Anda di halaman 1dari 14

Weeks 5-6 Optimality Theory Approaches to Opacity in OT

March 8, 2011

Contents
1 Constraint Conjunction 2 Two-level constraints 3 Stratal OT 4 Candidate Chain Theory 5 Output to Output Correspondence 6 Opacity is not a disease but a phenomenon Please read chapters 1-2 of Hidden Generalizations (2007) by John McCarthy posted on the course website. 1 3 4 7 8 12

1
(1)

Constraint Conjunction
Constraint Conjunction is one strategy people have used to handle saltation. a. Anna Lubowicz (1998) Derived Environment Eects in OT1 b. Ito, Junko and Armin Mester (1999) Recall the Campidinian Sardianian data: a.
1

(2)

> Intervocalic lenition of voiceless stops /p, t, k/ and the voiceless aricate /tS/ (underlying forms justied by appearance in isolation):

Lubowicz, Anna. 1999. Derived environment eects in OT. The Proceedings of the Seventeenth West Coast Conference on Formal Linguistics, ed. by Kimary Shahin, Susan Blake and Eun Sook Kim, 451-65. Stanford, CA: Center Study Language & Information. [Rutgers Optimality Archive 239]

Weeks 5-6: Optimality Theory Approaches to Opacity in OT /bEl:u piS:i/ [bEl:u BiS:i] nice sh /s:u trintaduzu/ [s:u Drintaduzu] the thirty-two /dE kuat:ru/ [dE Guat:ru] of four... > /s:u tSElu/ [s:u ZElu] the heaven Preservation of underlying intervocalic /b, d, g/: /s:a bia/ [s:a bia] the road /s:u gat:u/ [s:u gat:u] the cat /don:ia dominiGu/ [don:ja dominiGu] every Sunday

J. Heinz

b.

(3)

Lubowiczs answer, endorsed by Ito/Mester: constraints that are violated when two conditions are met in the same location (local conjunction). a. For Campidinian Sardinian: *Ident(voice) AND *V[-cont]V b. Intuition: Dont have a intervocalic stop if youre also changing voice Fill in constraints and reassure yourself that this gets you what is needed:
/apa/ aBa apa aba /aba/ Id(vce) & *V[-cont]V *V[-vce]V ID(voice) *B Id(cont) *V[-cont]V

(4)

a. b. c.

Id(vce) & *V[-cont]V

*V[-vce]V

ID(voice)

*B

Id(cont)

*V[-cont]V

a. b. c.

aba apa aBa /Ba/ ba Ba pa Id(vce) & *V[-cont]V *V[-vce]V ID(voice) *B Id(cont) *V[-cont]V

a. b. c.

(5) (6) (7)

Conjunction has to be local to avoid insane results like: /tIk/ [tIk], /pag-tIk/ [paktIx] (the assimilatory voicing of the rst /g/ permits the second /k/ to spirantize). As constraint families grow, the constraints grow. . . If there are m possible Markedness constraints and n Faithfulness constraints, there are now m n constraints. . . . Factorial Typology a. Form of the constraint: dont be both marked and unfaithful b. Lubowiczs intent: if youre unfaithful, dont be marked c. An additional consequence: if youre marked, be faithfulmeaning we can get special entities arise only in a marked context, which is going to be a problem 2

Weeks 5-6: Optimality Theory Approaches to Opacity in OT (8) An Apparent Disaster Case /ba/  pa ba /aptka/ aptka apka Id(voice) & *CCC Max(C) *[-son, +voice] *CCC

J. Heinz

Id(voice)

Id(voice) & *CCC

Max(C)

*[-son, +voice]

*CCC

Id(voice)

/apdka/  apdka aptka (9)

Id(voice) & *CCC

Max(C)

*[-son, +voice]

*CCC

Id(voice)

What is wrong? a. I think it is the primitive approach to Faithfulness in classical OTjust Ident, Max, Dep, and a few others. b. We need to expand the power of the theoryin a way that doesnt give it insanely excessive power. c. Some potentially relevant ideas: (i) Steriades view that unfaithful mappings prefer to be phonetically minimal (ii) The view that UG can impose bias, and not just absolute bans

2
(10) (11)

Two-level constraints
This strategy introduces constraints that are neither markedness nor faithfulness constraints, but some combination. They are constraints on alternations. Ref. John McCarthy (1996) Remarks on phonological opacity in Optimality Theory. In Jacqueline Lecarme, Jean Lowenstamm, and Ur Shlonsky, eds., Studies in Afroasiatic Grammar. Papers from the Second Conference on Afroasiatic Linguistics, Sophia Antipolis, 1994. The Hague: Holland Academic Graphics. Pp. 215243. Recall Polish counterbleeding opacity:

(12)

Weeks 5-6: Optimality Theory Approaches to Opacity in OT sg. a) b) c) d) e) f) g) h) i) j) k) l) m) n) (13) pl. horse bow sheaf cat nose juice club labor rubble lye crib ice cart horn

J. Heinz

trup trupi wuk wuki snop snopi kot koti nos nosi sok soki klup klubi trut trudi grus gruzi wuk wugi wup wobi z z lut lodi vus vozi ruk rogi

For Polish /voz/ [vus], we could propose a constraint like: A nonlow back vowel must be high on the surface, if it is followed by a voiced obstruent at the level of underlying representation.

