Domain, then x′must be adjacent to y′in the input. Let x, y ∈S2and x′, y′∈S1. If xℜx′and yℜy′, and x is adjacent to y then x′is adjacent to y′. In the Appendix I establish a taxonomy of segmental deviations from utterly faithful one-to-one correspondence, using largely the traditional terminology for phonetic «figures of speech». This is set out in a table showing which deviations are penalized by which constraints. 3. Trying again with deletion and coalescence in Navajo Lamontagne and Rice’s (1995) account of the Navajo D-effect requires a coalescence candidate to win in examples like /na+ii+d1+x2aa/ →[nei.1,2aa] ‘we look around’, while, for stop-initial roots, the winning candidate looks like a case of deletion: /ʔi+ii+d1+k2ááh/ →[ʔii.k2ááh] ‘we make a sand painting’. But the constraint ranking they offer, NOCODA » MAX » UNIFORMITY, actually entails that in the latter case the deletion candidate falls to some coalescence candidate, such as [ʔii.k1,2ááh]. Now, if the only coalescence candidate conceivable were precisely [ʔii.k1,2ááh], which is identical in pronunciation to Lamontagne and Rice’s preferred winner, the whole matter would be of little consequence. But as soon as it is accepted that some coalescence candidate can win, it is up to the analyst to demonstrate why it is this coalescence candidate that wins rather than some other, such as *[ʔii.1,2ááh]. Following up Lamontagne and Rice’s suggestions about featural alignment for the coalescence case, some appropriate feature Positional Faithfulness constraints can be proposed. The first is IDENTPA/RootInitial: “Correspondent consonant segments that are root-initial have the same Place of Articulation features”. In tableau (18) IDENTPA/RootInitial rules out coalescence candidates for /ʔi+ii+d+kááh/ which lack a root-initial velar, such as (18e). The second Positional Faithfulness conCluster Reduction: Deletion or Coalescence? CatJL 4, 2005 65 Cat.Jour.Ling. 4 001-252 7/2/06 11:43 Página 65 Tableau (25) illustrates the general pattern of affricate coalescence: some cluster constraint, such as here C*C¬ContrPA, together with MAX, outranks UNIFORMITY. Some faithfulness constraints concerning stridency and manner of articulation are also active. Tots [ œ tots] ‘all.MASC.PL’ with an affricate (25f) is better than alternatives with true deletion (25b, d) or non-affricate coalescence (25c, e). (Final coda voicing neutralization is undominated in Catalan.) (25) IDENTStrident, MAX, IDENTManner, C*C¬ContrPA » UNIFORMITY In the following tableaux I select examples that demonstrate which constraints are active in Catalan cluster reduction, and their ranking relative to one another. Tableau (26a) takes the citation form of verd ‘green’ to demonstrate that faithfulness to manner of articulation (IDENTManner), in fact, outranks C*C¬ContrPA; the winner (26a.iv) has a homorganic coda cluster. By contrast, in (26b), in pre-consonantal position within a phrase, the complex pre-consonant coda constraint *CC]σC comes into play, preferring a reduced candidate to a faithful one. By IDENTRhotic the candidate (26b.i) that preserves the rhotic wins over the alternative that preserves the stop. And the implied ranking IDENTRhotic » IDENTObstruent along with other constraints that favour sonorant codas, reflects the Syllable Contact Law —also an aspect of perceptual salience— by which the best pre-consonantal codas are sonorants, and the best onsets are obstruent stops. More precisely, there is an inherent ranking deriving from syllable-structure markedness IDENTCodaRhotic » IDENTRhotic, IDENTCodaObstruent. As mentioned previously, in the advanced variety of Catalan of Catalonia, for verd ‘green’, reduced [ œ bεr] is preferred to faithful [ œ bεrt]. This outcome follows from C*C¬ContrPA being promoted above IDENTManner. 