24.961 Harmony-2

[1] Autosegmental models: a multiply-linked feature represents simultaneity in time; same assumption for domain theories like McCarthy’s Spans; Cole & Kisseberth’s Optimal Domains

• Neutral/transparent vowels are a serious challenge since it appears that a feature has spread across a segment that could potentially bear it: a “gapped” structure (Archangeli & Pulleyblank 1994) Finnish kodi-kas \ / [+back]

[-back] [2] One response

• The contradiction is only apparent (Gafos 1996); the harmonic feature is just not audibly salient on a vowel or non-contrastive obstruent in sibilant harmony like Chumash (1982)

k-sunon-us ‘I obey him’ s-ixut ‘it burns’ k-ʂunot-ʂ ‘I am obedient’ ʂ-ilakʂ ‘it is soft’

• Challenged by various phonetic studies as a general solution to transparent segments (Beddor & Yavuz 1995 for Turkish, Walker 1998 for Guarani, Przezdziecki 2006 for Yoruba; cf. Benus & Gafos 2007 for Hungarian) • Alternative Syntagmatic Correspondence relations proposed by Rose & Walker (2004)

[3] Consonant Harmony vs. Vowel Harmony

• consonant harmony has fewer blocking segments than vowel harmony

Kikongo: suffixes with /l/ become [n] when root contains a nasal consonant; any other consonant or vowel may intervene (including a prenasal)

m-bud-idi ‘I hit’ tu-kun-ini ‘we planted’ n-suk-idi ‘I washed’ tu-nik-ini ‘we grounded sos-ele ‘searched for’ leem-ene ‘shown’ sakid-ila ‘congratulate for’ ku-dumuk-ina ‘to jump for’ bantik-idi ‘begun’ tu-mant-ini ‘we climbed’

• similarity effect: segments that harmonize are ones that are most similar to the trigger: in Kikongo voiced stops but not voiceless ones assimilate nasality; compare nasal spreading in Malay that affects vowels and glides or in Ijo that spreads nasal leftwards

1 from a nasal consonant or vowel but is blocked by obstruents: izõŋgo ‘jug’, abãmu ‘loft’, j̃ãrĩ̃ ‘shake’

[4] Hansson (2001) and Rose & Walker (2004) propose a syntagmatic correspondence constraint based on shared features (similarity); long distance consonant harmony arises from imposing an Ident-[F] on corresponding segments in this syntagmatic relation, termed Agreement-by- Correspondence (ABC)

• Corr C<->C: Let S be an output string of segments and Ci, Cj be segments that share a

specified set of features F. If Ci, Cj are in S then Ci is in the Corr relation with Cj and thus

Ci and Cj are correspondents of one another • A hierarchy of correspondence based on similarity (for nasality):

