allomorphy in korean noun particles:acms.bufs.ac.kr/yslee/research/papers/53allomorhy... ·...

65 Korean Linguistics, Volume 14, 65-89. © 2008 International Circle of Korean Linguistics.

ALLOMORPHY IN KOREAN NOUN PARTICLES:a

Multiple Inputs and Default Allomorph*

Yongsung Lee

Pusan University of Foreign Studies

Abstract. The Multiple Input approach in Optimality Theory for Korean allomorph alternations has been challenged by recent research. It is shown not only to have failed in dealing with alternations after ŋ-final stems and with the lo/ilo and the wa/kwa alternations, but also to have resulted in a ranking paradox between different allomorph sets. These problems lead researchers to argue that Optimality Theory is incapable of dealing with the Korean allomorphy or to propose a separate selection device outside of the evaluation component to explain the allomorph alternations in Korean. This paper, however, shows that the apparent problems come from the partially optimizing nature of Korean noun allomorphy. It argues that parallel evaluation with multiple inputs has no problem in dealing with Korean allomorphy. A universal allomorphy constraint, DEFAULT

ALLOMORPH, which prefers a phonologically simple allomorph, is proposed to remedy the problems found in the previous analyses. With DEFAULT ALLOMORPH and other relevant constraints, we do not have to resort to separate selection/elimination stages or to language-particular and morpheme-specific constraints to explain Korean noun particle allomorph alternations.

Keywords: Optimality Theory, Korean particles, suppletive allomorph, subcategorization, default allomorph, multiple input, partial optimization

* The abbreviations in this paper are from the editor’s guide. Other abbreviations used in this paper are OT (Optimality Theory), MIH (Multiple Input Hypothesis), EVAL (Evaluation Component), and CON (Constraint Component).

66 Y. LEE

1. INTRODUCTION

Korean noun particles show interesting alternations between i/ka (NOM marker), in/nin (TOP marker), i l/li l (ACC marker), wa/kwa (CONJ marker), and lo/ilo (GOL/INS marker) among others. There have been many attempts both in rule-based and constraint-based frameworks to deal with these alternations. At one extreme end of the proposals, one may posit /ka/ as an underlying form for the nominative marker and try to derive /i/ from /ka/ by two deletion rules and one insertion rule as in Odden (1993: 134). Paster (2006: 68), however, claims that the lack of surface resemblance between two allomorphs, /i/ and /ka/ in this case, makes it very difficult to explain the naturalness of the rules involved. In her terms, those which do not have derivative relations are “suppletive allomorphs.”

In the Optimality Theory (=OT) framework, we find two different approaches. One is the Multiple Input Hypothesis (=MIH) as enshrined in Lapointe’s (1999) and Sung’s (2005) analyses. MIH allows all the suppletive allomorphs in the input and leaves the final choice to constraint interaction. Another is to posit a separate allomorph selection stage, in line of Russel (2006), either before EVAL as in Kim (2006) or after it as in Bye (2007). Sung (2005) and Kim (2006) show that there are apparently irreconcilable problems with MIH, which lead Kim (2006) to posit a pre-selection stage based on subcategorization constraints of each and every allomorph.

The separate stage of allomorph selection, however, severely weakens the basic tenet of OT, the parallelism, in the sense that we need mechanisms other than EVAL for selecting the optimal form. For example, Kim (2006) posits an allomorph selection stage before EVAL. The selection is based on the subcategorization restriction that each allomorph has, and only one allomorph is chosen to be the input in EVAL, which is, in principle, very similar to Lapointe’s (1999) stem selection. Another approach is the post-selection proposal as in Bye (2007) employing Morpholexical Control Component. In this approach, allomorphs do not compete with each other, they go through separate evaluation and the results of each evaluation are put to Control Component where only one of them survives. In this model, there is a priority relation between allomorphs so that when the more specific allomorph is chosen, the other will be rejected automatically. Here again, the Control Component crucially references subcategorization information as in Kim’s (2006) pre-selection model.

The basic logic underlying the pre-EVAL selection of the allomorph or the post-EVAL elimination process is that without such devices, OT cannot deal with the allomorph alternations properly. In this paper, we go back to see

ALLOMORPHY IN KOREAN NOUN PARTICLES 67

whether OT really fails to deal with Korean particle alternations without additional devices. We will check back with the conspicuous problems in the noun particle allomorphy in Korean to see whether they can somehow be resolved. If MIH and direct evaluation can explain the allomorph alternations, we may not need any separate component for the allomorph selection. It will be shown in this paper that the apparent problems in the previous analyses come from two sources; the reliance on the constraint ONSET and the partial-optimizing nature of the Korean allomorphy. As will be shown in the following discussion, all ŋ-final stems fail to select the correct allomorph, if the crucial constraint is ONSET. All of the previous analyses in OT frameworks, to the best of my knowledge, employ ONSET as one of the crucial constraints in explaining the allomorph selection.

