A FORMAL REPLACEMENT FOR REDUPLICATIVE ...rkennedy.faculty.linguistics.ucsb.edu/papers/kennedy...2 The following data in (1) illustrate Marantz’s Generalization. In the first three

1

A FORMAL REPLACEMENT FOR REDUPLICATIVE ANCHORING

Robert Kennedy, UCSB

1. Introduction

This squib offers a proposal for restricting the typological predictions that the Optimality-

Theoretic model of phonology (Prince & Smolensky 1993, McCarthy & Prince 1993, 1995)

makes for reduplicative morphology and phonology.

A formal theory of reduplication should have two basic goals: to account for the attested

typology of reduplicative systems, and to preclude the existence of impossible systems, where

evidence of the impossibility of a system is inferred from its absence from attested examples.

In light of the extremely robust phenomenon known alternately as Marantz’s

Generalization or Edge-in Association, I argue that Optimality Theory’s use of ANCHOR

constraints to capture the phenomenon falls short of both goals. ANCHOR constraints are both too

weak to capture some attested reduplicative patterns and too strong to preclude the existence of

some unattested systems.

I introduce a formal replacement of ANCHOR-correspondent constraints that resolves the

shortcomings of formal ANCHOR. I also provide an argument in favor of this proposal over

similar proposals made by Nelson (2003) and Hogoboom (2004).

2. Marantz’s generalization

A well-known observation in the study of reduplicative phonology is that wherever

reduplication is partial, prefixes tend overwhelmingly to reduplicate the left edges of stems,

while suffixes tend to reduplicate the right edges of words. This observation has come to be

known as Marantz’s generalization, after Marantz (1982).

2

The following data in (1) illustrate Marantz’s Generalization. In the first three languages,

the reduplicated sequence (underlined and boldfaced) is a prefix, and its segmental content is a

copy of the left side of the stem. In other words, the reduplicant is partial, and fails to maximize

elements from the right end of the stem.

(1) Prefixing reduplication

Agta (Healey 1960:7)

bari ‘body’ bar-bari-k kid-in ‘my whole body’

mag-saddu ‘to leak’ mag-sad-saddu ‘leak in many places’

na-wakay ‘lost’ na-wak-wakay ‘many things lost’

takki ‘leg’ tak-takki ‘legs’

uffu ‘thigh’ uf-uffu ‘thighs’

ulu ‘head’ ul-ulu ‘heads’

Tagalog (Carrier 1979)

la:kad ‘walk’ pag-la-la:kad ‘walking’

kandi:lah ‘candle’ pag-ka-kandi:lah ‘candle vendor’

linis ‘clean’ mag-li:-linis ‘will clean’

t-um-akboh ‘run’ t-um-a:-takboh ‘will run’

Yaqui (Molina et al. 1999)

vusa ‘awaken’ vu.vusa ‘awaken’ (habitual)

vamse ‘hurry’ vam.vamse ‘hurry’ (habitual)

patta ‘cover’ pat.patta ‘cover’ (habitual)

chepta ‘jump over’ chep.chepta ‘jump over’ (habitual)

3

In the following three examples in (2), the reduplicant is a suffix that follows the stem.

In each case, the reduplicated content is a copy of the right side of the stem, and leaves the left

side of the stem uncopied.

(2) Suffixing reduplication

Woleaian (Sohn 1976)

faŋošo ‘current’ faŋošo-ŋošo ‘to have a little current’

fitiye-li ‘marry him’ fitiye-tiye ‘to marry’

masowe ‘hard’ masowe-sowe ‘to be strong’

perase ‘to splash’ perase-rase ‘to scatter’

tafiši ‘to trap’ tafiši-fiši ‘to sparkle’

fati ‘corner’ fati-feti ‘to be angular’

lape ‘big, great’ lape-lape ‘greater’

misi ‘fool’ misi-misi ‘tell lies’

Lakota (Shaw 1976)

iá iá-a ‘speak’

thaí t

haí-i ‘show’

ečhú eč

hú-č

hú ‘do work’

u spé uspé-spe ‘to learn’

kce kce-kče ‘to fart’

wačhí wač

hí-c

hi ‘to dance’

čoká čoká-ka ‘empty’

há ska há ska-ska ‘be tall’

yamní yamní-mni ‘three’