Why/How is this dierent from contextual markedness?

Why/How is this dierent from contextual faithfulness?

(14)

This seems most straightforward but has not been widely favored, perhaps because it seems like a brute-force solution.

What would the non-surface constraint(s) look like for Campidinian Sardinian?

3
(15) (16)

Stratal OT
Counterbleeding, counterfeeding, and saltation, are all manageable in OT with levels. The basic idea is to permit a limited, prespecied number of levels of structure. These, and only these, may be involved in serial derivation. (Goldsmith 1993, McCarthy and Prince 1993). 4

Weeks 5-6: Optimality Theory Approaches to Opacity in OT (17) (18)

J. Heinz

Goldsmith (1993) suggests an M(orpheme)-level, a W(ord)-level and a P(honetic)-level See also Kiparsky (2000?) for a version of Lexical Phonology incorporating OT (called LPM-OT), in which all derivationality follows from the structure of the grammar. The levels here are essentially stem, word, and postlexical. 2 Recall Catalan has processes of word-nal non-continuant deletion and nasal assimilatio, so that a UR like /bEn+k/ surfaces as [bEN] I sell.

(19)

Why do we call this relation opaque?

What kind of opacity is this, counterbleeding or counterfeeding?

(20)

With levels:

Stem level
+nas a place
!

/bEn+k/ doubly bad b/c of nk] assimilation xes only one problem deletion not allowed deletion not allowed bEnk  bENk

MaxC

+cons a place

*[+nas][-cont] ]

Ident (place)

bEn bEN

! !

So the output of the stem level [bENk] becomes the input to the word level:

The paper is here: http://www.stanford.edu/~kiparsky/Papers/vdhulst-ritter.pdf

Weeks 5-6: Optimality Theory Approaches to Opacity in OT Word level


+nas a place
(!)

J. Heinz

/bEn+k/ doubly bad b/c of nk] (plus no reason to change back to [n]) bad b/c *Nk] deletion xes remaining problem no reason to change back to [n] bEnk

*[+nas][-cont] ] (!)

+cons a place

MaxC

Ident (place)

bENk bEN

bEn

(21)

So the nal output is [bEN]. a. No further assimilation applies at postlexical level, since this is the only word in the utterance. Recall in Bedouin Arabic, underlying /katab/ [kitab], but /gabr/ [gabur].

(22)

Why do we call this relation opaque? Is this counterbleeding or counterfeeding?

(23)

With levels:

Earlier level (dont know whether stem or word) /gabr/ bad b/c of br] epenthesis not allowed epenthesis not allowed  gabr ga.bur gi.bur (!) ! (!) DEP-V *a] *COMPLEX CODA IDENT(hi) IDENT(lo)

Later level (perhaps postlexical)

Weeks 5-6: Optimality Theory Approaches to Opacity in OT /gabr/ bad b/c of br] epenthesis xes one problem but creates a new one raising not allowed gabr  ga.bur *COMPLEX CODA ! IDENT(hi) IDENT(lo) DEP-V

J. Heinz *a]

gi.bur

(!)

(!)

Discussion. What are your thoughts on Stratal OT?

4
(24) (25)

Candidate Chain Theory


Reference: McCarthy, John. 2007. Hidden Generalizations: Phonological Opacity in Optimality Theory. Equinox Publishing, Ltd. Essential idea: The constraints in CON do not evaluate (underlying,surface) pairs. Instead, they evaluate a sequence of derivational steps from underlying form to surface form. This sequence is a called a candidate chain and this version of OT is abbreviated OT-CC. GEN only produces candidate chains that are: a. gradually divergent (minimal change at each step) b. harmonically improving (the change cannot have led to increased ill-formedness) Examples of candidate chains for input /apa/ (given high-enough ranked *V[-cont]V, *V[-voice]V): a. <apa> b. <apa, aba> c. <apa, aba, aBa> d. <apa, aFa, aBa> Standard Markedness and Faithfulness constraints a. Markedness constraints apply as before; i.e only taking as arguments the nal surface form.

