24.961 Contrast and Perceptual Distinctiveness [1] Flemming (1995, 2004, 2006) • Auditory based features and constraints • Grounded in perception • Explicit reference to paradigmatic contrast • Constraints evaluate the distance between contrasting pairs of sounds in perceptual space optimizing for distinctiveness of contrasts [2] Markedness may depend on contrast (1995) • For nonlow vowels, lip rounding aligns with backness (an enhancement relation): [i-u] is more distinct on F2 than [i-y] or [i-ɯ]; • Theories of markedness that don’t invoke contrast posit *y,ɯ » *i,u (cf. Calabrese’s 1995 Filters) • Correctly states that a language will choose /i/ before /y/ • But Flemming claims that central vowels like [i] are only marked when compared to [i] and [u] • In a system that lacks [i] and [u] then [i] may be the optimal vowel on articulatory grounds, since it involves more minimal tongue displacement between consonants • Marshallese is parade example (Choi 1992) • Vertical vowel system with front and back and round determined by consonants, which are palatalized and velarized • Historical reanalysis of Autronesian five-vowel system: tjeʌpɯ < *tepo • Calabrese might argue that Marshallese vowels are underspecified for front vs. back rather than being central • What happens at word edges or when the vowel is long: do we see a central vowel quality steady state? Choi (1992) states that long vowels have an F2 target but does not say if it is a central vowel [3] dispersion theory • An inventory of sounds is a compromise between constraints maximizing the distance between the sounds along some auditory dimension (e.g. F1, F2; voicing duration in consonants) and constraints maximizing the number of sounds, with articulatory effort being a third factor • Given a fixed auditory space, the more sounds there are, the smaller the distance between them: cf. persons in an elevator 1 ! Sound are located in a multi-dimensional perceptual space where perceptual Forsavow asssum somceoul grd idbe with idieeadlizteodm spoarceet;ha F2nin barckesl;l.F1 doubles bark frequencies e•s, the me eIlPs A ymebol appl one a ranked set of een contrasting sounds. ints are opposed by two general (families of) constraints: Minare aD xiism ze t1he beVowels rt-of astspecified . ist-F2:3by ! Perceptual dimensions n-ary features. feature matrices: [F1 • ! *M t-iF2: » *num MinDis F2:c2ont » are *rMi nsD » .... » *MinDist-F2:6: this 1, markedness ! M ini m i z e a r ti c ul a t or y e f fort. F2 6, …]. hierarchy optimizes the distance between sounds along some auditory dimension Ma numb erbe ofrcof ontcront astsra(v • Ma xixmimi izezethe num stosw. els): more information can be stored and transmitted per uni t ofesspa ce tor timnc e, ybut t tIhe aterapos sibl !O ppos the ende of aM NDcIos STt of to gfarevor few meacxionf mauslliyondistinct contrasts. • Sound inventories arise from embedding Maximize Contrasts somewhere within the MinDist ! With more contrasting sounds, each sound spoken can distinguish more words (and thus nkinys g more information). onve cra • ! IEr taebl w: de reas e eac Min distts= onest mrplateaminent deabe s alopos iticve cons trahint tha elFe2ctval s tuhee blayrge 3 Japran r nt Kor • not Firsbe t tabl elghe e, sdeccon d tfroai en eealiumfor inaftievde-vow by hi -raesnke ons s)ean . inventory of contrasts (that has Flemming, Edward. “Contrast and Perceptual Distinctiveness.” In The Phonetic Bases of Markedness. Edited by Bruce Hayes, ! Kirchner, INDIST Robert and Donca Steriade. Cambridge University Press, 2004. © Cambridge University Press. This content is excluded from our Creative Commons license. For more information, see http://ocw.mit.edu/help/faq-fair-use/. M constraints impose the preference for contrasts between front unrounded and back rounded vowels. ! If there is no F2 contrast then these constraints are irrelevant (cf. vertical vowel 2 inventories). 