the followin! vowel a. [+anterior] { s z } before [i] b. [ - PDF document

the followin! vowel a. [+anterior] { s z
the followin! vowel a. [+anterior] { s z } before [i] b. [Ðanterior] { " # } elsewhere (4) Gnztn #2armony within the stem makes sibilants deviate from the reement, nd with each otherÕ For each distinct pair of output consonants X and Y, assiOne violation for each [+anterior] sibilant in the output.5 (37) The basic allophonic distribution requires *"i to dominate(and both must dominate faithfulness for anteriority too) *"i *s IDENT-[ant]made based on the anteriority value of the output b. As lon! as thereÕs a [+anterior] sibilant, a!reement will conver!e on it CCáANCHOR-Ôcause to be eatenÕ (Ôlon!Õ causative suffix invariably has [s]) (62) Both Hyman (2003) and Hansson (2010) note that [s] may occur before vowels other than [i] as the result of a covert ÔshortÕ causative suffix some formsConsider a hypothetical input /-sa!i/ (cf. -!ˆˆ!ˆ/-sˆˆs“r Ôbe in pain (perf.)Õ): here, rá[+sibilant] CCáIDENT-[ant](1~0) W (0~1) (74) There are 16 possibilitiesfor inputs with two sibilants, for all permutations of [+anterior] and [-anterior] on each, followed either by [i] (the neutralisation context) or [a] (representin! non-neutralisin! contexts). (75) In 15 of these 16 cases, a!reement that respects the ri!htmost sibilantÕs behaviour falls out from basic faithfulness and neutralization, and/or from These affixes cause an assortment of consonant mutations in various eastern Bantu lan(85) Not all suffixes with /-i/ cause systematic mutations:13 12 For details on Kinyarwanda, see Kimenyi (1979), and Walker et al. (2008); for others, see Hyman (2003). 13 Some applicative forms do seem to have at least the [anterior] to [+anterior] shift for sibilan

ts, but it doesnÕt appear to be consiste
ts, but it doesnÕt appear to be consistent. For instance, TaylorÕs (1959) dictionary !ives the form [left directionality can be understood as harmony respectin! the idiosyncratic phonolo!y that ORRá[+sibilant] CCáIDENT-[ant]NCHOR *"i IDENT-[ant]he vowels in all three of these suffixes were historically super-hi!h vowels NCHOR-R handles cases where the ri!htmost sibilant is faithful b. MORPHREAL handles cases where itÕs mutated by the morpholo!y Conclusions for the bi!!er picture: (101) At least some kinds of directional asymmetries can be derived without havin! the basic mechanism of ABC (CORR, CCáIDENT las Pulleyblank. (2012). Emerent phonolo!y: Evidence from En!lish. In Issues in En!lish Lin!uistics, ed. by Ik-Hwan Lee, Youn!-Se KanLee, Hye-Kyun! Kan!, & Sun!-Ho Ahn, 1Santa Cruz dissertation. Mpiranya, Feli, Diana, and Doulas Pulleyblank. (To appear b). Tonal allomorphy in Kinande. ical Shades within and across Lanes: Inspirations from the Theoretical Bennett, Wm. G. (2013). Dissimilation, Consonant Harmony, and Surface Correspondence. PhD Hansson, G. î. (2001). Theoretical and typoloical issues in consonant harmony. . î. (2010). Consonant Harmony: LonDistance Interaction in Phonolo. Number 145 in University of California Publications in Linuistics. University of California Hyman, Larry M. . Sound chane, misanalysis, and analoy in the Bantu causative. Kimenyi, A. (1979). Studies in Kinyarwanda and University of California Kurisu, Kazutaka, . They of Morpheme Realization. University of California Mpiranya, Fand Rachel Walker. (2005). Sibilant harmony in inyarwanda and coronal opacity. Handout of paper presented at GLOW 28, University of Geneva, March 31, Nevins, Andrew, . Condition