72 CatJL 4, 2005 Max W. Wheeler tots ‘all.MASC.PL’ /tot1+z2/IDENTStrid MAX IDENTManner C*C¬ContrPA UNIFORMITY a. œ tot1s2*! b. œ tot1*! c. œ tot1,2 *! * * d. œ tos1*! e. œ tos1,2 *! * ☞ f. œ tots1,2 * Cat.Jour.Ling. 4 001-252 7/2/06 11:43 Página 72 (26) MAX, IDENTRhotic, *CC]σC » IDENTManner » C*C¬ContrPA » UNIFORMITY Faithful pre-consonantal triomf [tɾi œ of] ‘triumph’ (22f) also shows some faithfulness constraints outranking the complex pre-consonantal coda constraint *CC]σC as displayed in (27). (27) IDENTStrid, IDENTNasal » *CCσC » UNIFORMITY Now I can show the constraint hierarchy accounting for the pattern of citationform cluster reduction in pont [ œ pɔn] ‘bridge’ (22a) and molt [ œ mol] ‘much’ (22b). Tableau (28) for molt also shows the preservative effect of IDENTLateral, which in concert with some previously mentioned IDENT constraints, favours sonorant codas over obstruent ones. (28) IDENTLateral, C*C¬ContrCont » IDENTManner » UNIFORMITY Cluster Reduction: Deletion or Coalescence? CatJL 4, 2005 73 verd ‘green’ a. /vəɾ1d2/ i. œ vəɾ1*! ii. œ vəɾ1,2 *! * iii. œ vət1,2 *! * * ☞ iv. œ vəɾ1t2* b. /vəɾ1d2#C/ ☞ i. œ vəɾ1,2#C * * ii. œ vət1,2#C *! * * iii. œ vəɾ1t2#C *! * MAX IDENTRhotic *CC]σC IDENTManner C*C ¬ ContrPA UNIFORMITY triomf ‘triumph’ /tɾiom1f2#C/ IDENTStrid IDENTNas *CC]σCUNIFORMITY ☞ a. tɾi œ o1f2#C * b. tɾi œ o1,2#C *! * c. tɾi œ of1,2#C *! * molt ‘much’ /mol1t2/IDENTLat C*C¬ContrCont IDENTManner UNIFORMITY a. œ mol1t2*! ☞ b. œ mol1,2 ** c. œ mot1,2 *! * * Cat.Jour.Ling. 4 001-252 7/2/06 11:43 Página 73 Morphological analogy plays some role in the realization of Catalan consonant clusters, though morphological analogy is often overridden by phonological markedness. I simplify the issue here by mentioning just a PARADIGMUNIFORMITY (PU) constraint for the nominal number paradigm PUsg/pl, which abbreviates several constraints like O-ODEPC, O-OMAXC, O-OIDENTManner, or their Optimal Paradigms versions (McCarthy 2005; Pons 2004). Tableau (29) considers the cases of verds [ œ vəɾts] ‘green.MASC.PL’, molts [ œ mols] ‘many.MASC.PL’ and triomfs [tɾi œ ofs] ‘triumphs’, whose final clusters display affrication, reduction and faithfulness respectively. In (29a.ii) [ œ vəɾts] wins as the plural of [ œ vəɾt]. However, in (29b) the parallel *[ œ molts], which beats [ œ mols] on C*C¬ContrCont (assuming [l] is [–cont] only next to a homorganic stop; see Wheeler 2005: chapter 10) and on IDENTManner, loses by PARADIGMUNIFORMITY: the singular is realized [ œ mol] so it is better to construct the plural on this form. The final complex cluster in [tɾi œ ofs] violates several clustermarkedness constraints, but it is better to maintain the form of the stem [tɾi œ of], and also the sibilance of the suffix /+z/, than to simplify the cluster.9 (29) IDENTSib, PUsg/pl » C*C¬ContrCont » IDENTManner » C*C¬ContrPA » UNIFORMITY 74 CatJL 4, 2005 Max W. Wheeler 9. In the more conservative variety in Catalonia which for verd ‘green’ has [ œ bεrt] in the singular but [ œ bεrs] in the plural, the complex coda constraint *THREE-MANNER CODA (*CCC]σ: “There is no more than one point where change of Manner of Articulation occurs within a coda”) outranks PUsg/pl. a. /vəɾ1d2+z3/ sg. [œvəɾt] i. œvəɾ1t2s3**! ii. œvəɾ1,2s*!**** ☞iii. œvəɾ1ts2,3 ** b. /mol1t2+z3/ sg. [œmol] i. œmol1t2s3*! * ** ☞ii. œmol1,2s**** iii. œmol1ts2,3 *! * * c. /tɾiom1f2+z3/ sg. [tɾiœof] ☞i. tɾiœo1f2s3** ii. tɾiœom1,2s*!* * iii. tɾiœof1,2s*!