16. Differences on multiple dimensions can combine to contribute to the overall distinctiveness • Candidate d is winner in second tableau; by moving Max Contrasts up (leftward) in the MinDist hierarchy, we derive more vowels but at the cost of closer spacing; by moving down (rightward) we derive fewer vowels but with larger spacing • Korean has /i/ vs. /ɨ/ vs. /u/ while Japanese has /i/ vs. /u/ (though /u/ is phonetically quite front, at least in Tokyo dialect; if it really is /ɯ/ then we would have to appeal to articulatory effort to choose /ɯ/ over the auditorily more optimal /u/. One the other hand, phonologically Jap /u/ patterns with labials, causing /h/ to be realized as [ɸ]). Perhaps F3 is relevant; see discussion of Cantonese below. Vowel height (F1) • Standard Italian (i,u, e,o,ɛ,ɔ,a) arises from ranking MinDist F1:2 » Max Contr » MinDistF1:3 Flemming, Edward. “Contrast and Perceptual Distinctiveness.” In The Phonetic Bases of Markedness. Edited by Bruce Hayes, Robert Kirchner, and Donca Steriade. Cambridge University Press, 2004. © Cambridge University Press. This content is excluded from our Creative Commons license. For more information, see http://ocw.mit.edu/help/faq-fair-use/. • Spanish demotes Max Contrast for greater dispersion: MinDist F1:3 » Max Contr » MinDistF1:4 Flemming, Edward. “Contrast and Perceptual Distinctiveness.” In The Phonetic Bases of Markedness. Edited by Bruce Hayes, Robert Kirchner, and Donca Steriade. Cambridge University Press, 2004. © Cambridge University Press. This content is excluded from our Creative Commons license. For more information, see http://ocw.mit.edu/help/faq-fair-use/. • Arabic {i,u,a} Flemming, Edward. “Contrast and Perceptual Distinctiveness.” In The Phonetic Bases of Markedness. Edited by Bruce Hayes, Robert Kirchner, and Donca Steriade. Cambridge University Press, 2004. © Cambridge University Press. This content is excluded from our Creative Commons license. For more information, see http://ocw.mit.edu/help/faq-fair-use/. [4] neutralization of contrasts • In unstressed syllables Standard Italian seven vowels reduce to five, with loss of the distinction between open and closed mid vowels 3 • EF sees this as a response to increased artic effort that would be required to realize vowels in shorter time span of unstressed syllables • Chief evidence is that low vowel /a/ is raised to [ɐ]; this encroaches on the vowel space; if the same Min-Dist and Artic effort constraints that define the stressed vowel/lexical inventory are imposed, then the number of distinctions decreases since the grammar now chooses the five-vowel system Flemming, Edward. “Contrast and Perceptual Distinctiveness.” In The Phonetic Bases of Markedness. Edited by Bruce Hayes, Robert Kirchner, and Donca Steriade. Cambridge University Press, 2004. © Cambridge University Press. This content is excluded from our Creative Commons license. For more information, see http://ocw.mit.edu/help/faq-fair-use/. Flemming, Edward. “The Role of Distinctiveness Constraints in Phonology.” Massachusetts Institute of Technology, 2006. © Edward Flemming. All rights reserved. This content is excluded from our Creative Commons license. For more information, see http://ocw.mit.edu/help/faq-fair-use/. [5] Dispersion and enhancement (Flemming 2006) • If dispersion constraints can freely interact with faithfulness and markedness constraints then the model over-generates: the inventory of contrasting segments should vary with context and targeted enhancement repairs should be able to rescue contrasts that are challenged by context (e.g. adding a schwa to final voiced obstruents). Another