s on (Dis)Harmony. Massachusetts Institu
s on (Dis)Harmony. Massachusetts Institute of TechnoloPoletto, Robert E.. Topics In Runyankore Phonolohe Ohio State UniversityPrince, A. (2002). Aroptimality. In Coetzee, A., Carpenter, A., and de Lacy, P., editors, Rose, S. (2004). A typoloy of consonant areement as Taylor, Charles V. A simplified Runyankorelish and EnEn Walker, Rachel. (2000a). Lon!-distance consonantal identity effects. WCCFL 19.532Ð545. Walker, R. (2000). Yaka nasal harmony: Spreadinor semental correspondence? Walker, R. (. Consonantal correspondence. In s of the Workshop on the Lexicon in Phonetics and Phonolo, volume 6 of Papers in Experimental and Theoretical Walker, R, DByrd, and Fidle Mpiranya. (2008). An articulatory view of Kinyarwanda Wilson, C. (2001). Consonant cluster neutralisation and tareted constraints. UC Berkeley Phonology Lab Annual Report: ABCConference92 5. Summary & conclusionKeypointsweÕveshowthistalk:(98)a raises an interestinconundrum for how directionality needs to be handled in ABCItÕs strictly rileft, with no apparent basis for beinreement for both valuesnot direction areement pattern like many other cases of directional harmonyNot explainable as faithfulnessdriven control, since itÕs allophonic(99)But actually,it only raisthat problem in some other possible worldSibilants donÕtreally have an allophonic distributionin KiMorpholoical effects just make it look that way(100)The pattern we actuallyfind in Kibe explained in ABCNCHORR handles cases where the rihtmost sibilant is faithfulORPHEAL handles cases where itÕs mutated by the morpholoConclusionsforthepicture:(101)At least some kinds of directional asymmetries can be derived without havinthe basic mechanism of ABC (ORR, CCáDE

NT constraints) refer to directionality(
NT constraints) refer to directionality(alonthe lines suested in Bennett 2013)(102)seems like one of the best cases to ue that the theory of Surface Correspondence needs torefer to directionality, but it doesnÕt hold up as such. (103)Maybe other cases of strict directionality can be handled in similar ways, without needinto adjust the theoryReferenceseli, Diana, andlas Pulleyblank. (2012). Emerent phonoloy: Evidence from lish. In Issues in Enlish Lin, ed. by IkHwan Lee, YounSe Kan, KyounKim, KeeHo Kim, IlKon Kim, SeonHa Rhee, JinKim, HyoKim, Kieli, Diana, and Doulas Pulleyblank. (To appear a. Autosments in EmerUC Berkeley Phonology Lab Annual Report: ABCConference91.2. Is there aprincipled basis for thismorpholoical story(92)For at least the short causative forms, the answer seems to be a definite yes.(93)In most cases, the only surface evidence for the // causative is the mutationit induces on a previous consoÔworkÕÔmake (X) workÕfundaÔbe narrowÕ funzaÔrestrictÕtaahaÔenterÕ taasjaÔtake inÕ(94)For this affix, failure to manifest the mutations means leavinthe morpheme totally unexpresseda canonical type of EALIZEORPHEMEviolati(95)For the other suffixes that cause mutations(perfective ire and nominal the situation is a little murkier(96)But, there still seems to be some basis for it:he vowels in all three of these suffixes were historically supervowels in proto(Hyman 2003)e, *, and *The hih vs. superh contrast would have been crucial for morpholoicaldisambiuation iprotoBantu (e.. applicative *vs. perfective As the hih vs. superh contrast was lost in Kia, the functional load for these distinctions would have shifted from vowel heiht to the spirantizations induced by superh vowel

sOn this interpretation, the mutations c
sOn this interpretation, the mutations caused by suffixes like perfective ire/ would have been crucial for realizinthese morphemes(even if itÕs been rendered redundantby subsequent chanes in Ki(97)The point: in cases where harmony is left, but cannot be explained by rihtmostfaithfulness, there is a demonstrably morpholoical source to the observed directionaliUC Berkeley Phonology Lab Annual Report: ABCConference90(86)The mutations caused by these suffixes have an important consequence for harmony: when the rihtmost sibilant mutated from // to [s], the reeinsibilants are not on equal footin(87)In a form like /ire [saasire] Ôbe in pain (perf.)Õ, one sibilant has its form dictated by the perfective suffix. The other sibilant does not(88)In this situation, rileft directionality can be understood as harmony respectinthe idiosyncratic phonoloy that oes with these morphemes(89)We can understand it as a morphemerealization effect:he mutations caused by the perfective, aentive nominal, and causative suffixes an interal part of realizinthese morphemes, Then failinto implement them would violate a constraint on morpheme realization call it ORPHEAL(90)ORPHEAL can break the directionality tie in exactly the same way as NCHORInput: /Output: [ORPHEALORRá[+sibilant]DENT[ant]NCHO*"i DENT[ant]! a. SCorr(0)(0)(1)(0)(2) b. SCorr(0~1!) (1~0(0~1(2~0c. SCorr (0~1! DENT(1~1) (2~1d. SCorr (0~1! ORR(1~1) (2~1(91)This is a familiar control type of interaction.ORPHEAL nails down the rihtmost (rootfinal) sibilant as [+anterior]By doinso, it forces other sibilants to assimilate to match the rootfinal one leadinto RiLeft harmonyWe intend this in the same spirit as KurisuÕs (2001)

