A nonagreement analysis of Harari long-distance V-C palatalization (talk)

Background ABC in Harari Assumptions Analysis Future Research

A Nonagreement Analysis of HarariVowel-Consonant Palatalization

Radu Craioveanu & Ross Godfrey

June 19, 2014

Radu Craioveanu & Ross GodfreyA Nonagreement Analysis of Harari Vowel-Consonant Palatalization


Agreement by Correspondence

Intro to ABC

Agreement by Correspondence was proposed to account for long-distanceconsonant agreement (LDCA) phenomena, where classes of segmentsdisplay featural harmony with complete transparency of all interveningmaterial.Harmony is implemented through two Optimality Theoretic elements:Correspond constraints and Agree constraints

⑴ ABC configuration (adapted from Rose and Walker 2004, ⑶)

C V C V

[±F] [±F]

Cx V Cx V

[αF] [αF]

CorrC↔C

AgreeC↔C




Similarity in Correspondence

Rose and Walker (2004), in their typology of long-distance consonantagreement (LDCA) patterns, note that interacting segments share majorfeatures like sonorant, continuant, or PlaceSegments in participating classes differ by one or more minor features(e.g., constricted glottis, retroflex, lateral)Rose and Walker hold that similarity is a requirement in LDCA, and setsthis phenomenon apart from other assimilatory systems

The authors note the algorithm of Frisch et al. (2004) is successful fordetermining corresponding consonant classessimilarity = shared feature classes

shared+ nonshared feature classes




Blocking in LDCA

Another major characteristic of LDCA is that intervening segments donot block agreementAssimilatory processes that do show blocking effects are characteristicallyunlike LDCA (Rose and Walker, 2004):

They are local.They affect contiguous strings of segments.Vowels may trigger assimilation or participate in harmony.

Summary: LDCA operates on segments within a class, at a distance,without affecting or being affected by intervening material; ABC realizesthese properties using constraints.

Similar segments are put into correspondence by CorrC↔C constraintsIdentity between corresponding segments is enforced by Agree constraints



Data

Long-Distance Palatalization (LDP) in Harari

In Harari (Ethio-Semitic), the 2nd person feminine singular suffix -itriggers palatalization of coronal obstruents and sonorants in the stem.This is easiest to see with imperatives.

⑵ Masc. Imp. Fem. Imp.a. lɨbəs lɨbəʃ-i ‘dress!’b. kɨfəl kɨfəj ‘pay!’c. kɨtəb kɨʧəb-i ‘write!’d. ħɨnək ħɨɲək’-i ‘strangle!’e. sɨxər ʃ ɨxər-i ‘be drunk!’f. a-t-a-bargɨ a-ʧ-a-barg-i ‘don’t startle!’



Data

Long-Distance Palatalization (LDP) in Harari

When a coronal sonorant is stem-final and a coronal obstruent appearsearlier in the stem, both get palatalized (Rose 2004, ⑺).

⑶ a. xɨdən xɨʤəɲ-i ‘cover!’b. gɨdəl gɨʤəj ‘kill!’

Palatalization of multiple coronal obstruents is optional (Rose 2004, ⑿).

⑷ a. bɨt’əs bɨt’əʃ-i, bɨʧ ’əʃ-i ‘rip!’b. sɨdəb sɨʤəb-i, ʃ ɨʤəb-i ‘insult!’



Previous analyses

Agreement analysis (Rose 2004)

Rose (2004) offers an analysis of these facts in Harari, settling on anagreement-based mechanism because of the transparency of interveningsegments and preference for coronal obstruents in palatalizationRose’s analysis hinges on Stem-Affix Identity (rephrased below):

SA-Ident: If a suffix vowel α is [+pal], there must be a stem consonant βsuch that αℜβ, and β must be [–pal] in the input and [+pal] in the output.Essentially, a palatal suffix vowel enters into a relationship with a stemconsonant such that it palatalizes it.

Rose’s analysis does not explicitly use Corr constraints, but SA-Identbuilds in a form of correspondence by establishing a relationship betweenthe trigger vowel and target consonant.