Corr T<->T » Corr T <->D » Corr K<->T » Corr K<->D

Identical stops same place same voicing any pair of oral stops A TYPOLOGYOF CONSONANT AGREEMENTAS CORRESPONDENCE 509 • for coronal anterior harmony shared continuancy makes two sibilants more similar than tionalityby a including stop andthe a fricativeleft/right target edge in their statement(examples for harmon- ies consonantsinclude Cole & Kisseberth1995, 1995a, Gafos 1996, involving• harmony arises from an Ident-[F] over the correspondingPadgett segments that dominates input- Nif Chiosaiin& Padgett 1997, cf. Walker 1998). Likewise, in rule-based approaches, the directionoutput is incorporated faithfulnessinto the rule's description(e.g. Poser 1982, Shaw 1991, Odden 1994). In the ABC approach,IDENT-CC constraints compel agreementbetween consonants, and it is here that we locate the directionalitystatement.33 In Kikongo, the rightward direction of nasal agreement indicates that IDENT- CLCR(nas) is prioritizedabove IDENT-CRCL(nas).Since IDENT-CLCRis always obeyed by the consonantsthat participatein Kikongo nasal agreement,we locate this constraint at the top of the hierarchy.The dominatedstatus of IDENT-CRCLwill become apparent when we examine forms in which an oral voiced stop or approximantconsonant pre- cedes a nasal in a stem. First,Rose, we Sharondetermine, and theRachel ranking Walker.of “A IDENT-OI(nas) Typology of Consonantwith respect Agreement to IDENT-CLCR(nas) as Correspondence.” Language 80 (2004): 475-531. © Linguistic Society of America. All rights reserved. This content is excluded from our Creative Commons and the nasal correspondencehierarchy. In Kikongo, any voiced stop becomes nasal license. For more information, see http://ocw.mit.edu/help/faq-fair-use/. if precededby a nasal in the stem. This signals that CORR-N+-Band IDENT-CLCR(nas) • in a language with anterior harmony between affricated stops and fricatives Ident-IO together outrankIDENT-OI(nas), as shown in 52. The winning candidatein 52a estab- lishes correspondencewould slipbetween below Corr the nasal t<-> andʃ suffixso underlying consonant, /talakand they-ʃ/ would agree for become nasal- /talak-s/ ity.34 In• 52b,analysis the stops of Kikongodo not correspond, (prenasaliincurring zed stopsa arefatal not violation in the ofcorrespondence CORR-N4-B, relation nor are and in 52c the stops correspondbut do not agree in nasality, an outcome ruled out by IDENT-CLCR(nas).prefixal segments) (52) IDENT-CLCR(nas), >> IDENT-OI(nas) CORR-N--B /sim-idi/ ID-CLCR(nas) CORR-N"-D CORR-N<-B ID-OI(nas)

a. a7 simxinxi b. simxidyi I

c. simxidxi ,

Rose, Sharon, and Rachel Walker. “A Typology of Consonant Agreement as Correspondence.” Language 80 (2004): agreement.475-531.This © Linguistic is captured Societyby situating of America.IDENT-OI(nas) All rights reserved.below This CORR-N+-L content is excludedas well. from We our Creative Commons locatelicense. For moreandinformation, CORR-N4-B see http://ocw.mit.edu/help/faq-fair-use/together in the nasal correspondence. hierarchy, as CORR-N+--L their similarity rankings are very close. The outcome is illustratedin 53. Candidates 53bfaithful and 53c, candidate which do b violatesnot nasalize the /1/constraint in the output, defining are eliminatedthe syntagmatic on the corresponbasis of dence and faithful CORR-N4-4Lc obeys correspondenceand IDENT-CC(nas), but violatesrespectively.35 the constraint requiring identity for [nasal]; IO faithfulness

3"for Compare the root another or Max kind of-[nasal] approach > in Depwhich-[nasal] apparent directionality blocks denasalizationeffects seen in certaincandidate kinds ofsib x-idxi assimilation are accomplished via positional faithfulness constraints without statement of direction (e.g. 2 Padgett 1995c, Beckman 1997, 1998, Lombardi 1999, Walker 2001b). 34 Whether we posit /1/ or /d/ as the input suffix consonant does not figure here. Either way, it will be realized as [d] before [i] if oral, which we attributeto a contextual markednessconstraint that we refer to descriptively as *[li]. 35 We regardit as unsurprisingthat under nasal agreementthe affected consonant becomes a plain nasal stop at the cost of mannerfeatures that might be active in the segment. When, for instance, [1] becomes a nasal segment it does not retainits approximantnature (attributable to the feature[lateral]). In addition,there are no reportsof nasal continuantsin languages where continuantapproximants are affected by nasal agree- ment in MSCs. The explanationhere is twofold. First, the dispreferencefor nasalized continuants/approxi- mants is well documented(Padgett 1995b). Constraintson such configurationswill triggerthe hardeningof approximantsto stops undernasalization. Second, the formationof nasal stops ratherthan nasalizedsegments better satisfies IDENT-CCrequirements. These constraintswill promote the closest match in strictureand other propertiesbetween agreeing segments.