*** iv. tɾiœoms2,3 *! * v. tɾiœof2,3 *! * * IDENTSib PUsg/pl C*C ¬ ContrCont IDENTManner C*C ¬ ContrPA UNIFORMITY Cat.Jour.Ling. 4 001-252 7/2/06 11:43 Página 74 Yet, while we observe some complex clusters can be retained in Ibiza Catalan in utterance-final forms, reduction is more widespread in forms uttered before a following consonant. In this context (30), illustrating pre-consonantal porcs ‘pigs’, we see that the *THREE-MANNER PRE-CONSONANT CODA (*3MAN]σC) constraint outranks PARADIGMUNIFORMITY. The faithful candidate (30a) which also contains the singular [ œ pɔɾk], loses on *THREE-MANNER PRE-CONSONANT CODA (*3MAN]σC); candidate (30c) which also recapitulates the singular form, at the cost of losing the plural marker, is excluded by high ranking IDENTSibilant. (30) IDENTSib, IDENTRhotic, *3MAN]σC » PUsg/pl » *CC]σC The overall ranking of the constraints considered in this section is as follows (31): (31) MAX, IDENTNasal, IDENTLateral, IDENTRhotic, IDENTStrident, IDENTSibilant, *THREE-MANNER PRE-CONSONANT CODA (*3MAN]σC) | PUsg/pl | C*C¬ContrCont, *CC]σC | IDENTManner | C*C¬ContrPA | UNIFORMITY The faithfulness constraints at the top of the ranking protect coda sonorants, stridents ([s, f]) and sibilants [s, ʃ] from cluster reduction effects. High-ranking MAX means true deletion is not acceptable as a repair to coda complexity of any type. PARADIGMUNIFORMITY requiring singular and plural stems to match in form ranks high, but below at least one coda cluster constraint *THREE-MANNER PRECONSONANT CODA (*3MAN]σC). Other cluster constraints stand above UNIFORMITY, though IDENTManner is interleaved between them to protect a cluster like [ɾt] that Cluster Reduction: Deletion or Coalescence? CatJL 4, 2005 75 /pɔɾ1k2+z3#C/ sg. [ œ pɔɾk] IDENTSib IDENTRhotic *3MAN]σC PUsg/pl *CC]σC a. œ pɔɾks#C *! * b. œ pɔk1,2s#C *! * * c. œ pɔɾk2,3#C *! * ☞ d. œ pɔɾ1,2s#C * * e. œ pɔɾ1,2,3#C *! * Cat.Jour.Ling. 4 001-252 7/2/06 11:43 Página 75 lacks a place contrast. An important conclusion from these observations is that, when UNIFORMITY is ranked relatively low (and in any case below MAX), cluster reduction that looks like deletion is in fact coalescence. It is not necessarily the case, however, that in a language like Catalan reduced clusters always display coalescence. In the Majorcan variety I examine in more detail in Wheeler (2005), gust ‘taste’ [ œ ust] is realized [ œ us1] (true deletion) in pre-consonantal position, while its plural gusts [ œ uts1,2,3] is realized [ œ ut1,2,3] (coalesced) in pre-consonantal position. In Majorcan, while MAX outranks UNIFORMITY, as elsewhere in Catalan, several constraints outrank MAX, and some cluster constraints (*3MAN]σ, *2MAN]σC) are undominated. 5. Concluding observations One aspect of the current conception of Optimality Theory is that the GEN component may supply candidates that are pronounced identically but that differ either in prosodic organization (syllableand foot-structure, and so on) or in correspondence relations.10 Here I have drawn attention to the latter type of alternatives differing in correspondence, and have attempted to demonstrate that a coherent account of phonological patterns cannot simply ignore the alternatives not favoured by the analyst. Is the theory too rich, in allowing such a plethora of candidates? Probably not, in that there are good reasons why both true deletion and coalescence have been identified as effective «repairs» to violations of well-founded complexity constraints. Though I have not investigated the issue in this paper, the same logic requires that GEN freely offers «breaking» candidates, that is, those with an INTEGRITY violation. Thus, in a language where breaking candidates can sometimes win —for example, when gemination of vowels or consonants is a means of satisfying the Stress-to-Weight principle— it is up to the analyst to demonstrate what constraints outrank INTEGRITY so as to prevent breaking candidates from winning