Theory & Goals

Theoretical Motivations

Agreement by Correspondence (ABC) emphasizes similarity ofcorresponding segments.Vowels and consonants are not terribly similar, and a language should notestablish vowel-consonant correspondences (V↔C) if it doesn’t havecloser correspondences (V↔V or C↔C) as well.

Harari does not display these correspondences.

Thus, long-distance VC interactions in languages like Harari should not beanalyzed as cases of ABC.



Theory & Goals

Theory & Goals

We seek to provide a more satisfactory analysis of Harari long-distancepalatalization, and also strengthen the idea that all agreementrelationships should be mediated by similarity-based correspondence.

We will therefore be pursuing an analysis based on spreading and floatingfeatures.But first, some further data must be introduced.



More data

Spreading, Docking, and Blocking

Harari also has a local palatalization process, seen in words with agentive-i, which is amenable to a spreading analysis.

⑸ Root Agentivea.

√hrs həraʃ-i ‘one who ploughs’

b.√gdl gədaj (*gəʤaj) ‘one who kills’

c.√dlg dəlag-i (*ʤəlag-i) ‘one who works’



More data

Spreading, Docking, and Blocking

While Rose argues that Harari LDP does not display blocking effects,these could simply be difficult to detect due to rampant optionality ofmultiple palatalization.The data in ⑹ is amenable to a floating feature analysis.

⑹ Preferred Dispreferreda. a-t-bɨrər a-ʧ-bɨrər-i a-t-bɨrər-i ‘don’t fly!’b. a-t-bɨʃək’ a-t-bɨʃək’-i a-ʧ-bɨʃək’-i ‘don’t be soaked!’

If underlying /ʃ/ gets vacuously palatalized, the acceptability of[a-ʧ-bɨʃək’-i] in (6b) follows from the same mechanism that allowsoptional multiple palatalization of coronal obstruents (which is alsodispreferred for many speakers).



Structures

Underlying Forms

In our analysis, the LDP-triggering suffix has the underlying form in ⑺,while the agentive suffix has the form in ⑻.

⑺ i [+pal]

×

[+pal]/ /

LDP suffix

⑻ i [+pal]

×

/ /

Agentive suffix

This suffix consists of a floating [+palatal] feature, followed by thephoneme /i/ (which is itself [+palatal]).Wolf (2007) also has an OT treatment of affixes with both segmental andfloating content.



Constraints

Constraints

We use the following constraints in our analysis:1 *[±pal]i : Nonepenthetic [i] and a segment immediately to its left maynot bear individual [±pal] features. This constraint enforces localspreading.

2 Ident[+pal]-IO: (i) A segment that is [+pal] in the input must be[+pal] in the output; (ii) If there is an unlinked [+pal] feature in theinput, there must be an instance of [+pal] in the output belonging solelyto a stem segment.

3 Ident[+pal]-OI: If there is an instance of [+pal] belonging solely to astem segment in the output, then either (i) that segment must be [+pal]in the input, or (ii) the input must contain an unlinked [+pal] feature.



Multiple Palatalization

Preferred Multiple Palatalization

Cases of preferred multiple palatalization are reanalyzed as the result oftwo processes:

Local spreading of the [+pal] feature from /i/Docking of the floating [+pal] feature on the suffix

Segments that share a [pal] feature are indicated with parentheses.

⑼ /fɨt’ən-[+pal]i/ *[±pal]i Id[+pal]-IO Id[+pal]-OIa. fɨʧ ’əni *!b. fɨt’ə(ɲi) *!

� c. fɨʧ ’ə(ɲi)




Optional Multiple Palatalization

We use competing grammars to capture optionality. In the alternativegrammar, we see the emergence of ABC effects; we analyze optionalmultiple palatalization as agreement of consonants by correspondence.Additional palatalization does not incur extra violations ofIdent[+pal]-OI when there is a floating feature in the input.