This content downloaded by the authorized user from 192.168.82.207 on Tue, 13 Nov 2012 19:35:50 PM All use subject to JSTOR Terms and Conditions LANGUAGE,VOLUME 80, NUMBER 3 (2004)

(53) > IDENT-OI(nas) LANGUAGE,CORR-N+--L VOLUME 80, NUMBER 3 (2004) /nik-ulu/ ID-CLCR(nas) CORR-N*-4D CORR-N"-B ID-OI(nas) CORR-N--L agreement between nasal and voiced oral stop pairs is similarly a. agrn,ikun,u * ranking:IDENT-OI(nas) >> CORR-N-~D. to notice about the constraint for is that IDENT- b. nikulyu ranking Ngbaka between the CORR-C+-Cconstraints requiring correspondence be- nasal/prenasalc. nikulu stops *!and the ones that enforce correspondencebe- heterorganicRose, Sharon nasal/prenasal, and Rachel stops Walker. and “A Typology nasal/voiced of Consonant oral stopAgreement pairs. as IDENT- Correspondence.” Language 80 (2004): which475-531. Observepromotes © Linguisticthat nasal the identity voicelessSociety ofbetween America.stop in corresponding 53All rightsdoes reserved.not participatesegments, This contentin stands nasal is excluded un- agreement. from ourThis Creativ e Commons license. For more information, see http://ocw.mit.edu/help/faq-fair-use/. follows from its lack of similarity to the nasal: the nasal and oiceless stop are not sufficiently similarto provoke correspondencein Kikongo, and hence agreementis not IO faithfulnessAND HETERORGANIC for nasalityNASAL slips AGREEMENT. below theWe constra turn ourint attention placingnow nasals to and liquids in a enforced between them. Candidatesin which these consonants are in correspondence where nasal holds between nasals and voiced oral both correspondenceare screenedagreement out relationby faithfulness leading constraints: to nasalization of liquidsincursstops, by a thegratuitous remainingviolation constraints but a and A TYPOLOGYOF CONSONANTconsonants AGREEMENT[nxirlxunxu] alsoAS are CORRESPONDENCE 511 voiof celessIDENT-OI(nas),heterorganic stop is pairs.tooand dissimilar likewise Approximant with nasalwith and respect so is not totargeted. IDENT-CC(nas). placed First,in the correspondenceCorrespon- relation are in order. As the [nxikxunxu] does not include dence between consonantsinKikongo the output inventorythus occurs only when compelledprenasal by similarity- (54) constraintsIDENT-CLCR(nas) IDENT-OI(nas)are not ofIDENT-CRCL(nas) relevance so we driven constraints,and involving the >>neutrality prenasalsof voiceless >> consonantshere, (and vowels and voiced constraintsfrom tableaux. We also and CORR- fricatives) followsKikongo straightforwardly. omit CORR-N+--N these /ku-dumuk-ila/constraints ID-CLCRredundant i CORR-nasal CORR- and CORR- do notID-OI ID-CRCL The tableau inproduce 54 addressesonly (nas) directionality.agreementNasal agreementN<->L in this(nas)play word produces (nas) in the alternationswe examine. i N-*D nasalization in the 1/ to the right of the nasal butN--B leaves the oral quality of the /d/ to nasal shows as in /ku- its leftagreementa. agintact. The resultingrightward output idirectionality, sequence obeys exemplifiedIDENT-CLCR(nas), which requires** -- kudxumxukinxa We that directional that[kudumukina] corresponding*[kunumukina]. consonants following propose a nasal be specifiedagreement [nasal], but it violates an evaluationb. kunumukinaof faithfulness sensitive to the dimension * IDENT-CRCL(nas),which requires a [nasal]left/right specification in corresponding(i.e. prece- consonants constraintsthat versus precedinga nasal. It distinguishis the interleaving progressiveof IDENT-OI(nas) regressivebetween agreement these constraints that 51. c. kunxumxukilya thatif a is *!* * achievesIDENT-CLCR(nas) unidirectionalrequires agreement. The segment ranking [nasal],of IDENT-CLCR(nas)any correspondentover IDENT-OI(nas) to its RIGHTin the sequence of segments must also be [nasal]. IDENT- compelsd. rightwardkudumukinanasal agreementin the! /1/, as evident** in comparison* of 54a and 54e. for in the leftwarddirection. Unidirectional responsibleThe one-way directionalityagreement is seen in 54a versus 54b,c. In 54a, as opposedagree- to 54b, both under asymmetrical of* these constraints.Prioritization of IDENT- * candidatese. kudumxukilaestablish rankings correspondence between all threevoiced consonantsand both obey feature as in and dominance of IDENT-CLCR(nas).progressive But theagreement, leftward agreement Kikongo, affecting the first voiced consonant in resultsf. in regressive agreement, as in Kinyarwanda*(!) sibilant *) agreement 54b incursud,umxukila a fatal violation of IDENT-OI(nas).Even though this candidatefares better positRose, that Sharonth left/right, and Rachelsensitive Walker. constraints “A Typologytogether of Consonantreplace Agreementnondirectional as Correspondence.” Language 80 (2004): with475-531. respect © Linguistic to IDENT-CRCL(nas), Society of America.this All constraint rights reserved.is dominated This contentby isIDENT-OI(nas), excluded from ourand Creativ e Commons NCsin CON in Kikongo,(the set ofAo universal (1991:195, constraints). n. 2) pointsIn out Ngbaka, that nasals where devoice there before is no a voiceless hencelicense. a For faithful more information, mapping of see the http://ocw.mit.edu/help/faq-fair-use/ leftward/d/ is favored. Candidates. 