Further problems are witnessed in the wa/kwa alternation and the lo/ilo alternation. As for the wa/kwa alternation, /kwa/ is used after a consonant and /wa/ after a vowel. As noted in Bye (2007: 77), “the Markedness-driven approach would lead us to expect -wa across the board.” The unexpected behavior of allomorphs poses serious problems in dealing with the alternation based on MIH. In the lo/ilo alternation, we find /lo/ not only after a vowel but also after a lateral. On the surface, the grouping of vowels and laterals seems to necessitate a subcategorization approach as in Kim (2006). All these problems combined together seem to argue against MIH and point to the need of separate components for the allomorph selection based on the subcategorization information.

To deal with these problems, this paper argues that there is a priority relation between allomorphs as noted in Bye (2007) or in McCarthy and Wolf (2005). The priority relation, however, is not specified as a lexical property of allomorphs as in Mascaró (2007) or McCarthy and Wolf (2005: 28). Instead, it will be shown that the priority is due to the default value of the allomorph (DEFAULT), which can be predicted based on the universal phonological simplicity. The default allomorph is the phonologically simpler one. In addition to that, it will be shown that we need the constraint, NOHIATUS, apart from ONSET. It will be shown that the constraint NOHIATUS, not ONSET, plays a crucial role in selecting the correct allomorph among multiple inputs.

DEFAULT and NOHIATUS which interact with relevant constraints present a consistent analysis for the Korean particle alternations. The logic is that if we could deal with the allomorphy without resorting to any other components such as pre-EVAL allomorph selection stages or post-EVAL elimination processes, it should be the preferred solution. And this paper shows that there is such a solution. In section 2, we will briefly survey the data and the analytic problems.

68 Y. LEE

Section 3 introduces DEFAULT and NOHIATUS and shows how they serve to present a simple and comprehensive explanation to the allomorph selection in the Korean particles. Section 4 wraps up the discussion and concludes the paper.

2. DATA AND PROBLEMS

One of the issues in the Korean allomorphy discussion is whether to treat the noun particle allomorphy and the verbal conjugation allomorphy all together or in a separate manner. Consider the following sample allomorphic variations found in Korean.

(1) Korean allomorphy alternation

a. Noun particle allomorphy b. Verb conjugation allomorphy

Nominative i/ka Conditional myən/imyən Topic in/nin Effective ni/ini Accusative il/l il Purposive lə/ilə Conjunctive wa/kwa Past s’/əs’ Goal/Inst. lo/ilo Perfective n/in

At a glance, we see a noticeable regularity in the verbal conjugation allomorphy. Allomorphs are very much similar to each other. The only difference is that the longer form has an extra vowel, /i/ or /ə/, in the suffix-initial position. These allomorph alternations in verbal conjugation may be dealt with by either of the two following options. First, based on the observation that the verbal conjugation allomorphy resembles the alternation of Goal/Instrument marker in the last row of (1a), we may extend the noun particle analysis to verbal conjugation. Or we may separate out the verbal conjugation and take a different approach. In this approach, verbal conjugation allomorphs are not viewed as suppletive. Then we may posit one form in the input and account for the emergence of the other form with constraint interaction as shown in Hong’s (2001) analysis.1

1 We leave open the question whether to treat the verbal conjugation separately from the noun parti-cle alternations or to treat them as an extension of the analysis presented for the noun particle allo-morphy in this paper. The superficial regularity of suffixes can be deceiving in the verbal conjuga-tion, since there are many irregular stem changes in the verbal conjugation. Consider the following variation of an irregular verb stem, /mut/ (to ask), in comparison with other regular types focusing on the stem vowel changes.


2.1. Data

In this paper, we will focus on the noun particle alternation. The noun particle allomorph alternation looks different from the verbal conjugation allomorphy. There is no natural way to derive one form from the other, perhaps except for Goal/Instrument marker, lo/ilo. Neither insertion analysis nor deletion analysis can account for the noun particle allomorphy in a consistent way. The rules in either of the approaches, posited to capture the phonological relation between allomorphs, may not have any independent motivation in Korean. They are truly “suppletive.” Let us first take a closer look at the noun particle allomorph alternations.