⑽ Alternative grammar/sɨdəb-[+pal]i/ Id[+p]-IO IdCC[pal] Id[+p]-OI Corr-T↔T

a. ʃɨʤəbi *!b. sɨdəbi *!c. sɨʤəbi *!

� d. ʃɨʤəbi




Optional Multiple Palatalization

We need to prevent possible overapplication of palatalization harmony inthe alternative grammar.The formulation of the Ident constraints allows this agreement to belimited to cases where palatalization is triggered by a floating feature.

⑾ Alternative grammar/saʧa/ Id[+p]-IO IdCC[pal] Id[+p]-OI Corr-T↔T

� e. saʧa *f. sata *!g. saʧa *!h. ʃaʧa *!



Sonorant behaviour

An Outstanding Issue

How do we account for the special behaviour of sonorants, which we sawearlier in ⑶?

⑶ a. xɨdən xɨʤəɲ-i ‘cover!’b. gɨdəl gɨʤəj ‘kill!’

Our analysis predicts that stems with both medial and final coronalobstruents should undergo preferred multiple palatalization, contrary toreports by Rose.This could be explained if local spreading preferentially affects sonorants,but this is yet to be confirmed.



Bonus: Towards an alternative analysis



An Analytical Alternative

Conjunctions and Disjunctions

We have relied on a coǌunctive Ident[+pal]-IO and disjunctiveIdent[+pal]-OI to feed agreement in optional multiple palatalization.

Ident[+pal]-IO: ⒤ segments stay palatal & (ii) floating [pal]Ident[+pal]-OI: either ⒤ input segment was palatal OR (ii) floating [+pal]

The use of coǌunctions and disjunctions in our constraint definitionsmay raise eyebrows (and already has!).Our justification is that they provide instructions in the case of floatingfeatures in the input, which are otherwise invisible to segment-focusedIdent-IO constraintsBut perhaps this feeding effect can be captured another way. We give anincredibly embryonic sketch in the remaining time.



An Analytical Alternative

Multiple Correspondence

We follow in large part Finley (2009) in providing this alternative analysis.She hypothesizes that input floating features can correspond to multipleoutput segments, and accounts for many patterns of “morphologicalharmony” in this way.Let’s see if we can derive this with as few unmotivated stipulations aspossible.



Stipulations & Consequences

Defective Inputs

Let’s start with a motivated stipulation. Say Gen doesn’t generate outputswith floating features.¹This could be expressed as an interface condition, e.g., features must belinked to segments in order to be interpreted by the phonetic component,and Gen doesn’t generate any candidates that can’t be interpreted.This means that some inputs with floating features would be defective, inthe sense that fully faithful outputs would not be generated.

¹ Note that one of the selling points of McCarthy and Prince’s (1995) original formulation of Cor-respondence Theory was that outputs with unlinked features were no longer necessary. In Princeand Smolensky’s (1993) Containment Theory, input features could never truly be deleted.




A Harari Example

Now imagine a grammar with high-ranked Max-IO. In this case, itwould be bad to delete the floating feature (violating the highly-rankedconstraint), but impossible to leave it unlinked (because that candidateisn’t generated). What do you do?

Here’s a possibility, using Harari data. We revise classic CorrespondenceTheory slightly by tying correspondence to root nodes (“feature bundles”)rather than segments (i.e., skeletal ×-slots).We consider Ident-IO to be violated only when corresponding input andoutput segments have contradictory values for a feature. Deletion of afeature (e.g., with underspecified outputs) doesn’t induce a violation; thus,no violation is incurred by an output segment [b] (unspecified for [pal])that corresponds with a [+pal] floating feature.




A Harari Example

⑿ /k₁ɨ₂t₃ə₄b₅-[+pal]₆i₇/ Max-IO *NonCorPal Ident-IOa. k₁ɨ₂t₃ə₄b₅i₇ *!