54a and 54c incur unidirectionalCequal word-initially violationsnasal (e.g. with agreement, nkosi respect 'lion'), to bothIDENT-OI(nas), but constraints not word-mediallybutare 54c situated violatesin normalat tCORR-N+-L same rate speechwhile (e.g. 54a zinkosiSince there 'lions').and are likewise He no attributes headed for the-thisdo relevantmains difference in ABC to the, directionalityIDENT-CC(F) differentsyllabic constraints must roles be ofbuilt the intonasals. the constraints hierarchy,obeys it. Nasalization of the first voicedlaryngeal consonant is thus again suboptimal.CORR- andWord-initial Bolivian NCs are analyzed as tautosyllabic sequences. Note that with respect to changingconstraints Aymara.the segmentsalso rule in outthe candidatescorrespondence54d and relation: 54f. note that a > b in (54). neutrality,C+--C our attentionis limited to medial NCs only, because in initial NCs the nasal a. IDENT-CLCR(nas):The final thatLet weCL addressbe a segment in our in the outputof and CRnasal be any derives from apoint prefix, which stands outsideanalysis the stem domainKikongo in which nasal correspondentof NCof clusters.CLsuch thatWe attribute CRfollows the CL in the sequenceof nasals of in segments NC operates.neutrality neutrality in the outputfrom (R>L). the If CL is [nasal],affected then oral CR consonants. is [nasal]. We considertwo dissimilarityThe preference for relationspotentiallyto exist between segmentswith matchingroles IDENT-CRCL(nas):for theLet behaviorCL be a ofsegment in nasals the output from and nasals CR in be NCs: any structuredistinguishingis accomplished by the constraintsingletonin 55 (after Gafos 1996, correspondentreferenceRose, Sharonto , theirandof RachelCL differentsuch Walker. that CR“A Typologyfollows CLof Consonantin and the the sequence Agreementother involves of segments as Correspondence.” reference to their Language 80 (2004): McCarthy& Prince 1993, 1994,syllable Suzuki positions, 1999). indifferent475-531. the output ©release Linguistic (R>L). status. Society IfWe CR of develop is America. [nasal], the All then former rights CL reserved.approach is [nasal]. Thishere, content but alsois excluded sketch from the latterour Creativ e Commons license.(55) For SROLE-CC: more information,Corresponding see http://ocw.mit.edu/help/faq-fair-use/consonants must have identical. syllable roles. of unidirectional the choice callsbetween on a them distinction to further researc avaiable in possibility,The scenario leaving is laidagreement out in 56. SROLE-CCdominates already CORR-C constraintsso as thisof TurningCORRESPONDENCE constraintto the forces syllableTHEORY: change position for of a approach,segmentfaithfulness inwe the speculateconstraints Corr relation thatmay --C nasals target that in followsmedial NCsa nasal but says to inhibit correspondencebetween o•t&, similar consonants that have differentsyllable roles. Inbelong the case in a syllableof input-output coda. Incorrespondence, contrast, examples nasals of constraintsto an onset,that as do Thisnothing ranking aboutis responsible what precedes;for eliminating IO faithfulnesssingletoncandidates thenbelong56b prevent and 56c,s change where onthe segmentssingletonnasal in preceding the oralthe stops. MAX-IO If syllablefamily position (McCarthy is a factor & Prince contributing 1995) andto segments' the IDENT-IO relative. similarity, thenasal; NC a cluster corresponding is in correspondence constraint withIdent a-C singletonRCL controls onset right consonant.-to-leftCandidate harmony 56c Paterthen the 1999, coda while nasals' those difference onin thisare fromconstraints the onsetand IDENT-could rise to also fails to show nasal agreementfocusing between 3 regard DEP-IOcorresponding consonants, stops violating give IDENT- their constraintsIn in the 51 area of the studiesof faithfulness, have found that CLCR(nas).generalneutrality. remark:Candidate it is 56a unclearsimilarly is optimal:language how distinguish we it obeys performance,restrict SROLE thetarget ABCat the technology cost of to justsegments the(It dependentis features of theand same in terms of the are moredimension. to in errors assumedoccupying hereoa that3 the syllable input contains positionprecedence a high vowel likely in theparticipate suffix, CORR-N+--L.which speech is lowered of coronal consonantsinto1983, andthe nasal;This also lendsof do the we constraint haveto featurethethat notion drives spreading that or feature spans for other through(Shattuck-Hufnageldirectionalitythe operation of vowel1987).expression harmony.) support agree-matching syllable harmoniesrolesa tacticcontribute thatsimilar torespect segments' to that locality? taken similarity. in otherHowIn workdo support we dealing tellof whichthe with heterosyllabic directionality representationstatus isof appropriatemedial in any given In work on feature>> the constraintconstraints has been case?(56) SROLE-CCspreading, CORR-C->C driving harmony terms of alignment (Kirchner1993). Such constraintsaccomplish direc- /-somp-ila/ IDENT- SROLE- CORR- CORR- CORR- IDENT-OI [5] Nevins (2004, 2009)CLCR CC (nas) (nas) N<--D N-->B N--4L This content downloaded by the authorized user from 192.168.82.207 on Tue, 13 Nov 2012 19:35:50 PM • All use subject to JSTOR Terms and Conditions a. QrVowel u harmony is treated by the same principles that define syntactic agreement: /1 A A 3 This contentsomxp downloaded elya from 18.7.29.240 on Wed, 5 Nov 2014 09:00:11 AM All use subject to JSTOR Terms and Conditions b. a ur a /NAAA somxp enxa