(26) (27)

(28)

(29)

Weeks 5-6: Optimality Theory Approaches to Opacity in OT b.

J. Heinz

Faithfulness constraints apply as before too (evaluated by considering underlying form and surface form). But note this is the sum of all the individual changes in the chain b/c of harmonically improving requirement. So we can think of each step in the chain as violating some faithfulness constraint.

(30)

OT-CC introduces a new kind of constraint called PREC constraints. These constraints determine the well-formedness of chains. a. Prec constraints have the form PREC(F1,F2) which intuitively means a chain in which a violation of F2 occurs before a violation of F1 is ill-formed. (see McCarthy 2007:98) Violations are assigned in two instances: a. Each instance of a F2 violation not preceded by a F1 violation. b. Each instance of a F1 violation that is preceded by a F2 violation. So a chain with only one F1 and F2 violation where the F2 violation occurs earlier in the chain than the F1 violation incurs two violations of PREC(F1,F2). Commentary on this theory: see Colin Wilson (ms., 2006) Counterfeeding from the Past (available upon request), shows that Candidate Chain theory allows: Rule-ordering theory makes the following prediction about the interaction of these processes: if [word-nal consonant] deletion feeds [word-nal raising (/pek/ pe [pi]) and raising feeds assibilation of /t/ to [s] before [i] (/te/ ti [si]), then instances of raising that are fed by consonant deletion also feed assibilation (/tek/ te ti [si]).

a.

b. (31)

(32)

The bizarre case predicted by OT-CC a. /pek/ pe [pi] b. /te/ ti [si] but c. /tek/ te ti [ti] d. would follow from Dont violate Ident(continuant)

Max(k)

before

you

violate

5
(33)

Output to Output Correspondence


Output-to-output correspondence was proposed independently of any need to handle opacity. It was originally proposed to address issues which, in rule-based theories, were addressed with cyclity. I will cover this rst. References: a. Benua, Laura. 1997. Transderivational identity: Phonological relations between words. Doctoral dissertation, University of Massachusetts, Amherst. Rutgers Optimality Archive ROA-259, http://ruccs.rutgers.edu/roa.html. b. Burzio, Luigi. 1996. Multiple correspondence. Ms. Johns Hopkins University, Baltimore, Md. 8

(34)

Weeks 5-6: Optimality Theory Approaches to Opacity in OT c.

J. Heinz

d.

e.

Kager, Ren. 1999. Surface opacity of metrical structure in Optimality Theory. e In The Derivational Residue in Phonology, ed. Ben Hermans and Marc van Oostendorp. Amsterdam: John Benjamins. Kenstowicz, Michael. 1997. Uniform exponence: Exemplication and extension. 1997. In V. Miglio & B. Moren, eds., Selected papers from the Hopkins Optimality Workshop 1997, University of Maryland Working papers in Linguistics 5, 139-54; and on line. Steriade, Donca. 2000. Paradigm Uniformity and the phonetics-phonology interface. Papers in laboratory phonology V. Cambridge: Cambridge University Press.

(35)

Kager (1999) argues for an output-ouput-correspondence-based analysis of Arabic stress, syncope, and epenthesis. In particular, he argues for correspondence between the base and the axed form. Denition of base (Kager p. 282) a. The base is a free-standing output form of a word. b. The base contains a subset of the grammatical features of the derived form. Usually, the base has one ax fewer than the axed form. So, in most cases, to nd the base, remove the most recently added ax. If the result is a real word, thats the base target output cats condensation unfathomable fathomable subhuman substrate base cat condense fathomable fathom human (*strate is not a word)

(36)

(37)

(38) (39)

Let us recall the Palistinian Arabic data we looked at last semester which motivated the cycle: Classic example: Palestinian Arabic (data originally from Brame (1974)) a. Verbs without objects subject 2sg. masc. 2sg. fem. 3sg. masc. 3sg. fem. 1pl. 2pl. 3pl. study da.rs+t a da.rs.+ti a d.ras a d.ra.s+at a da.rs.+na a da.rs.+tu a d.ra.s+u a understand fh m+t fh m.+ti f .him f h.m+at fh m.+na fh m.+tu f h.m+u

Weeks 5-6: Optimality Theory Approaches to Opacity in OT (40)

J. Heinz

The following processes appear to be occurring: a. i-Syncope: unstressed /i/ deletes in open non-nal syllables b. Stress rule: if penult is light, stress falls on antepenultotherwise it falls on penult

In the data above, which rule must apply rst?