4

škokpá škokpá-kpa ‘hollowed out’

gmigmá gmigmá-gma ‘spherical’

špahá špahá - ‘broken off’

xloá xloá - ‘slovenly’

Malagasy (Keenan and Polinsky 1998)

lèhibé lèhibè-bé ‘big’

vovó vovò-vó ‘barking’

hadíno hadìno-díno ‘forget’

àlahélo àlahèlo-hélo ‘sadness’

2.1 Copy, Associate, Stray Erasure

In derivational morpho-phonology (e.g., Marantz 1982, McCarthy & Prince 1986),

reduplication is modeled as a sequence of operations: Copy, Associate, and Stray Erasure.

Reduplication first creates a Copy of the stem phonemic melody, either before the stem for a

prefix or after the stem for a suffix. The copied melody then Associates to a reduplicative

template in a “phoneme-driven” manner, scanning the phoneme melody and finding an

appropriate slot for each phoneme in turn. Phonemes for which there is no room in the template

go unassociated, and are discarded through Stray Erasure.

Marantz’s generalization is attributed to the principle of “Edge-in” association. Since

prefixes appear to the left of the stem, association of the copied melody proceeds left-to-right.

Likewise, since suffixes follow the stem, association of the copied melody proceeds right-to-left.

McCarthy & Prince (1986) replace Marantz’s CV skeleta with prosodic units; otherwise, the

intuition remains the same. Association proceeds edge-in, and is phoneme-driven. I illustrate

the sequence with a Tagalog example in (3).

5

(3) pag-la-la:kad C V + l a: k a d

Copy C V

l a : k a d + l a: k a d

Associate C V

l a : k a d + l a: k a d

Stray Erasure C V

l a : k a d + l a: k a d

The Edge-in principle prevents Association from initiating inside the copied melody, at

the k perhaps, which would predict *ka-la:kad. It also prevents Association from initiating at the

end of the copied melody, which would also predict *ka-la:kad.

2.2 Marked association

Marantz acknowledges a small number of marked cases which seem to contradict Edge-

in Association, either where prefixing association is right-to-left or suffixing association is left-

to-right. In other words, it appears superficially in these cases that association proceeds

“juncture-out”. I provide examples in (4) below.

(4) Chukchee (Krause 1980)

jilʔe- gopher jilʔe-jil abs. sg.

nute- earth, ground nute-nut abs. sg.

inu- part of reindeer leg inu-un abs. sg.

Tzeltal (Berlin 1963)

6

-peč ‘to flatten it’ -peč-pu ‘to flatten it out’

-b’ah ‘strike with hammer’ -b’ah-b’u ‘strike repeatedly’

-t’aš ‘strike with open hand’ -t’aš-t’u ‘strike repeatedly’

-t’os ‘snap fingers’ -t’os-t’u ‘snap fingers repeatedly’

-čih ‘shake it’ -čih-čin ‘shake it vigorously’

-p’eh ‘unique constituent’ -p’eh-p’en ‘cut it vig. into pieces’

-c’al ‘make it ready for carrying’ -c’al-c’ n ‘continue carrying’

-t’oh ‘peck it’ -t’oh-t’on ‘peck it vigorously’

-wuk’ ‘break it into pieces’ -wuhk’-wun ‘chew up hard foods’

-t’im ‘pluck guitar strings’ š-t’im-t’u ‘sounds of guitar strings’

-čep ‘hoist cargo’ š-čep-ču ‘movement of cargo’

-t’an ‘take off clothes’ š-t’an-t’u ‘walk removing clothes’

-poč ‘unique constituent’ š-poč-pu ‘sounds of wings’

-wul ‘unique constituent’ š-wul-wu ‘rapidly talking’

Madurese (Stevens 1968: 34)

búwáq-an ‘fruit’ wáq-búwáq-an ‘fruits’

neat ‘intention’ yat-neyat ‘intentions’

sɔɔn ‘request (noun)’ ɔn-sɔɔn ‘requests’