across the board. In my attempt to fix up Lamontagne and Rice’s account of the D-effect in Navajo in the light of these observations, I observed that the version of the account where UNIFORMITY outranks MAX, allowing a true deletion candidate to win in some circumstances, also requires invoking constraints of the MAX(Feature) type. While MAX(Feature) constraints are not shown to be necessary in the account I offer here of cluster reduction in Ibiza Catalan, where MAX(Segment) ranks high and many coalescence candidates win, they are generally likely to be appropriate in languages where MAX(Segment) ranks lower. And in fact the account of cluster reduction in Majorcan Catalan in Wheeler (2005: chapter 7) where MAX(Segment) ranks much lower than in the Ibiza variety, though still above UNIFORMITY, does have recourse to MAX(Feature) constraints. I believe the necessary approach to cluster reduction in general adds weight to the case not yet 76 CatJL 4, 2005 Max W. Wheeler 10. In Wheeler (2005) I do attempt to show what constraint rankings exclude inappropriately syllabified candidates. Cat.Jour.Ling. 4 001-252 7/2/06 11:43 Página 76 universally accepted for including in phonological theory constraints of the MAX(Feature) type. The description and analysis I have given of cluster reduction in Ibiza Catalan takes for granted the position universally adopted in the literature on Catalan that there is an affrication process involving coalescence, when a coronal obstruent stop is followed by a coronal strident fricative, so that, for example, /tot1+z2/ tots ‘all.MASC.PL’ is realized [ œ tots1,2] with coalescence, or, before a vowel-initial word [ œ tod1.dz1,2] with coalescence and breaking (Wheeler 2005: §3.1). It may be that this interpretation should be re-examined, as it is not clear what objective observations it is founded on. Is there any empirical consequence of the choice between the representations [ œ tots1,2] («affrication with coalescence») and [ œ tot1s2] («no affrication or coalescence») (or indeed [ œ tot1s2] —«affrication without coalescence»), or of the choice between [ œ tod1.dz1,2] and [ œ tod1.z2] (or [ œ tod1.z2]) for the pre-vocalic realization? My suspicion is that there is not, once assimilation of place is assumed —Catalan coronal obstruent plosives are typically dental, while coronal sibilants are alveolar ([s]) or alveolo-palatal ([ʃ]), but clusters of coronals always share place. Perhaps it is right, then, in the analysis of languages like Catalan which display an affrication process, to no longer assume without argument that winning candidates are those with a representation such as [ œ tots1,2], rather than [ œ tot1s2] or [ œ tot1s2]. Cluster Reduction: Deletion or Coalescence? CatJL 4, 2005 77 Cat.Jour.Ling. 4 001-252 7/2/06 11:43 Página 77 Appendix Table 1 consists of a taxonomy of segmental deviations from faithful one-to-one correspondence. The input strings given are assumed to comprise the whole domain, where this is relevant. Asterisks indicate which constraints are violated by the deviation in question; (*) denotes possible, even likely, penalization, contingent on the content of the corresponding strings. 