problem is an infinite regress: “the wellformedness of a candidate word [pad] might depend on whether or not [pat] is a possible word. But to determine whether [pat] is a possible word, we have to determine whether or not is statisfies MinDist constraints, requiring it to be adequately distinct from its neighbors, and so on”. • EF’s claim is that we don’t in general find these effects and the only response to a nonoptimal contrast is neutralization: in unstressed syllables distinctions are lost and the set of vowels shrinks rather than shifts (e.g. by introducing length); the only response to final voiced obstruents is devoicing (loss of contrast towards articulatorily less effortful sound) • Proposal is to restrict the role of dispersion constraints to defining the phonemic inventory that encodes the lexicon and as a final “quality check” on the output of the input-output mapping in the ESC (Evaluation of Surface Contrast) module; in particular, dispersion constraints cannot interact with (be ranked with) the markedness and faithfulness constraints that define the input-output mapping. 4 • In the ECS there are (apparently) just the MindDist constraints and a general *Merger constraint [6] Example from Cantonese: high vowels: i y u UG space: F2 F3 5 4 3 2 1 i y ɨ ʉ u 4 3 2 1 i ɨ y,u ɹ constraint ranking for inventory MINDIST= F2:2 or F3:2 (14) a. i ! b. ! i y c. iy ! d. iy " e. i u u u u u *! *! MAXIMIZE CONTRASTS !!! !!! !!!! !!!! !!! MINDIST= F2:3 or F3:3 **! * *** *** MINDIST= F2:4 ** ** ***** ***** Flemming, Edward. “The Role of Distinctiveness Constraints in Phonology.” Massachusetts Institute of Technology, 2006. © Edward Flemming. All rights reserved. This content is excluded from our Creative Commons license. For more information, see http://ocw.mit.edu/help/faq-fair-use/. Given an inventory of contrasting segments, the goal is to realize the full inventory of contrasts in all contexts. So possible underlying forms consist of all sequences of segments from • front rounded vowels found after dentals and velars but not labials: ti, tu, ty; ki, ku, ky but the inventory and the goal is to realize these underlying forms faithfully. In some cases pi, pu, contrasts *py underlying are neutralized because they cannot be realized with sufficient • could be repaired by shift to a new phoneme -> pɨ) but this is in general not found; distinctiveness in a particular context, as in final(/py/ neutralization of obstruent voicing contrasts. Following Flemming only merger to [i] (2002), we will analyze the restriction on front rounded vowels in β Cantonese in similar terms. Coarticulation with anofadjacent labial renders [i] too[isimilar to front ] that is • in input-output mapping there is coarticulation /i/ with /p/ creating a vowel rounded [y], so the contrast is neutralized in this context. In both cases the evaluation of too close to /y/ and the response is to neutralize the [iB] - [y] contrast; the outcome is distinctiveness must apply to the surface realizations of the contrasts in order to take contextual determined by lowering ranking dispersion constraint maximizing distance from [u] and effects such as labial coarticulation into account. However, MINDIST and MAXIMIZE CONTRASTS [i] freely with contextual markedness constraints if we are to account for the must choosing not interact (17) Realization: relative stability/pin/of inventories*Lacross contexts. these two generalizations motivates ABIAL IDENT(F2) IReconciling DENT(F3) COARTICULATION the distinctiona. between Realization and Evaluation of Surface Contrasts (ESC). The Realization *! pin * b. component maps an input string onto its phonetic realization while the ESC assesses the ! pi n **! *** c. p" n distinctiveness of contrasts based on these phonetic