(41)

Verbs with objects object 1sg. 2sg. masc. 2sg. fem. 3sg. masc. 3sg. fem. 1pl. 2pl. 3pl. he understood X .h m.+ni f h.m+ak f h.m+ik f h.m+u .h m.+ha .h m.+na .h m.+kum .h m.+hum

(42)

In particular interest is the following form: a. /him++na/ [.h m.+na] He understod us

Why doesnt the i-Syncope apply yielding *[fh m.na]?

(43) (44)

Brame (1974) observes that forms in which [i] fails to delete are all based on a free form (here: [f him]). In classic OT, there is no concept of the cycle, and so the theory of Output-toOuput (OO) theory was developed. In addition to Input-Output correspondence constraints, this theory proposes that there are also Base-Axed correspondence constraints. IO-MAX-V/stress: BA-MAX-V/stress: every stressed vowel in the input must have a correspondent in the output every stressed vowel in the base must have a correspondent in the axed form (both are outputs)

(45)

Heres how Kager (1999) blocks syncope in h mna he understood us:

10

Weeks 5-6: Optimality Theory Approaches to Opacity in OT


input: /him+na/ base: [f him]  he understood us he understood h mna fh mna

J. Heinz

BA-MAX-V/stress !

*i]

IO-MAX-V

Why is it dierent in fh mna we understood ?


input: /him+na/ base: none he understood us (theres no such word as understood in this language) h mna fh mna

BA-MAX-V/stress

*i]

IO-MAX-V

If Arabic had a word that just meant understood, perhaps pronouned f him, would there still be a dierence between the two na suxes under Kagers approach?

(46)

Heres how Kager gets the fact that stress seems to ignore epenthetic vowelsit doesnt have anything to do with the base: IO-DEP-V/stress: every stressed vowel in the output must have a correspondent in the input IO-DEP-V/stress WSP * *!

input: /Pakl+ha/ her food base: [Pakil] food  P.kil.ha a Pa.k l.ha (47)

Incidentally, Kiparsky has argued against this analysis, saying that it misses a key point about epenthesis in Arabic: its not just that epenthetic vowels cant be stressed. Rather, they are ignored by a lot of phonology. Kiparsky proposes that we can explain it all if we say that epenthesis happens at the postlexical level, and the other changes happen earlier (before the epenthetic vowel is there). (Kiparskys data seem to be from mix of varieties of ArabicIm not sure which is which.)

(48)

11

Weeks 5-6: Optimality Theory Approaches to Opacity in OT

J. Heinz

6
(49) (50)

Opacity is not a disease but a phenomenon


It is useful to think pretheoretically and abstractly: what are the data patterns we seeks to describe? We should think of opacity as patterns of alternation. Saltation. (Note : a < b indicates a minimal change from a to b.)

Example: (underlying, surface) (apa, aBa) (aba, aba) ...


Schema: apa

<

aba

<
3

aBa

Abstract Schema: A

<

<
3

(51)

Counterfeeding opacity:

Example: (underlying,surface) (katab,kitab) (gabr,gabur) ...

Schema: [u] epenthesis


/

katab gabr gabur

[a] raising
/

kitab

Abstract Schema: A
/

B
\

7/

12

Weeks 5-6: Optimality Theory Approaches to Opacity in OT (52) Counterbleeding opacity: horse club mansion cat ice cf. pl trupi cf. pl. klubi cf. pl dvori cf. pl koti cf. pl. lodi

J. Heinz

Example: (underlying, surface) (trup, trup) (klub, klup) ... (dvor, dvur) (kot, kot) ... (lod, lut) ...

Schema: dvor lod LL

raising

dvur klub

LLL  devoicing LLL LLL klup raising,devoicing LLL LL%

lut

kot R

Abstract Schema: A x / B when A @@ @@


@@ @ x,y @@  
y

would be ne.

C R

@@y @@ @

(53)

Thinking about opacity as a form of alternation should helpthis directly engages the phenomena, and may help avoid theories that lead to bizarre predictions. a. We need a theory like the one OT developed for phonotactics (i.e. constraints on surface forms) but instead which constrains alternations b. Schematic:

13

Weeks 5-6: Optimality Theory Approaches to Opacity in OT

J. Heinz

c. d. (54)

a Set a Relation a (a,b) b (c,d) c (e,f) d (g,h) ... ... Phonotactics: What are the constraints on phonologically possible sets (i.e. sets of well-formed words)? Alternations: What are the constraints on phonologically possible relations (i.e. sets of well-formed alternations)?

Like Bruce Hayes, Im also taken with the idea of a bias toward minimal alternation because this has explanatory valueagain, it suggests a way to constrain alternation.

References
Brame, Michael. 1974. The cycle in phonology: stress in Palestinian, Maltese and Spanish. Linguistic Inquiry 5:3960.

14