Tillamook (Reichard 1959)

tq ‘break’ dæš q-t q- n ‘they tried to break it away’

tł ‘tell’ da s-ł-túł- n ‘they went and told him’

dak’ ‘lie’ niš-k-duk’ ‘they put her in their canoe’

ga·ł ‘eye’ an-s-ł-ga·ł ‘my eye ‘

7

cqil ‘climb’ q-ʑúq-il ‘they climb’

łaqil ‘sit’ nš-q-ł q-il ‘he is sitting in it’

æš ‘hold’ š-áš-un ‘he is holding it’

Though it may seem stipulative to leave these as “marked” cases in which association is

backwards, a solution for each exists which preserves the edge-in character of association, as

other processes interact to obscure this generalization. McCarthy & Prince (1986) and Nelson

(2003) offer means of reconciling such cases with Marantz’s generalization.

Note that the observable description need not characterize these as juncture-out

association: for example, not only does the Chukchee suffix begin with a copy of the first

consonant of the stem, but it also ends with a copy of the second-last consonant of the stem.

Viewed as such, the process remains edge-in, right-to-left association. The same is true of the

Tzeltal cases. Likewise, though it is true that the Madurese and Tillamook prefixes begin with

the final consonant of the stem, they also begin with a copy of the pen-initial consonant. In

every case, the outermost segment of the reduplicant is a copy of the next available segment

inward from the edge. Consequently, these four cases are not compelling counterexamples to

Marantz’s generalization.

Such cases are handled with blocking mechanisms such as the Obligatory Contour

Principle. This prevents the first consonant from being copied, forcing the template to be

satisfied with the next available consonant. I illustrate with an example from Tillamook:

(5) dæš q-t q n

Input C + t q

Copy C

t ә q + t q

8

Associate C

t q + t q

blocked by OCP

Associate C

t q + t q

Stray Erasure C

t q + t q

Similar solutions are possible for Chukchee, Tzeltal, and Madurese. As result, it is

reasonable to maintain Marantz’s Generalization as an exceptionless phenomenon, insofar as

reduplicative association can always proceed left-to-right for prefixes and right-to-left for

suffixes.

3. Edge-in association in OT: Anchor constraints

The non-serial model of Optimality Theory (Prince & Smolensky 1993, McCarthy &

Prince 1993) relinquishes the steps of Copy, Associate, and Erase; instead, the output of a

reduplicated stem is one that best satisfies a ranked set of Markedness (structural) and

Faithfulness (structure-preserving) constraints. Standard OT captures Edge-in association with

ANCHOR constraints, which require the edges of phoneme substrings to be in correspondence.

(6) ANCHOR-LEFT The left edge of the base and the left edge of the reduplicant are in

correspondence

ANCHOR-RIGHT The Right edge of the base and the Right edge of the reduplicant

are in correspondence

The phenomenon of Edge-in association is replaced with the use of ANCHOR-LEFT for

prefixing systems and ANCHOR-RIGHT for suffixing systems. I illustrate with a Tagalog example

9

below, using an undominated ranking of the Size-Restriction constraint defined briefly in (7); I

leave its exact formalization undefined as it does not bear on the argument.

(7) SIZE-RESTRICTION: The reduplicant is a syllable.

Given the choice between two candidates that satisfy Size-Restriction, the optimum is the

one whose prefix is properly left-anchored. In other words, la-la:kad is optimal because the first

segment of the prefix is in a correspondence relationship with the first segment of the base. In

contrast, the same cannot be said of the first segment in *ka-la:kad. Tableau (8) offers a

summary.

(8) Anchoring for Edge-in association: Tagalog /Red + la:kad/ la-la:kad

SIZE-RESTRICTION ANCHOR-LEFT ANCHOR-RIGHT

a. la-la:kad *

b. ka-la:kad *!

c. la:kad-la:kad *!

3.1 ANCHOR and overprediction

ANCHOR constraints make an over-prediction of possible reduplicative systems. Given

the premise of constraint universality, it is formally possible that a language may require prefixes

to be right-anchored, or suffixes to be left-anchored, contrary to Marantz’s exceptionless

generalization. I refer to this as “opposite edge anchoring”.