realizations. The Realization component incorporates(18)theESC: contextual markedness constraints that motivate contextual variation in the MINDIST= *MERGE MINDIST= MINDIST= /pyn , pin , pun / F2:2 not include MF2:3 F2:4 realization of contrasts but does INDIST constraints. The MINDIST constraints or F3:2 or F3:3 evaluate the outputs in ESC but cannot directly influence the realization of a given pin ,Realization pun / /pyn ,of a. *! * ** pyn pi n pun input. ! /pin , pun / b. * pun model to neutralization of front rounding contrasts is illustrated by piof n this The application /pyn , pun / c. (15) and (16). It is easier * to se ethe overall *! structure of the analysis by the tableaux in pyn pun considering ESC first andofthen turn to the detailsin of Realization. Flemming, Edward. “The Role Distinctiveness Constraints Phonology.” Massachusetts Institute of Technology, 2006. © Edward Flemming. All rights reserved. This content is excluded from Commons license. The ESC the same ranking of of MINDIST constraints asittheis Inventory – there our is no re-consider To evaluate theemploys distinctiveness contrasts necessary toCreative a target input in For more information, see http://ocw.mit.edu/help/faq-fair-use/. ranking of constraints between components, although only certain classes of constraints apply in relation to aeach setcomponent. of minimally contrasting inputs. This setplace must also occupies the same in the at least contain all input forms that The initial hypothesis is that *MERGE ranking as MAXIMIZE CONTRASTS. This implies that the minimum level of distinctiveness that is differ from the target by changing, inserting deleting a single 5 acceptable for a contrast in the inventory should also be theor threshold below which a surface segment, but generally must contrast is neutralized. In thewill analysisbe above, the smallest acceptable F3 contrast must have a contain additional forms as discussed further below. Here we are interested in the input distinctiveness of F3:2 since MINDIST = F3:2 ranks just above MAXIMIZE CONTRASTS (14). If *M ERGE occupies the samevowel position as adjacent MAXIMIZE CONTRASTS the rankingThe of MINDIST /pyn/ with aconstraints, front rounded to a inlabial. contrast set for this input includes any contextual influence that would reduce the distinctiveness of an F3 contrast ! ! 1 2 3 1 ! 2 1 3 2 ! 1,2 1,2 1,2 1,2 3 3 3 3 3 Gallagher (2010) • two types of cross-linguistic root co-occurrence constraints on laryngeally marked consonants such as ejectives • dissimilatory: C’VCV, CVC’V, CVCV, *C’VC’V (cf. Lyman’s Law in Japanese) • assimilatory (less common): *C’VCV, CVC’V, CVCV, *C’VC’V (Chaha) • Repair of dissimilation eliminates second ejective; repair of assimilation eliminates C’VC by distributing ejection through the root 436 Gillian Gallagher pattern I call ‘mixed ’,*T’-K’ heterorganic and homorganic pairs pattern differT-K (1) a. dissimilation T’-K ently. ‘ T-K’ stands for any pair of heterorganic consonants, while ‘T-T’ T-T *T’-T’ T’-T stands for any pair of homorganic consonants; the apostrophe here stands T-K b. assimilation T’-K’ *T’-K or implosive. T-T T’-T’ *T’-T T-K c. mixed *T’-K’ T’-K Gillian Gallagher a. Shus wap: kw’alt ‘to stagger’ qet’ ‘to hoist’ kwup ‘to push’ qmut ‘hat The constraints b. Chaha: ji-t’ək’ir ‘he hides’ ji-kəft ‘he opens *C’VC or CVC’ co-occurrenceT’-T’ restrictions reflect asymmetries in the perceptual T-T *T’-T Gallagher, Gillian. “Perceptual Distinctness and Long-Distance Laryngeal Restrictions.” Phonology 27, no. 3 (2010), strength of contrasts roots This with different laryngeal 435-80.