EALIZEORPHEMEbut abstract away from form
EALIZEORPHEMEbut abstract away from formal details in implementation that are outside the scope of the Kia harmony direction puzzle. The exact model we are currently explorin is one proposed in recent work by Archaneli & Pulleyblank (2012, in press, to appear a, tappear b). In that model, choosin between allomorphs is central to the theory, and the choice may be dictated by phonotactics in some cases and by morphemespecific selectional requirements in others. The cases covered here by ORPHEALwould be the latter kind of pattern.UC Berkeley Phonology Lab Annual Report: ABCConference894.1. Morpholoicallyconditioned mutations(81)The examples that crucially show rootfinal / surfacinas [s z] come overwhelminly from forms with three specific suffixes:Perfective /ire/entive nominalizer /Causative /(82)These affixes cause an assortment of consonant mutations in various eastern Bantu lanes, e.. Kinyarwanda,Kinande, Haya, Bemba, and others.(83)a is no exception: thethree affixes inducea host of consonant includinAffrication of stopsAssibilation of /r/ and /h/Frontinof // to [s z](84)Consonant mutations causedby certain suffixes with /(85)Not all suffixes with /i/ cause systematic mutations For details on Kinyarwanda, see Kimenyi (1979), and Walker et al. (2008); for others, see Hyman (2003).Some applicative forms do seem to have at least the [anterior] to [+anterior] shift for sibilants, but it doesnÕt appear to be consistent.For instance, TaylorÕs (1959) dictionary ives the form [] as in (), but in ylorÕs (1985) rammar, the form [] can be found as well, alonwith a few other such examples (such as s -iziraÔcome for/toÕ, from [a] ÔcomeÕ). Taken t

oether with the nonmutated forms in (),
oether with the nonmutated forms in (), it appears that these are not systematic chanes, and this is corroborated by the lack of mutation for other consonants.But even if applicatives did exhibit mutations, this could potentially be explained as analoical extension.UC Berkeley Phonology Lab Annual Report: ABCConference88(76)Inputs and explanatory covera(77)itraryleft directionality issueonly arises in one input, /i/.(78)Why?The rihtmost sibilant must be unfaithful, otherwise directional harmony can be handled with CNCHORThe precedinsibilant must also be unfaithful; if it isnÕt, it means harmony falls out just from neutralizationof the rihtmost one onlyand no assimilation is crucially happeninThe precedinsibilant must be in a where we expect itto be faithful we only see a crucial effect of harmony if it deviates from the normal mappinOnly /,e,o,u}i/ inputs fit this profile(79)PointThe only case that really needs explaininin order to handle the Kidata are inputs like /i/ that surface as [4. The role of morpholo(80)The explanation we propose for the /i/[sasi]type cases is that they are in part the result of morpholoical effectsUC Berkeley Phonology Lab Annual Report: ABCConference87(71)The same is true for inputs like /isa/: neutralization yields areement, even if the harmony constraints donÕt play a roleInput: /isa/Output: [sisa]ORRá[+sibilant]DENT[ant]*"i DENT[ant]*s! a. s1is1a SCorr!:{s s} (0)(0)(1)(2) b. "1i"1a SCorr!:{" "} (0~1!)(1~1)(2~0)c. "1is1a SCorr!:{" s (0~1! DENT(0~1 (2~1)d. "1is2a SCorr!:{"}{s (0~1! ORR(0~1 (2~1)(72)Point: if itÕs neutralization instead of allophony, then some of the mappinrequired to reement fall out for free(73)We can fu