ANCHOR formally allows languages in which opposite-edge anchoring occurs over

exceedingly long stems. However, no marked case has an edgemost reduplicative segment that

copies from beyond the second available position. OT’s Anchoring theory thus leaves the

relationship between affix position and directional association as an accident. I offer an example

10

of such a system below. In formal terms, the reduplicant is required to be a prefix, as demanded

by ALIGN-RED-LEFT:

(9) ALIGN-RED-LEFT: the reduplicant is a prefix

However, the ranking of ANCHOR-RIGHT over ANCHOR-LEFT forces a situation in which

the prefix is anchored to the end of the stem and not the beginning. In other words, given a

choice between a left-anchored prefix and a right-anchored prefix, the system chooses the second

option because of the higher priority of ANCHOR-RIGHT. Tableau (10) illustrates this.

(10) Overprediction: opposite-edge anchoring in OT

Unattested language: /Red + tulasep/ sep-tulasep

ALIGN-RED-LEFT ANCHOR-RIGHT ANCHOR-LEFT

a. tul-tulasep *

b. sep-tulasep *

c. tulasep-sep *

Although such long-distance opposite-edge examples are not computationally difficult to

imagine, they are not known to exist. The fact that ANCHOR predicts their existence suggests that

a theory that includes it is too powerful.

3.2. ANCHOR and underprediction

In several ways, ANCHOR constraints are also too predictively weak. First, they cannot

handle what I refer to as “gradient ANCHOR effects”. Second, they have trouble with systems

that use reduplicative prefixes and suffixes as distinct morphemes.

3.2.1 Gradient anchor effects

11

The kinds of gradient ANCHOR effects that are problematic for formal anchoring are seen

in reduplicative systems in which Edge-in association is clearly observable, but edges of

substrings are not in correspondence. The marked cases of Section 2.2 fit this description, as do

the Chuukese and West Tarangan patterns exemplified in (11) below.

(11) Chuukese (Goodenough & Sugita 1980)

kin ‘separated’ kini-kin ‘partition’

m n ‘blow’ m nɨ-mәn ‘severe storm’

ŋon ‘behold’ ŋono-ŋon ‘ogling’

nuk ‘haul on line’ nuku-nuk distributive

p k ‘chopped’ p ku-pәk ‘chop’

pwuc ‘crazy’ p

wuco-p

wuc distributive

roŋ ‘hear’ roŋo-roŋ distributive

kurupw ‘joint’ kurup

wu-rup

w ‘full of nodes’

seniŋ ‘earlobe’ seniŋe-niŋ ‘hear what one wants to hear’

Rebi West Tarangan (Spaelti 1997)

tapúran ta-r-púran middle

payláwa-na pay-law-láwana friendly-3s

bit m-na bi-m-t mna small-3s

Doka Timur West Tarangan (Spaelti 1997)

kar p ka-p-r p many

takúr ta-r-kúr coconut shell -> sago/coconut mix

gasíra ga-r-síra old

12

In the Chuukese case, a pattern of word-final vowel deletion holds of all stems, whether

reduplicated or not. Thus /seniŋe/ is realized as [seniŋ] when unsuffixed, but under

reduplication, the stem-final vowel remains, and what would be its reduplicative correspondent

is not realized. I attribute final-vowel deletion to a constraint FREE-VOWEL; the pattern is related

to the placement of primary stress (Rehg 1991).

The problem of ANCHOR here is that it cannot choose between two poorly-anchored

candidates. For example, in both seniŋe-niŋ and *seniŋe-sen, the rightmost segment of the

suffix is not in correspondence with the rightmost segment of the stem, so both forms violate

ANCHOR-RIGHT. Consequently, ANCHOR-LEFT, ranked lower, ends up incorrectly cancelling the

optimal form. This is summarized in Tableau (12) below.

(12) Underprediction: mis-anchoring

Chuukese /seniŋe + Red/ seniŋe-niŋ

FREE-VOWEL ANCHOR-RIGHT ANCHOR-LEFT

a. seniŋe-ŋe *! *

b. () seniŋe-niŋ * *!

c. *seniŋe-sen *

The formal definition of ANCHOR-RIGHT makes no allowance for the second or third best

base correspondent. Wherever the rightmost base segment is not maximized in the reduplicant,

ANCHOR does not prefer maximization of the second-last segment (as in seniŋe-niŋ) over

maximization of any other preceding segment (as in *seniŋe-sen).