78 CatJL 4, 2005 Max W. Wheeler Table 1. Deviations from perfect segmental correspondence with the correspondence constraints that penalize them. I-CONTIG O-CONTIG I-ADJ O-ADJ R-ANCHOR L-ANCHOR LINEARITY UNIFORMITY INTEGRITY IDENT(F) MAX DEP Cat.Jour.Ling. 4 001-252 7/2/06 11:43 Página 78 Cluster Reduction: Deletion or Coalescence? CatJL 4, 2005 79 Table 1. Deviations from perfect segmental correspondence with the correspondence constraints that penalize them. I-CONTIG O-CONTIG I-ADJ O-ADJ R-ANCHOR L-ANCHOR LINEARITY UNIFORMITY INTEGRITY IDENT(F) MAX DEP Cat.Jour.Ling. 4 001-252 7/2/06 11:43 Página 79 80 CatJL 4, 2005 Max W. Wheeler Table 1. Deviations from perfect segmental correspondence with the correspondence constraints that penalize them. I-CONTIG O-CONTIG I-ADJ O-ADJ R-ANCHOR L-ANCHOR LINEARITY UNIFORMITY INTEGRITY IDENT(F) MAX DEP Cat.Jour.Ling. 4 001-252 7/2/06 11:43 Página 80 References Beckman, Jill N. (1998). Positional Faithfulness. University of Massachusetts, Amherst, doctoral dissertation. [Available on Rutgers Optimality Archive #234, http://roa.rutgers.edu/. Published as Positional Faithfulness: An Optimality Theoretic Treatment of Phonological Asymmetries in 1999 by Garland.] Carpenter, Angela C. (2002). «Noncontiguous Metathesis and Adjacency». In: Carpenter, Angela C.; Coetzee, Andries W.; de Lacy, Paul (eds.). University of Massachusetts Occasional Papers in Linguistics 26: Papers in Optimality Theory II. Amherst, MA: Graduate Linguistic Student Association, pp. 1-25. Côté, Marie-Hélène (2001). Consonant Cluster Phonotactics: A Perceptual Approach. MIT, doctoral dissertation 2000, revised. [Available on Rutgers Optimality Archive #548, http://roa.rutgers.edu/.] Dols Salas, Nicolau (2000). Teoria fonològica i sil·labificació. El cas del català de Mallorca. Universitat de les Illes Balears, doctoral dissertation. Jiménez, Jesús (1999). L’estructura sil·làbica del català. València: Institut Interuniversitari de Filologia Valenciana; Barcelona: Publicacions de l’Abadia de Montserrat. Kager, René (1999). Optimality Theory. Cambridge: Cambridge University Press. Lamontagne, Greg; Rice, Keren (1995). «A correspondence account of coalescence». In: Beckman, Jill; Walsh Dickey, Laura; Urbanczyk, Susan (eds.). University of Massachusetts Occasional Papers in Linguistics 18: Papers in Optimality Theory. Amherst, MA: Graduate Linguistic Student Association, pp. 211-223. Lombardi, Linda (2001). «Why Place and Voice are Different: Constraint-Specific Alternations in Optimality Theory». In: Lombardi, Linda (ed.). Segmental Phonology in Optimality Theory. Cambridge: Cambridge University Press, pp. 13-45. McCarthy, John J. (1995). «Extensions of Faithfulness: Rotuman Revisited» [title page: «Faithfulness in Prosodic Morphology & Phonology: Rotuman Revisited»]. University of Massachusetts, Amherst, MA, unpublished manuscript. [Available on Rutgers Optimality Archive #110, http://roa.rutgers.edu/.]. McCarthy, John J. (2003). «OT Constraints are Categorical». Phonology 20: 75-138. McCarthy, John J. (2005). «Optimal Paradigms». In: Downing, Laura J.; Hall, T.A.; Raffelsiefen, Renate (eds.). Paradigms in Phonological Theory. Oxford: Oxford University Press, pp. 170-210. [Prepublication 2001 version available on Rutgers Optimality Archive #485, http://roa.rutgers.edu/.] McCarthy, John J.; Prince, Alan S. (1993). «Generalized Alignment». In: Booij, Geert; van Marle, Jaap (eds.). Yearbook of Morphology 1993. Dordrecht: Kluwer, pp. 79153. [Available on Rutgers Optimality Archive #7, http://roa.rutgers.edu/.] McCarthy, John J.; Prince, Alan (1995). «Faithfulness and Reduplicative Identity». In: Beckman, Jill; Walsh Dickey, Laura; Urbanczyk, Suzanne (eds.). University of Massachusetts Occasional Papers in Linguistics 18: Papers in Optimality Theory. Amherst, MA: Graduate Linguistic Student Association, pp. 249-384. [Available on Rutgers Optimality Archive #60, http://roa.rutgers.edu/.] McCarthy, John J.; Prince, Alan S. (1999). «Faithfulness and Identity in Prosodic Morphology». In: Kager, René; van der Hulst, Harry; Zonneveld, Wim (eds.). The Prosody-Morphology Interface. Cambridge: Cambridge University Press, pp. 218-309. Cluster Reduction: Deletion or Coalescence? CatJL 4, 2005 81 Cat.Jour.Ling. 4 001-252 7/2/06 11:43 Página 81