© Cambridgebetween University Press. content is excluded from our configuraCreative Commons license. For more Co-occurrence restrictions on reported laryngealin features have various forms, The experimental results w4 reveal a three-tiered hierhttp://ocw.mit.edu/help/faq-fair-use/ . information, see Gillian Gallagher whichinat glance distinctness seem to be of contradictory. Assimilatory and disperceptual laryngeal contrasts in pairs of roots. thefirst • Not restrictions a simple markedness hierarc since what is ruled one tof ype is preferred in the similatory are opposites – hy what is prohibited inout oneintype The constraints [k’ap’i-kapi] > is required [k’api-kapi] > Mixed [k’ap’i-k’api] language isrprecisely what in another. restriction lanothe co-occurrence restrictions reflect asymmetries in the perceptual guages are particularly puzzling, because the antagonistic requirements to B([T’-K’]:[T-K]) B([T’-K]:[T-K]) B([T’-K’]:[T’-K]) • Gaof llagcontrasts her treats between this as arroots ising frwom par adigmlaryngeal atic disperconfigurasion constraint over the permitted ithadi fferent assimilate and dissimilate coexist in different corners of the same lanThe results reported in wo4n reveal a three-tiered hier454 le xicexperimental apuzzling lGillian root tyGallagher pes wit hsystemic respect to ejeis ctiunderstandable guage. This array of restrictions if markedness hierarchy projects two constraints, defined in distinctness in the perceptual ofmarkedness laryngeal contrasts in pairs of roots. constraints evaluate contrasts between sets of roots in a language of corresponding to the two asymmetries in (15). I refer the“di con5.2alyTsis heisconstraints • An supported by a speech perception task (“samto e” instead or fferent”) that finds it is more individual forms. that evaluate laryngeal contrasts between roots as laryngeal dis[k’ap’i-kapi] > [k’api-kapi] > [k’ap’i-k’api] Laryngeal co-occurrence eflect inerthe di t to constraints. judg e apaper C’ vs3.isCrestrictions contrast in trhe preasymmetries sence of anoth ejeperceptual ctive than when the The argument this that laryngeal co-occurrence restrictions ARDIST in or Lfficul B([T’-K’]:[T-K]) B([T’-K]:[T-K]) B ([T’-K’]:[T’-K]) strength of contrasts between roots with di ff erent laryngeal configuraare a unified andt is areplain best: i.e. understood accompphenomenon, anying consonan more errowhen rs on the C’VCperceptual ’ vs. C’VC than on CVC’ vs. CVC or tions. The experimental w4 reveal a three-tiered a. LarDist(2v1)-[F] properties of laryngeal contrasts results betweenreported roots arein taken into account. The hierhierarchy projects two systemic markedness constraints, defined in C’ Vtw C ovsin CV C perceptual archy the distinctness of laryngeal contrasts i n pairs If c on tra sting root s each have an [F ] segm ent, then th ey do tof roots. paper presents experimental support for a hierarchy of perceptual no discorresponding to the two asymmetries in (15). I refer to the conminimally di‰er in [F]. tinctness of laryngeal contrasts among roots, and develops an analysis of • that evaluate laryngeal > contrasts between roots dis(15) [k’ap’i-kapi] [k’api-kapi] > as laryngeal [k’ap’i-k’api] laryngeal co-occurrence based on markedness Let xR1, yR1, xR2,restrictions yR2 be segments thatsystemic can be specified for conthe D([T’-K’]:[T-K]) IST constraints.3 or LAR B B ([T’-K]:[T-K]) B ([T’-K’]:[T’-K]) straintslary projected from this idea ising that laryngeal ngeal feature [F]. hierarchy. Let R1 and The R2 becentral two contrast roots, such • a. th co-occurrence restrictions are not prohibitions against certain configuraat xR1, yR1ŒR1 and xR2, yR2ŒR2 and R1=R2, except that LarDist (2v1)-[F] hierarchy projects two systemic markedness constraints, defined in ≠ tions ofyThis laryngeal features, but rather are restrictions on the perceptual R1 y R2 . I f x R1 an d xR2 ar e [F], then R 1 an d R 2 ca nnot di‰e r on lyt