The West Tarangan forms show a similar problem, despite the intricately unique pattern

of infixation involved. In nearly every form, the reduplicant is a copy of the second consonant of

13

the final foot, and appears as a coda of the syllable that precedes this foot. Spaelti (1997)

attributes the minimal size of the reduplicant to an overriding effect of ALL-SYLLABLE-RIGHT,

which prefers reduplicated forms with fewer syllables – thus, the ideal reduplicant is one that

does not introduce an additional syllable. Spaelti attributes forms that copy more than a single

consonant, such as pay-law-láwana, to the phonotactics of the syllable that precedes the main

stress foot. A single consonant in such a case would introduce an illicit glide sequence, as in

*pay-w-láwana

Otherwise, the reduplicant is a single consonant, but is never a copy of the initial

consonant of the head foot, which we can attribute to *GEMINATE. Thus, except where the final

foot is a single syllable, the reduplicated consonant does not correspond to either edge of the

final foot. For example, in ta-r-púran, the reduplicant is poorly anchored at both edges. No

ranking of ANCHOR-LEFT and ANCHOR-RIGHT can choose this form over the well-anchored

competitor *ta-n-púran, as I show in Tableau (13) below.

(13) Underprediction: mis-anchoring

/ tapúran + Red/ ta-r-púran

ALL-SYLL-

RIGHT

*GEMINATE

ANCHOR-

LEFT

ANCHOR-

RIGHT

a. ta-pur-púran ***! *

b. ta-p-púran ** *! *

c. () ta-r-púran ** * *

d. ta-n-púran ** *

3.2.2 Languages with multiple reduplicants

14

A second underprediction of ANCHOR constraints is seen in languages that use both

prefixing and suffixing reduplication, notably where the two processes signify different functions

and any stem could take either affix. An example of this is observable in Woleaian (Sohn 1976),

in which prefixing reduplication derives progressive aspect, but suffixing reduplication derives

intransitive or stative verbs. I provide several examples in (14) below; note an incidental

alternation of with k and r with č.

(14) Woleaian

erae kek-kerae erae-rae ‘crawl’

kepate kek-kepate kepate-pate ‘word, language’

metafe mem-metafe metafe-tafe ‘become clear’

raŋe čeč-čaŋe raŋe-r ŋe ‘apply yellow powder’

The appeal to ANCHOR-RIGHT and ANCHOR-LEFT in this case results in a ranking

paradox: the ranking that predicts the proper prefix then precludes proper suffix. For example, to

predict the segmental content of the prefix, ANCHOR-LEFT must outrank ANCHOR-RIGHT, as

Tableau (15) shows.

(15) Woleaian prefix ranking for /Red + metafe/ mem-metafe

ANCHOR-LEFT ANCHOR-RIGHT

a. mem-metafe *

b. tafe-metafe *!

However, this ranking cannot hold for the suffix pattern. The higher rank of ANCHOR-

LEFT incorrectly predicts opposite-edge anchoring for the suffix; as a result, the ranking is

paradoxical. Tableau (16) illustrates this.

15

(16) Woleaian paradox for /metafe + Red/ metafe-tafe

ANCHOR-LEFT ANCHOR-RIGHT

a. () metafe-meta *

b. metafe-tafe *!

4. Alternative: PROXIMITY

All fundamental problems with ANCHOR constraints are attributable to the formal use of

Edge values <Left> and <Right> in their definition. As long as we allow direct reference to

these values, we cannot rule out the appearance of rightmostness and leftmostness in a formal

sense from a prefixing system. Furthermore, as long as we refer to specific Edges, we are

powerless wherever edgemost segments are not in correspondence.

I propose an alternative formalization of ANCHOR that retains the notion of strings in

correspondence, but requires no formal specification of the Edge values <Left> and <Right>, and

in fact requires no reference to the notion of Edge at all. I refer to this revised ANCHOR as

PROXIMITY. Defined in (17) below, PROXIMITY simply requires that segments in correspondence

be close to each other.

(17) PROXIMITY: no material intervenes between segments in correspondence.

This formalization is sensitive to the amount of structure that intervenes between two

segments in a correspondence relationship. Thus, it prefers the candidate t1-x-t1ulasep to the

candidate t1-xx-t1ulasep, since the two correspondent ts in the former are separated by a single

element and not by two.