� b. k₁ɨ₂t₃₆ə₄₆b₅₆i₇ * (t₆)� c. k₁ɨ₂ʧ₃₆ə₄₆b₅₆i₇ * (ʧ₃)

Without additional stipulations, this tableau predicts a tied outcome:⒝: /t/ does not palatalize, violating Ident-IO w.r.t. index 6⒞: /t/ palatalizes, violating Ident-IO w.r.t. index 3

It could be that violating Ident-IO with respect to “6” is worse, since100% of its input features have been changed. If we allow for that sort ofgradable violation, then we correctly predict the winner.




MaxFloat and its Kin

We confess that such an analysis may still need something like MaxFloat(Wolf 2007, but the original idea can be traced back to Zoll 1996) to getoff the .Otherwise, the candidate in ⒝ would win, and count as fully faithful,even though the [+pal] does not surface visibly.

⒜: floating [+pal] is deleted outright, violating Max-IO⒝: floating [+pal] corresponds only to /b/, where it is not contrastive

⒀ /k₁ɨ₂t₃ə₄b₅-[+pal]₆i₇/ Max-IO *NonCorPal Ident-IOa. k₁ɨ₂t₃ə₄b₅i₇ *!

� b. k₁ɨ₂t₃ə₄b₅₆i₇




Contiguity & Directionality of Correspondents

How are contiguity and directionality enforced in multiple outputcorrespondence to an input root node?What does it mean for an input and an output to have contradictoryprecedence relations, when one input bundle is floating?

Contiguity: a reworking of the Linearity constraint could help define thisfor floating feature associationDirectionality: could be established with additional Align constraints




Conclusion

This is obviously an extremely rough sketch, but it’s where we hope totake this in future work.We wish to show that the special behavior of floating features is due tothe fact that fully faithful outputs are not generated in these cases,allowing ordinarily blocked cases of multiple correspondence to surface.Cases of morphological harmony, where presence of morphemic floatingfeatures feeds agreement in the root, could arise from high-rankedCorrX↔X constraints in combination with the kind of constraint ranking(and defective input) seen in ⑿.The drastically different behavior of prototypical consonant harmony(e.g., no blocking effects) could be captured by maintaining an ABCanalysis in these cases, where Linearity isn’t a concern (because thecorrespondence in this case is purely an output correspondence), andproblematic segments can simply be skipped, exempt from thecorrespondence relation.



Thanks

Our sincerest thanks goes to:

Alexei Kochetov for guiding and funding for our research,our primary consultant Mohammed Abdella for his patience and enthusiasm,and the participants of the 2014 Agreement by Correspondence Conference at

UC Berkeley for their insightful comments.



References

Finley, Sara. 200⒐ Morphemic harmony as featural correspondence. Lingua 119:478–50⒈Frisch, Stefan, Janet Pierrehumbert, and Michael Broe. 200⒋ Similarity avoidance and the ocp.

Natural Language and Linguistic Theory 22:179–22⒏McCarthy, John, and Alan Prince. 199⒌ Faithfulness and reduplicative identity. In Papers in

Optimality Theory, ed. Jill Beckman, Laura Walsh Dickey, and Suzanne Urbanczyk, 249–38⒋Amherst, MA: Graduate Linguistic Student Association, University of Massachusetts.

Prince, Alan, and Paul Smolensky. 199⒊ Optimality Theory: Constraint interaction in generativegrammar.

Rose, Sharon. 200⒋ Long-distance vowel-consonant agreement in Harari. Journal of AfricanLanguages and Linguistics 25:41–8⒎

Rose, Sharon, and Rachel Walker. 200⒋ A typology of consonant agreement as correspondence.Language 80:475–53⒈

Wolf, Matthew. 200⒎ For an autosegmental theory of mutation. In Papers in Optimality TheoryIII , ed. Leah Bateman, Michael O’Keefe, Ehren Reilly, and Adam Werle, 315–40⒋ Amherst:University of Massachusetts, Graduate Linguistic Student Association.

Zoll, Cheryl. 199⒍ Parsing below the segment in a constraint-based framework. DoctoralDissertation, University of California, Berkeley.


A nonagreement analysis of Harari long-distance V-C palatalization (talk)

Documents

Transcript of A nonagreement analysis of Harari long-distance V-C palatalization (talk)