rther restrict the problem by considerin
rther restrict the problem by considerinNCHORin cases where the rihtmost sibilant is faithful, rileft harmony can be handled by this constraintInput: /Output: [a]ORRá[+sibilant]DENT[ant]NCHOR*"i DENT[ant]*s! a. "1i"1a SCorr!:{" "} (0)(0)(1)(1)(0) b. s1is1a SCorr!:{s s} (0~1(1~0(1~1(0~2c. s1i"1a SCorr!:{s "} (0~1! DENT (0~1(1~0(0~1d. s1i"2a SCorr!:{s}{"} (0~1! ORR (0~1(1~0(0~1(74)here are 16 possibilitiesfor inputs with two sibilants, for all permutations of [+anterior] and [anterior] on each, followed either by [i] (the neutralisation context) or [a] (representineutralisin(75)In of these reement that respects the rihtmost sibilantÕs behaviourfallout from basic faithfulness and neutralization, and/or from faithful protection of the rihtmost correspondentUC Berkeley Phonology Lab Annual Report: ABCConference86(65)Key points from this data:[s] is attested before all the vowels, and so is [{s z} are underrepresented before [e] and [u], but not [a] or [o]} are stronly underrepresented before [i] (but [i] sequences arettested, and #] is also slihtly represented before [e])(66)This isnÕt indicative of systematic and active allophonyThe ban on * and * before [i] (and [j]) seems clearly supportedThe ban on *s and *z beforeallother vowels isnÕt supportedhis oks much more like neutralization of /i/ to [si]rather than the full twoway neutralization of an allophonic pattern. Lack of allophony chanes the problem(67)If the basic distribution of [s]~[] and [z]~[isnÕt allophonic, then it doesnÕt set up the arbitrary directionality problem in the same way.(68)To illustrate this, consider an system like the one reported for Kia, but with way neutralizationinstead of

twoway allophonyi/and /si/ neutralizeto
twoway allophonyi/and /si/ neutralizeto [si]and /a/ surfacefaithfullyneutralization for /s z/(69)If there is neutralization of this sort, then some subcases of areement be explained without harmony beindirectionspecific(70)Consider a hypothetical input /i/ (cf. ˆˆˆ/sˆˆs“r Ôbe in painperfhere, rleft areement falls out just from neutralizationInput: /i/Output: [sasi]ORRá[+sibilant]DENT[ant]*"i DENT[ant]*s! a. s1as1i SCorr!:{s s} (0)(0)(1) (2)b. "1a"1i SCorr!:{" "} (0~1!)(1~1)(2~0)c. s1a"1i SCorr!:{s "} (0~1! DENT(0~1 (2~1)d. s1a"2i SCorr!:{s}{"} (0~1! ORR(0~1 (2~1) UC Berkeley Phonology Lab Annual Report: ABCConference85()zˆndÔits (cl.10, poss. cl.9)Õ Ôthat (cl.10)ÕÔthis (cl.10)Õ(class 10 prefix invariably has [z], never *[Ôpasture; cause to eatÕibwÔcause to be eatenÕ(ÔlonÕ causative suffix invariably has [s])(62)Both Hyman (2003) and Hansson (2010) note that [s] may occur before vowels other than [i] as the result of a covert ÔshortÕ causative suffix The [sa] from /sa/ analysis is possible if unintuitive for some formslike [˜rsˆ] ÔpermissionÕ (maybe from Ôin vainÕ?But a covert causative // analysis doesnÕt work in most cases:Nonrootfinal [s], as in [ks˜˜mˆ] Ôdo astampindanceÕInvariable class 10 [z] in] inz˜] and [z“], which demonstrativeshe ÔlonÕ causative would need to be treated as a double causative in all cases(63)Observed/Expected ratios calculated from TaylorÕs (1959) dictionary11 654 stems (nonlemmatized), diitized by CBOLDn O/E value of 1.0 means a combination occurs as often as expected from the individual frequencies of e