Note that PROXIMITY measures distances between every pair of correspondents. Thus, a

form like tu-tulasep incurs two violations of PROXIMITY: one because of the segment that

16

intervenes between both instances of t and one because of the segment that intervenes between

both instances of u. Likewise, tul-tulasep incurs six: two violations each for the instances of t, u,

and l. Longer reduplicants therefore incur more violations.

More crucially, misanchored reduplicants also incur more violations. For example, the

candidate la-tulasep incurs six violations: three each for the pairs of l and a. Likewise, sep-

tulasep incurs fifteen violations: five for each of the prefix’s three segments. The effect of this

second property of PROXIMITY is that it avoids the over-predictive problem of opposite-edge

anchoring, and instead allows a formal interpretation that adequately captures gradient anchoring

effects.

Let us first look at PROXIMITY’s treatment of opposite-edge candidates such as those seen

in Section 3. A crucial difference between PROXIMITY and ANCHOR is that while ANCHOR

requires correspondence of particular edge-oriented segments, PROXIMITY simply measures

distance between correspondents. As a result, we can keep an overt requirement for the position

of the reduplicant as prefix or suffix, without risking opposite-edge anchoring. Tableau (18)

shows how opposite-edge anchoring violates PROXIMITY more than same-edge anchoring does.

(18) No opposite-edge anchoring

Hypothetical language: /Red + tulasep/ tul-tulasep; sep an impossible prefix

ALIGN-RED-LEFT PROXIMITY

a. tul-tulasep 2

b. sep-tulasep 6!

c. tulasep-sep * 2

17

A second benefit of PROXIMITY is that its measurement of distance allows it to capture

gradient anchor effects. As such, it can properly differentiate between opposite-edge affixes and

nearly well-anchored affixes, such as the Chuukese and West Tarangan examples seen in Section

3.1.1. In Chuukese, the effect of FREE-VOWEL is to disrupt right-edge anchoring, but PROXIMITY

then prefers a suffix that copies from closer to the right edge over one that draws from the left

edge. Tableaux (19) and offers a summary.

(19) Chuukese /seniŋe + Red/ seniŋe-niŋ

FREE-VOWEL PROXIMITY

a. seniŋe-ŋe *!

b. seniŋe-niŋ 3

c. seniŋe-sen 5!

Likewise, in West Tarangan, *GEMINATE prevents a left-anchored infix, but Proximity

then prefers the candidate that copies the next available consonant over one that copies the

rightmost one. I show this in Tableau (20) for ta-r-púran.

(20) West Tarangan / tapúran + Red/ ta-r-púran

ALL-SYLL-RIGHT *GEMINATE PROXIMITY

a. ta-pur-púran ***!

b. ta-p-púran ** *!

c. ta-r-púran ** **

d. ta-n-púran ** ****!

18

Third, PROXIMITY successfully fits into Woleaian-like systems that use both prefixes and

suffixes, since it makes no explicit reference to left and right edges. Wherever the reduplicant is

a prefix, PROXIMITY prefers correspondence to the left edge of the stem, and wherever the

reduplicant is a suffix, PROXIMITY prefers correspondence to the right edge of the stem. In other

words, the single constraint PROXIMITY replaces the work of conflicting ANCHOR constraints, so

the ranking paradox of ANCHOR-LEFT and ANCHOR-RIGHT is therefore avoided. Tableaux (21)

and (22) illustrate.

(21) Woleaian /Prefix + metafe/ mem-metafe

PROXIMITY

a. mem-metafe 2

b. tafe-metafe 5!

(22) Woleaian / metafe + suffix/ metafe-tafe

PROXIMITY

a. metafe-meta 5!

b. metafe-tafe 2

5. Conclusion

The current proposal replaces the work of reduplicative Anchoring constraints in

Optimality-Theoretic model of reduplication. As a result, it resolves three theoretical problems

with the use of formal ANCHOR constraints. First, it eliminates the typological prediction of

opposite edge anchoring, because its lack of reference to explicit edges precludes the appearance

of right-edge anchoring in left-side affixation, and vice versa. Second, it adequately captures

19

gradient anchoring effects, since its formalization measures the distance between two segments

in correspondence, not simply whether two segments are in correspondence. Third, it

successfully handles languages with both prefixing and suffixing reduplication.