c. 0om F i ) / Ai i A somxp elxa

This content downloaded by the authorized user from 192.168.82.207 on Tue, 13 Nov 2012 19:35:50 PM All use subject to JSTOR Terms and Conditions in Italian “I ragazzi sono andati” the auxiliary needs to have its person and number features specified while the participle needs to specify its number and gender features— both looking to the subject—the closest relevant element (relativized minimality)

• There are parametric choices imposed on the search algorithm; cast in a derivational rule-based model (a la Calabrese 2005) where ordered rules can be parameterized for whether they see all features, just marked features, or just contrastive features

Turkish

pul pul-lar pul-un pul-lar-in ‘stamp’ el el-ler el-in el-ler-in ‘hand’

[−high] [l] – [l][−high] [r] ‘hands’ el-ler < /el-In/ −round −back −back filled by copy from closest specified feature matrix

- search parameter: left-to-right/right-to-left - for pul-lar-in ‘your stamps’ the search in order to value [round] looks to the preceding plural morpheme –lar- rather than to the root pul- since the former is closer is closer

• an opaque vowel is one that is lexically specified for the relevant feature and hence does not have to seek a value but will instead donate its value to another segment that needs one; earlier redundancy rule assigns predictable [−ATR] to low vowel

/ɗob-Uɖ-nA-m-gO/ -> ɗob-Uɖ-na-m-gO -> ɗob-uɖ-na-m-gɔ [ATR] + + − + + − −

[6] Neutral vowels are treated as reflexes of parameters on the search procedure: the procedure may look for contrastive values or for marked values (cf. Calabrese 2005) For Finnish [back] minimally contrasts a vs. ä, o vs. ö and u vs. y; i and e lack back counterparts

i e ä y ö o a u high + − − + − − − + low − − + − − − + - back − − − − − + + + round − − − + + + − + • Search to value a suffix for the closest contrastive value for [back]

kodi-kas ‘cozy’ vs. väri-käs ‘colorful’