ach soundO/E values less than 1 mean a c
ach soundO/E values less than 1 mean a combination is underrepresented; reater than 1 means itÕs overrepre(64)O/E ratios for sibilantvowel combinationsin stems i e a o u j s 0.90.31.61.00.11.9z 1.10.21.41.10.00.7# 0.11.61.51.83.40.0$ 0.00.71.32.66.40.0t!s 0.90.50.60.40.08.8t!" 1.41.70.10.71.40.11.5 and 0.5 in boldface; 0.5 also shadedMore on this in ¤4It doesnÕt explain why there seems to be a threeway [s]~[]~[sj] contrast, thouh.UC Berkeley Phonology Lab Annual Report: ABCConference84 3. Reevaluatinthe data: a problem that isnÕt3.1. The empirical basis for allophony(58)Earlier descriptions of the Nkorea sibilant patterns come from Hyman (2003) and Hansson (2001), and the orial source data comes from a dictionary and rammar by Taylor (1959, 1985)(59)A search of TaylorÕs rammar turns up numerous examples that undermine the notion that sibilants in Nkorehave an allophonic distribution.(60)Some minimal and nearminimal pairs:(Taylor 1959, 1985)ÔchopÕÔstrike (of lihtni“zˆÃ”ease, darkenÕÔcomeÕÔspoil, do wron, sinÕs’“sˆÃ”ask for meatÕs˜˜mˆÃ”do a stampindanceÕ˜mˆÃ”read, attend churchÕ˜mÃ”swarm, beehiveÕ˜mÃ”ambitionÕÔwander, bedyinÔbecome soddenÕ˜rÃ”permissionÕ(˜)Ôin vainÕ˜rœsjˆÃ”newÕ(61)While some examples do show [s]~[] alternations, there are a number of other morphemes which systematically fail to do so:“rÃ”usefulÕ/Ôuseful (perf.)ÕÔable toÕ/ÔpossibleÕ, but not PolettoÕs (1998) Ph.D. dissertation explores the phonoloy of Runyankore in detail, but makes no mention of sibilant harmony or the allopho

ny.UC Berkeley Phonology Lab Annual Rep
ny.UC Berkeley Phonology Lab Annual Report: ABCConference83(55)In contexts where the rihtmost sibilant is faithful, and still controls harmony, this explanation simply doesnÕt et the desired resultInput: /sisa/Exp: [a]NCHORORR[+sib]DENT[ant]*"i *sRemarks! a. s1is1a SCorr!:{s s} (0)(0)(0)(0)(2)R harm b. "1i"1a SCorr!:{" "} (0~1!) (0~1)(2~0)L harmc. s1i"1a SCorr!:{} (0~1!) (0~1!)(0~1!)(2~1)corr, no harmd. s1i"2a SCorr!:{s} (0~1!)(0~1) (0~1)(2~1)nocorr2.4. Section recap(56)The harmony pattern reported in Kia has a strict directionality that seems to be totally arbitrary.The rihtmost sibilant needs to be allophonically conditioned, yet still control harmonyThis doesnÕt emere from a basic ABC analysis: the two areement possibilities [sÉs] and [] tie on the harmony constraintsUsindirectionspecific harmony constraints doesnÕt break the tie(in order to disambiuate, the harmony constraints would need to penalize areement)Lettineneral markedness constraints break the tie leads to valuedominant harmony, with flexible (not fixed) directionalityHavinfaithfulness break the tie doesnÕt work in all cases, because sometimes we need the controllinsibilant to be unfaithful(57)How do we solvethis problem?wo possibilities:he areementbased approach of ABC is fundamentally wron, and we should abandon it for a more processbased theory . Nevins 2004)he facts of Kia are wron. (WeÕre oinitÕs this one.)UC Berkeley Phonology Lab Annual Report: ABCConference82(50)A directional reformulation of CCáDENT[anterior]:DENTT+anterior]: ÔPrecedincorrespondents aree for [+anterior]For each distinct pair of output consonants X and Y, assin a violation if:X and Y are in the sam

e surface correspondence class Y precede
e surface correspondence class Y precedes XX is [+anterior]iv.Y is [anterior](51)ReplacinDENT[anterior] with desinatedly RiLeft CDENTconstraints doesnÕt solve the problem(52)DENTonstraints are still satisfied by both areement candidates:Input: /Exp: [a]ORR[+sib]DENTTÐant] CRCLáIDENT[+ant]*"i *sRemarks! a. s1is1a SCorr!:{s s} (0)(0)(0)(2)R harm b. "1i"1a SCorr!:{" "} (0~1!)(2~0)L harmc. "1is1a SCorr!:{" (0~1! [+ant])(0~1)(2~1)corr, no harmd. "1is2a SCorr!:{"}{s} (0~1!) (0~1)(2~1)no corr(53)Option 3:Bennett (2013) proposes a positional faithfulness constraint, NCHORR, that protects the rihtmost correspondent in a class(54)NCHOR[anterior]: Ôif a C is rihtmost amonits SCorr class, it is faithful for anteriorityÕFor each distinct pair of output consonants X and Y, assin a violation if:X and Y are in the same surface correspondence class X precedes YThere is no other Z that corresponds with Y and is preceded by Yiv.The [±anterior] value of Y differs from the [±anterior] value of its input correspondent YUC Berkeley Phonology Lab Annual Report: ABCConference81The harmony constraints are satisfied equally well by both ways of doinment; they only rule out (c) & (d)the choice between [sÉs] and [] ets passed down to it favors the harmonic candidate with fewer [i] sequences(which is (a), [sÉs])(43)This is valuedominant armony, not directional harmony:The choice betweenthe areeincandidates is beinmade based on the anteriority value of the outputAs lonas thereÕs a [+anterior] sibilant, areement will convere on it(44)Point: if *i is what makes the decision, then the directionality of harmony will chane to suit the preferred form of areement (to avoid [i])and