Nelson (2003) makes a similar proposal to address some of the same problems, but rather

than rephrase ANCHOR, she argues for the removal of ALIGN-RED from the set of constraints.

The placement of the reduplicant is instead achieved indirectly through the use of Locality, as

defined in (23).

(23) LOCALITY The reduplicant and copied portion of the base are adjacent

A constraint set that includes LOCALITY (and not ALIGN-RED) precludes opposite-edge

Anchoring from being a possible system. Opposite-edge anchoring always violates LOCALITY,

so the optimal affix is one that is both local and well-anchored. Thus, the position of the

reduplicant is necessarily a function of Anchoring and LOCALITY. Regardless of the rank of

LOCALITY, right-anchoring systems will always prefer suffixes, and left-anchoring systems will

always prefer prefixes. Tableau (24) illustrates this effect for a right-anchoring system.

(24) No opposite-edge anchoring

Hypothetical language: /Red + tulasep/ tul-tulasep; sep an impossible prefix

ANCHOR-RIGHT ANCHOR-LEFT LOCALITY

a. tul-tulasep *

b. sep-tulasep * *!

c. tulasep-sep *

LOCALITY and PROXIMITY are similar in that they prefer strings in correspondence to be

close to one another. However, LOCALITY requires direct adjacency of reduplicant and copied

20

strings, and without further enrichment, does not capture nearly well-anchored affixes like the

Chuukese suffix.

More critical for LOCALITY is that it requires an additional labeling function for the

evaluation of candidates, to be able to discern whether particular substrings of candidates are

copied. For example, in Tableau (24), LOCALITY needs to know that in the candidate *sep-

tulasep, the Base substring –tula- is not copied, whereas the substring –sep is. In other words,

beyond detecting adjacency of Base and reduplicant, LOCALITY also needs to know whether each

segment of the Base is maximized.

In comparing PROXIMITY and LOCALITY, then, we can note the following: both avoid the

undesired prediction of opposite-edge anchoring. It is less clear whether LOCALITY can handle

languages that use both prefixing and suffixing reduplication as PROXIMITY does. Only

PROXIMITY clearly distinguishes degrees of mis-anchoring, since it measures distance between

correspondent elements. Finally, PROXIMITY’S means of evaluating violations is more direct, as

it is sensitive only to whether segments are in correspondence, not additionally to whether

particular segments lack correspondents in the same surface string. Because of these advantages,

as well as the obviation of those issues specific to ANCHOR, PROXIMITY is a better formal

approximation of Marantz’s generalization than the alternatives.

References

Berlin, B. 1963. Some semantic features of reduplication in Tzeltal. International Journal of

American Linguistics 29, 211-218.

Goodenough, W., & H. Sugita. 1980. Trukese-English dictionary.

21

Keenan, E. and M. Polinsky. 1998. Malagasy (Austronesian). In A. Spencer and A. Zwicky

(Eds.), Handbook of Morphology, 563-623. Malden MA: Blackwell Publishers.

Krause, S.R. 1980. Topics in Chukchee Phonology and Morphology. Doctoral dissertation,

University of Illinois, Urbana-Champagne IL.

Marantz, A. 1982. Re reduplication. Linguistic Inquiry 13.

McCarthy, J., and A. Prince. 1986. Prosodic Morphology.

Molina, F., H. Valenzuela, and D. Shaul. 1999. English-Yoeme Yoeme-English Dictionary.

New York: Hippocrene Books.

Nelson, Nicole. 2003. Asymmetric Anchoring. Doctoral dissertation, Rutgers University.

Reichard, G.A. 1959. A comparison of five Salishan languages. International Journal of

American Linguistics 25, 239-253.

Shaw, P. 1976. Dakota phonology and morphology. Doctoral dissertation, University of

Toronto.

Sohn, H-M. 1976. Woleaian reference grammar. Honolulu: University of Hawaii Press.

Spaelti, P. 1997. Dimensions of Variation in Multi-Pattern Reduplication. Doctoral

Dissertation, University of California-Santa Cruz.

Stevens, A.M. 1968. Madurese phonology and morphology. American Oriental Series vol. 52,

American Oriental Society, New Haven CT.