4 • Even though [i] in kodi is specified [−back], it is not minimally contrastive in the phonemic inventory for [back] since there is no /ɯ/and so the search continues until it finds such a vowel, in this case [o] or [ä] • If the search is unsuccessful then a default value is inserted: [−back] • For Finnish it seems like the default of [−back] matches the neutral vowels but Nevins cites Uyghur as language with the same phonemic inventory as Finnish but stems in [i] and [e] take back vowel suffixes; thus on this view the [−back] on the suffixes in kään- tää ‘to turn’ and piir-tää ‘to draw’ arise by different mechanisms: the former by copying the root vowel’s [−back] and the latter by insertion of default value

[7] Reference to marked values: Sibe (Xibe) is a Tungusic language of China

Sibe [−low] i ü i u

[+low] ɛ ö a ɔ

[+low] is marked: evidence

• Xibe disfavors successive low vowels in root • Most suffixes are high • Epenthetic vowel is high • Diachronic shift of final to initial stress raises unstressed low vowels

[8] Dorsal consonants contrast velar [k] and uvular [q] as [−low] vs. [+low]

• harmony: uvular occurs when preceding stem has a low vowel with any nonlow vowels allowed to intervene

tɔndɔ-qun honest udzin-kin heavy sula-qin loose ildin-kin bright adzi(g)-qin small farχun-qun dark

• Thus the search is for a marked value and any unmarked (high vowel) may intervene • default is [−low] velar in case search does not find a marked segment as in udzin-kin

[9] In Sanjiazi Manchu the search is relativized to contrastive features

[-low] i ü i u

[+low] æ a ɔ

• Harmony for [low] in suffix

5 qa-χa obstruct-past sæ-χa bite davi-xi stride matʃu-xu grow thinner tæri-xi plant

• search for [+low] cannot cross a [−low] vowel • no examples of stems of structure CaCü • the analysis predicts that search will look past [ü] to [a]

/ sæ-XI/ /tæri-XI/ low + 0 + − 0 UR + + + − − search (contrastive)

• this approach depends on establishing (learning) the contrastive and marked status of segments and that phonological processes can freely refer to either as well as fully specified ones

[10] In Hungarian (and Finnish) there are gradient effects in B N- V words where B = a back vowel, N a neutral (transparent i, e) and V a suffixal vowel (Hayes & Londe Phonology 2006, Ringen & Heinamaki 1999, and Kimper 2011)

pallér-nak [pɔlle:r-nɔk] ‘foreman dat. arzén-nak ≈ arzén-nek ‘arsenic’ mutagén-nek ‘mutagen’

• google searches and wug (novel) word testing probe this variation • distance effect: B N N – V has less backness harmony than B N – V • height effect: the neutral vowel transparency effect allowing harmony by a preceding back vowel increases with height and length: ɛ< e: < i • competition between N and B in determining the suffixal vowel; seems to track F2 with [i] having higher F2 (more front vowel) and [ɛ] lowest; • Hayes & Londe model with a stochastic (probabilistic) grammar where ranking between neighboring close constraints can change in a given input – output mapping based on a probability function tracking frequency in the speaker’s lexicon/experience

Selected References Gafos, Diamandis. 1996. The articulatory basis of locality in phonology. Baltimore, MD: Johns Hopkins University dissertation. [Published, New York: Garland, 1999.] Hansson, Gunnar. 2001a.Theoretical and typological issues in consonant harmony. Berkeley: University of California, Ph.D. dissertation. Hayes, Bruce and Zsuzsa Londe. 2006. Stochastic phonological knowledge: the case of Hungarian vowel harmony. Phonology 23, 59-104. Nevins, Andrew. 2010. Locality in Vowel Harmony. Cambridge, MA: MIT Press. Rose, Sharon and Rachel Walker. 2004. A typology of consonant agreement as correspondence. Language 80, 475-531.

6 MIT OpenCourseWare http://ocw.mit.edu

24.961 Introduction to Phonology Fall 2014

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.