that isnÕt the riht eneralization.2.3.
that isnÕt the riht eneralization.2.3. ThereÕs no simple (45)We know of no straihtforward way to et a strictly directional pattern to fall out from this kind of system without radical chanes to the constraints(46)Option 1: Hansson (2001/2010) uses tareted constraintsWilson 2001DENT[anterior] favors areement only if itÕs attained by chanthe sement on the left to match the one on the riht in anteriorityeted constraints like thisnecessarily do not assin countable violationsThis analysis also requires other markedness constraints that interact with harmony (i.e. *i) to be tareted constraints as well(47)Option 2Rose & Walker (2004) build directionality into CCáDENTLeftht CDENT[+F]alizes [+FÉLeft CDENT[+F]alizes [FÉ+F](48)Option Other work (Walker 2000b, 2001; Hansson 2001/2010) builds directionality into the correspondence relation instead, and has it percolate throuto the areement constraints, but the result seems to be the same.(49)directional constraints can restrict where harmony applies, but they canÕt impart directional control of the sort we seem to need for KiWe arenÕt the first to point this kind of indeterminacy out; see Hansson (2001:341).NB: in order to make a meaninful directionality distinction, these areement constraints need to be valuespecific. Rose & Walker did this usin a privative feature rather than a binary one like IÕm doin, but this is just a notational variance.UC Berkeley Phonology Lab Annual Report: ABCConference80(38)n order for harmony to force sibilants to deviate from the usual allophonic pattern, the ORRand CCáDENTconstraints that drive harmony must dominate(some of)the constraintsresponsible for the allopho(39)But, rankinORRá[+sibilant] and CCáDEN

T[anterior] over *i and *s canÕt produc
T[anterior] over *i and *s canÕt produce strictly directional harmony(40)i favors rileft harmony in inputs endinin /ÉSi/:Input: /asi/i/-sasi]ORRá[+sib]DENT[ant]*"i *sRemarks! a. s1as1i SCorr!:{s s} (0)(0)(0)(2)L harmb. "1a"1i SCorr!:{" "} (0~1!)(2~0)R harmc. "1as1i SCorr!:{" (0~1!) (2~1)corr, no harmd. "1as2i SCorr!:{"}{s} (0~1!) (2~1)no corr(41)But *i doesnÕt favor rileft harmony in inputs endinin /ÉSa/:Input: /Exp: [a]ORRá[+sib]DENT[ant]*"i *sRemarks! a. s1is1a SCorr!:{s s} (0)(0)(0)(2)R harm b. "1i"1a SCorr!:{" "} (0~1!)(2~0)L harmc. s1i"1a SCorr!:{s "} (0~1!) (2~1)corr, no harmd. s1i"2a SCorr!:{} (0~1!) (2~1)no corr(42)WhatÕs oinon here:Candidates (c) & (d) are disharmonic:they either donÕt have correspondence between sibilants (d), or have correspondents that disaree (c); theyÕre ruled out by theharmony constraintsCandidates (a) & (b) do harmony: the sibilants correspond and The difference between (a) and (b) is the result of areement: either [sÉs], or [].The tableauxs here, and most others later, use a hybrid comparative format. Winnincandidates are iven in row (a), and other rows represent comparisons between that winner and an alternative, losin, candidate. Inteers in parentheses show the constraint violations incurred by the winner, and how they compare to violations incurred by the losinalternative. Losers that match the data are marked with ÔÕ. ÔWÕ, ÔLÕ, and ÔeÕ indicate a constraintÕs preference for the Winner or the Loser, or neither (Prince 2002).To reduce visual clutter, e values are suppressed when both candidates have zero violations; blank cells indicate Ôe (0~0)Õ. UC Berke