an [F] segmen If twocorresponding contrasting root s the eachtwo have t,(15). thenIthey dotono (16), to asymmetries in refer the condistinctness of contrasts among possible roots in a language. The operative in a single instance [F]. minimally di‰er inof [F]. straints that evaluate laryngeal contrasts between roots laryngeal disconstraints in laryngeal co-occurrence restrictions not as standard •b. LarDist (1v0)-[F] Let xor R1,LyAR R1D , xIST R2,constraints. yR2 be seg3ments that can beare specified for the tance Optimality (OT; Prince Smolensky markedness conIflaryn twoTheory ge roots eac have tw thtat ma y be gcont eal rastin featur [F]. L ethR& 1 and Ro2 segments be two1993) con ras ting rospec ots, ified such straints.for Rather, they are systemic constraints that evaluate the marked[F], then they do not minimally di‰er in [F]. that a.xR1 , yR1ŒR1 and xR2, yR2ŒR2 and R1=R2, except that (16) LarDist (2v1)-[F] • of Le ness the set of possible language. This follows t≠ xR1 ,Ify. tw R1 ,tra ycontrasts bearrsoot m tsh thhave at1 an caann b]e2 seg sapproach pnnot em ciefie dthe fo rnon tthhly e y do not yR1 yR2 If,oxxR1 dR2 xing R2 eeg[in Fs],ean then R d[F R ca di‰e cR2 onan st ac nt, much previous work integrating systemic contrast markedness into Gallagher, Gillian. “Perceptual Distinctness and Long-Distance Laryngeal Restrictions.” Phonology 27, no. 3 (2010), phonological theory, including Contrast Preservation Theory roots, (Kubowicz lainryngea l feature [Fdi‰er ].ofLet R1[F]. and R2 be two contrasting such a single instance [F]. minimally in 435-80.© Cambridge University Press. This content is excluded from our Creative Commons license. For more 2003)b.information, and the Dispersion Theory of Contrast (Flemming 1995, 2006, that xR1, y R1 Œ R1 and x R2, yR2Œ R2. R 1 a n d R 2 c a n n o t d i‰ e r see Le t http://ocw.mit.edu/help/faq-fair-use/ xR1, yR1, xR2, yR2 be segmen. ts that can be specoifinelyd for the LarDist (1v0)-[F] Padgettin 2003, Sanders Chiosáin a single instance of Nı́ [F]. If two cont eac h].have w&1oaPadgett segments laryrastin ngea2003, lgfroots eatur e [F LettR nd R2 b2009). eth twatoma conytbe rasspec tingified roots, such the The first cpaper onstrain t is intended as to pe nalize just {ktypology ’ap’, k’api}ofbularyngeal t allows {k’a pi, kapi}, which does not is organised follows. The coat xthey and xR2,di‰er and R1=R2, ex cept that R1, ydo R1Œ yR2ŒinR2 then notR1 minimally [F]. for [F],th Loccurrence (2v1)-[F] penalises a an minimal given out w2, w3 inoutlines thelaryngeal contrast sa tAR isfD y IST theLet “if”restrictions uRyse ≠ 1R1 .isI, fylaid xR2 R1 dsexgin R2 rcontrast ets[and F], then Rba1ean deR nnot di‰er only xcla R1y,two ,yR2 xtoR2 b e which meacontains n tha can sp ci2fica edof for the feature between roots each of another segment specimarkedness approach the phenomenon. w4 reports the results a perinl feature a single[F]. instance of [F].R2 be two contrasting roots, la ry ngea Let R 1 and such ISTindividual’s (1v0)-[F] is ability more to general ; it penalises a fied for that feature.testing LARDan ception experiment distinguish laryngeal arDist (1v0)-[F] xstr R1L ,ain yinR1 Œgiven R1 and xR2, ytR2afeature Œ 1of nlaryngeal dd R2 odte d{ik‰er th e seconthat dcontrast conb. t ais m ore stringen ndR2. it inR te nade to two ecxacnln uroots ’api,only kofapi} and {k’api, kap’i} as minimal laryngeal between each contrasts between pairs of roots. The analysis co-occurrence If two contrasting roots each have two segments that may be specified in a with single instance of [F]. well as {k’ap’i, k’api} and allows just {kapi, k’ap’i}in w5, building on the rerestrictions systemic markedness is given for [F], then they do not minimally di‰er in [F]. sults of the perception experiment. w6 compares the systemic markedness approach with previous analyses of laryngeal restrictions, and w7 concludes. 