ley Phonology Lab Annual Report: ABCConf
ley Phonology Lab Annual Report: ABCConference79(29)On the basis of the allophony pattern, we would expect forms in (28) to have disareeinsibilants but they donÕt.(30)Harmony in these cases is crucially resolved in a RiLeft way:The rihtmost sibilant follows the allophony its quality is predictable from the followinvowelPrecedinsibilants deviate from the allophony in order to aree with the rihtmost they arenÕt predictable from the followinvowel(31)Point: The rihtmost sibilant seems to be controllinthe areement, irrespective of its qualitywhich is determined based on the followin2.2. The problem, illustrated(32)The combination of allophonic conditioninand strict directionality makes the Kia pattern problematic for theories like ABC, where harmony is driven by an imperative to have areementKeyConstraints(33)ORRá[+sibilant]: ÔSibilants correspond with each otherÕ For each distinct pair of output consonants X and Y, assin a violation if:X and Y are both [+sibilant]X and Y are not in the same surface correspondence class(34)DENT[anterior]: ÔCorrespondents aree for anteriorityÕ For each distinct pair of output consonants X and Y, assin a violation if:X and Y are in the same surface correspondence class X is [+anterior]Y is [anterior](35): Ôno [i] or [i] sequencesÕOne violation for each [anterior] sibilant followed by [i] in the output.(36): Ôno [] or [One violation for each [+anterior] sibilant in the output.(37)he basic allophonic distribution requires *i to dominate(and both must dominate faithfulness for anterioritytoo) *"i *sDENT[ant]a. i/ si ~ W (0~1)L (L (1~0)b. a ~ sa W (0~1)L (1~0) This issue eneralizes to other aree

mentbased harmony theories besides ABC (
mentbased harmony theories besides ABC (Pulleyblank 2002, e..), but we will consideronly ABC here.These specific definitions of the allophony constraints are not essential to the problem. For example, instead of i *s, we could have *sa,se,so,su , and the puzzle arises in the same way.UC Berkeley Phonology Lab Annual Report: ABCConference78(25)Representative examples:s ~ z ~ ÔscoldÕÕ k-z“r-ˆ ÔforbidÕÔlauhÕ [e]Ôbecome soddenÕÔkickÕ [a]ÔlapÕÔreadÕ [o]Ôlare water potÕÔflatterÕ[u]et wetÕ(26)Within the stem (root + suffixes), sibilants aree in anteriorityIn some cases this harmony is trivial, since areement is expected from the allophony anyway pattern (2In others, sibilant harmony overrides the expected allophony (2(27)Simple distributional harmony examples:[ s, z ]Ôsin(s), evilÕphlee", # ] b. -!" ÔdysenteryÕ Ôtie looselyÕ Ô(African) lonhairÕÔto nurseÕ Ôpull outÕÔelephantsÕ Ôseem, lookÕÔtsetse fliesÕ(28)Sibilant harmony overrides allophonic distribution Expected by allophonyAttested L harm.) Unattested R harm.) a. *-si"a -!’!ˆ sisa Ôbe fatÕb. *-siia -!’“! siisaÔear of milletÕc. sii-!“!ˆsiiÔserve (used of cattle)Õd. *-si"wa -!’! swaÔshrink fromÕe. aasire-sˆˆ“r *-"aaireÔbe in pain (perf.)Õ f. *-"asi -sˆs“ *-"a"i ÔporchÕg. *-"asi -s‡“ *-"a"i ÔbulletÕh. *-"aa -s‡ˆ *-"aai Ôspace outside a kraalÕ Harmony doesnÕt extend across the stem boundary into prefixes: -a-!i#a"] Ôthey (cl.10) have arrivedÕ (Taylor :122;tones not iven). This