6 e minimal contrast in a given laryngeal feature between two roots each of y kapi]}, {[kap’i, kapi]} and {[k’api, kap’i]}. Like candidate (e), candidates (f)–(j) satisfy highest-ranked LARDIST(2v1)-[ejective], but incur excessive violations of faithfulness, and are thus eliminated by high-ranking IDENT[ejective]. The tableau in (27) shows the evaluation of contrasting pairs with homorganic stops. With H-LARDIST(1v0)-[ejective] lowranked, dissimilation holds regardless of place of articulation. (27) Shuswap: dissimilation in ejection (homorganic) {/k’ak’i, k’aki, kak’i, kaki/} LarDist Ident LarDist H-LarDist (2v1)-[ej] [ej] (1v0)-[ej] (1v0)-[ej] **! ***** ***** a. {k’ak’i, k’aki, kak’i, kaki} Gillian Gallagher * *** *** canb. {k’aki, k’aki, kaki} ™{[k’ati, kati]}, {[kat’i, kati]} and {[k’ati, kat’i]}, eliminating kat’i]}, **! shows dissimilation, loses c. {k’ak’i, didates (a)–(d), (f)kaki} and (g). Candidate (c), which due to the contrasts {[k’ati, kati]}, {[kat’i, kati]} and {[k’ati, kat’i]}. Gallagher, Gillian. “Perceptual Distinctness and Long-Distance Laryngeal Restrictions.” Phonology 27, no. 3 (2010), DIST(1v0), but candidate Candidates (e) and (h)–(j) allThis satisfy 435-80.© Cambridge University Press. content L is AR excluded from our Creative Commons license. For more (e)information, preserves two forms instead of one, and is thus preferred by see http://ocw.mit.edu/help/faq-fair-use/. of Shuswap in (26) and is a specific instantiation of the [ejective]. The formulation of L(27) ARDIST(1v0) to penalise positional IDENTanalysis (2v1)-[F]7I ranking schema for dissimilation : LARDIST • typo: b is {k’aki, k’i, kaki}of assimilation, contrasts is crucial to thekaanalysis as candidates (e)DENT and[F] (f) ARDIST(1v0)-[F], H-LARDIST(1v0)-[F]. Given this ranking, neutral7L AR D IST (2v1). do equally well on faithfulness and low-ranking L • if athe pair of rootscontrast each havevsa.n1) ejeisctiv e then th ey cannot cothe ntrastvsm by virtue of the isation weakest required. Neutralising . inimally 1 The of tableau in (33) shows (2 that assimilation also results when 2homcontrastpr to one laryngeal eseforms nce or with absenc econsidered. of ejection:feature excludeallows s C’VC’for V vst.hree C’VCcontrasting V organic pairs of stops are forms, as opposed to the two-way contrast that results from neutralisation (33) Chaha: assimilation in ejection (homorganic) to a form with two laryngeal features. When homorganic and heterorganic {/k’ak’i, k’aki, kak’i, kaki/} LarDist Ident LarDist H-LarDist (1v0)-[ej] [ej] (2v1)-[ej] (1v0)-[ej] ** ***** a. {k’ak’i, k’aki, kak’i, kaki} *!**** *!** * *** b. {k’aki, kak’i, kaki} ** ™ c. {k’ak’i, kaki} Gallagher, Gillian. “Perceptual Distinctness and Long-Distance Laryngeal Restrictions.” Phonology 27, no. 3 (2010), 435-80.© Cambridge University Press. This content is excluded from our Creative Commons license. For more LARDinformation, IST(1v0)-[ejective] forces neutralisation of contrasts in ejection in see http://ocw.mit.edu/help/faq-fair-use/. roots with other voiceless stops, but not other voiced stops, as can be seen • tableau if a pair in of roots iffer inallejection then they differvoiceless in ejection in the (34). dHere, combinations of must ejective, andmaximally, i.e. at each C voiced stops are evaluated in parallel. The contrasts {k’adi, kadi} and (1v0), constraints because [g m d]ustbear are not outtobysystemic LARDIST A{gat’i, majorgati} question theseruled appeals contrast facethe is what is the candidate set feature [voice] and thus cannot contrast for [ejective]. over which the constraints are operating? This remains an outstanding research question. (34) Chaha: no assimilation between ejectives and voiced stops ----{/k’at’i, k’ati, kat’i, k’adi, gat’i, LarDist Ident LarDist H-LarDist (2v1)-[ej] (1v0)-[ej] kadi, gadi, kati/} (1v0)-[ej] Choi, John.gati, 1992. Phonetic underspecification and[ej] target interpolation: an acoustic study of *!**** ** a. {k’at’i, k’ati, kat’i, k’adi, gat’i, Marshallese vowel allophony. UCLA Working Papers in Phonetics 82 gati, kadi, gadi, kati} *20, 287-333. *!** {k’at’i, kat’i,Feature k’adi, gat’i, gati, Phonology Clements,b.G.N. 2003. Economy. kadi, gadi, kati} Flemming, Edward (1995, 2002). Auditory Representations in Phonology. Garland Press, New York ** ™ c. {k’at’i, k’adi, gat’i, gati, kadi, gadi, kati} Flemming, Edward (2004). Contrast and perceptual distinctiveness. Bruce Hayes, Robert ***!* Bases of Markedness. Cambridge d. {k’at’i, gati, kadi, Steriade gadi, kati}(eds.) The Phonetic Kirchner, and Donca In Chaha, the ranking of LARDIST(1v0)-[ejective] favours assimilation, University Press, Cambridge. 232-276. as in candidate (c), over dissimilation, as in candidate (b). Candidate (d), which eliminates all forms with only a single ejective, incurs excessive Fl emming, of Edw ard (2006).The The winning role of dicandidate stinctivene(c) ss cneutralises onstraints ithe n ph violations faithfulness. 1onol vs. 0ogy. MIT ms. contrast in roots with two voiceless stops, {[k’ati, kati]}, {[kat’i, kati]} and {[k’ati, roots Marymaintaining Beckman, &the Jan1 Evsd.w0acontrast timeone is necessary but not always Ko ng, Eukat’i]}, n Jong,while rds. 2012in. Vo ice with onsetonly voiceless stop {[k’adi, kadi]} and {[gat’i, gati]}. The violations of LARDIST sufficient to describe acquisition of voiced stops: The cases of Greek and Japanese. Journal constraints in (34) are given in (35). of Phonetics 40, 725-44. Gallagher, Gillian. 2010. Perceptual distinctness and long-distance laryngeal restrictions. Phonology 27, 435-80. Padgett, Jaye. 2003. Contrast and post-velar fronting in Russian. Naturtal Language and Linguistic 7 Theory 21.1, 39-87. Appendix on [voice] Various phonologists (Iverson, Ringen, Jessen, Vaux, and others) argue that the Germanic languages differ in the feature that contrasts /p,t,k/ vs. /b,d,g/: English, German [spread gl], Dutch [voice]. This position is critiqued by Kingston and Lahiri (K&L). • proponents of [spread gl] point to the fact that in many contexts there is no phonetically observed voicing and therefore this is prima facie evidence that [spread gl] is distinctive • K&L term this an “essentialist” view of a feature: this is a core set of phonetic correlates that should appear in every realization [±F] of the feature (cf. structuralists invariance condition on phonemes). They deny this, at least for [voice], claiming that a voicing contrast can be implemented by a variety of phonetic gestures whose distribution and magnitude vary according to context and no one gesture has privileged status. What unites them is a perceptual integration. • voicing in sonorants, stops, and fricatives is phonetically diverse but yet they pattern as a natural class for the past tense and plural allomorphs in English (assumes z and d are not the defaults). • passive voicing is really actively and purposely produced • perceptual integration to give an “intermediate perceptual property” IPP. Performed by the auditory system and can occur even when sound in not heard as speech: low frequency of F0 (and F1) with vocal fold vibration and duration of consonant with preceding vowel duration. • permits a more abstract view of a distinctive feature; so apparently VOT will integrate with high F0 to define the voiceless value signaling the open glottis gesture in the adjacent vowel. 8 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.