boundinparallels sibilant harmony in oth
boundinparallels sibilant harmony in other Bantu lanes, such as Kinyarwanda (Kimenyi 1979, Mpiranya & Walker 2005, Bennett 2013), and can be handled in the same wayUC Berkeley Phonology Lab Annual Report: ABCConference772. The Kia Puzzle2.1. Kia in a little more detail(18)Nkorea is actually a family ofclosely related dialects spoken primarily in anda, classified as E.1314, in the Interlacustrine roup. (19)We follow TaylorÕs (1985) convention of referrinto Runankore and Rukiether as Nkorea, and the eneralisations data reported in previowork enerally t attributed more specifically within this roup(20)The phonemic consonant inventory of Kia (after Taylor 1985)labialdentalalveolaralveolarvelarlottalp b t d t" d#k ! f v s z " # h m n $ % w & j (21)Sibilants donÕt occur before consonants; the only possible consonant clusters are NC{w,j} in form.[w] and [j] seem to pattern like [u] and [i], respectively.(22)revious work (Hyman 2003, Hansson 2001) reports that the anteriority of sibilants is conditioned by the followinowel, in a ÒbackwardsÓ fashion[+anterior] sibilants [s z] occur before [i] [i]-anterior] sibilants occur before all other vowels(23)Allophonic sibilant distribution(24)This distribution is unusual (cf.Japanese, Nupe, Kikono, ; /s//"]/_ihat isnÕtrelevant for the directionality issue in harmony, though.Taylor analyzes [t] and [d] as palatalization of /k / before hih vowels, and [d] as an allophone of /r/ after /n/. Taylor doesnÕt ive /ts/ as a sement in his inventory, nor does Poletto(1998), but itÕs attested in examples iven by both.Taylor (1985:205, 210) mentions that vowel reduction in normal speech can yield ÔquasieminatedÕ ["(], and

[St] clusters, but we donÕt consider t
[St] clusters, but we donÕt consider these here.UC Berkeley Phonology Lab Annual Report: ABCConference761.2. Context: Directionality in ABC(8)Previous approaches to directionality ofharmony in ABC:Build directionality into the Correspondence elation and/or the ORRconstraints (Hansson 2001/2010; Walker 2000b, 2001)Build directionality into the CCDENTconstraints that drive areement (Rose & Walker 2004)Appeal to other factors, like positional faithfulness (Bennett 2013; see also Hansson 2010)(9)None of these turn out to handle the Kia pattern correctly.The core issue is the direction of assimilationBuildindirectionality into the basic ABC mechanism canÕt control the direction of assimilation it only restricts when harmony is enforcedPositional faithfulness doesnÕt work if the sibilant controllinharmony isnÕt faithful (i.e. is allophonically determined)1.3. Goals and structure of this talkAimsthistalk(10)Explicate the problem Kia poses for an ABCtheory that doesnÕtdirectionalityas primitiveand parametrically determinable(11)Demonstrate that the problem actually isnÕt well supported by the factsDistribution of { s z } is nothtforwardly allophonicBut it heavily affected by morpholo(12)Proposa: the directionality isnÕt arbitrary; it follows at least in part from morpholoicallyintroduced asymmetries(13)Broader conclusion: this is an interestinproblem, but one we donÕt need to adjust the core of the theory to solveRoadmap:(14)¤2 explorese problem in more detail, and shows how it isnÕt explained by proposals(15)¤3 exploresthe data in more depth, and shows that sibilants arenÕt really allophonicThis puts a much different spin on the problem.(16)¤4 considers the effects of morpholohi

ch handle one crucial case(17)¤5 summar
ch handle one crucial case(17)¤5 summarizes and points out the broader consequence: the Kia case doesnÕt require an ABC theory to refer to directionalityspecificallyUC Berkeley Phonology Lab Annual Report: ABCConference75Directionality in Nkorea sibilant harmony: arbitrary or emerent?1. IntroductionOverviewand the question(1)The focus of this talk is sibilant harmony in Nkorea (henceforth just ÔKiaÕ), a Bantu lane spoken primarily in U(2)Previous work on Kia reports three eneralizations (Hyman 2003, Hansson 2001/2010).eneralizations:(3)Gnztn #1: Anteriority in sibilants is normally allophonic, conditioned by the followinvowel[+anterior] { s z } before [i] [i]Ðanterior] { } elsewhere(4)Gnztn #2armonywithin the stemmakes sibilants deviate from the normal allophonic pattern (5)Gnztn #3Harmony operates strictly left he rihtmost sibilant in the stem is conditioned in the normal way, and other, precedin, sibilants assimilate to match it.(6)Taken toether, these eneralizations yield a pattern of consonant harmony that defies explanation in Areement By Correspondence (Walker 2000a, 2000b; 2001, Hansson 2001/2010; Rose & Walker 2004).Sibilant allophony determines the result of areementBut, areement otherwise overrides allophonic conditionin(7)Question: if sibilants are allophonic in the first place, how can harmony always prefer a strictly Left directionality?Acknowledements: We want to thank Dr. Connie Tukwasibwe, Blake Allen, and Gunnar Hansson for assistance in obtain and processin Kia data, and oranisers and participants at the 3rd African Linuistics School in Ibadan, where the problem addressed in this talk came to our attention.UC Berkeley Phonology Lab Annual Report: ABCConference