(2) Noun particle allomorphy

a. Nominative marker: /i/ after a C-final stem, /ka/ after a V-final stem Stem i-suffix ka-suffix Gloss sok so.gi *sok.k’a ‘inside-NOM’ waŋ waŋ.i *waŋ.ga ‘king-NOM’ mul mu.ri *mul.ga ‘water-NOM’ so *so.i so.ga ‘cow-NOM’

b. Topic marker: /in/ after a C-final stem, /nin/ after a V-final stem Stem in-suffix nin-suffix Gloss sok so.gin *sog.nin ‘ inside-TOP’ waŋ waŋ.in *waŋ.nin ‘king-TOP’ mul mu.rin *mul.nin ‘water-TOP’ so *so.in so.nin ‘cow-TOP’

mut (to bury)-regular mut (to ask)-irregular mul(bit)-regular -(i)ni mu.di.ni mu.ri.ni mu.ni

-ko mut.k’o mut.k’o mul.go -ni mun.ni mun.ni mu.ni

-(ə)s’ mu.dəs’ mu.rəs’ mu.rəs’ -(ə)sə mu.də.sə mu.rə.sə mu.rə.sə

As shown above, the irregular verb, mut (to ask), behaves like the regular form, mut (to bury), before a C-initial suffix, and it behaves like another regular verb, mul (to bite), before a V-initial suffix, where the vowel is not /i/. And before the i-initial suffix, it behaves differently from either of the regular forms. The choice of the stem allomorphy seems to be influenced by the suffix. Due to this complication of the irregular verbs, the uniform analysis of the noun particle and the verbal conjuga-tion allomorphy may not be easily obtainable. Further research in this area is in order.

70 Y. LEE

c. Accusative marker: /il/ after a C-final stem, /lil/ after a V-final stem Stem il-suffix lil-suffix Gloss sok so.gil *sog.lil ‘inside-ACC’ waŋ waŋ.il *waŋ.lil ‘ king-ACC’ mul mu.ril *mul.li l ‘water-ACC’ so *so.il so.lil ‘cow-ACC’

d. Conjunctive marker: /kwa/ after a C-final stem, /wa/ after a V-final stem Stem kwa-suffix wa-suffix Gloss sok sok.k’wa *so.gwa ‘inside-CONJ’ waŋ waŋ.gwa * waŋ.wa ‘king- CONJ’ mul mul.gwa *mul.wa ‘water- CONJ’ so *so.gwa so.wa ‘cow- CONJ’

e. Goal/Instrument marker: /ilo/after a nonliquid C-final stem and /lo/ after a V-final and a l-final stem

Stem ilo-suffix lo-suffix Gloss sok so.gi.ro *sog.lo ‘inside-GOL/INS’ waŋ waŋ.i.ro *waŋ.lo ‘king- GOL/INS’ mul *mu.ri.ro. mul.lo ‘water- GOL/INS’ so *so.i.ro so.ro ‘cow- GOL/INS)’

A few things must be cleared away to see a simpler picture. In Korean, plain stops, /p, t, k/, are voiced between two sonorant sounds, /l/ is realized as /r/ in between two vowels and plain stops are glottalized after another obstruent.2 These are due to high ranking sequential markedness constraints which penalize a plain stop between two sonorants, an intervocalic lateral, and a post-obstruent plain stop. We will not be concerned with these variations for the sake of simplicity of exposition.

Now, let us take a look at the data focusing on the distribution of the allomorphs in regard to their phonological simplicity. The data in (2a), (2b) and (2c) have one thing in common. The phonologically simple forms, /i/, /in/, and /il/ are taken after C-final stems, while more complex forms, /ka/, /nin/, and /lil/ are found after V-final stems. As for (2d) and (2e), we see the reverse situation. Simpler allomorphs are found after V-final stems, while longer forms are found after C-final stems. The distribution is generally governed by the nature of phonological composition of suffixes. In (2a), (2b) and (2c), the simpler forms begin with vowels, while in (2d) and (2e), the simpler forms have C-initial

2 The glottalized consonants are represented as [C’] in (2) and thereafter.


allomorphs. Goal markers in (2e) need some special comment. Note that the shorter form, /lo/ is used not only after vowels but also after liquids. In all the examples in (2), except those in (2e), we see that liquid final stems behave in the same way as other C-final stems in choosing the suffixal allomorphs. This inconsistent behavior of liquid-final stems was one of the problems in the previous analyses.

There has been much discussion in the literature on how to explain these alternations in the OT framework. The discussion starts with Multiple Input Hypothesis along the line of Lapointe (1999: 267):

(3) Multiple Input Hypothesis (Lapointe, 1999: 267)

The GEN function operates in such a way to contribute to the candidate set all prosodic parses based on all of the forms listed as possible phonological representations for a morpheme or a lexeme.

The basic idea behind MIH, and its further elaboration in Mascaró (2007: 718), is that all the suppletive allomorphs are present in the underlying representation. GEN produces all the possible combinations of stems and allomorphs, and they are put to EVAL, where the optimal form is chosen through the parallel evaluation. Though MIH presents the simplest solution to suppletive allomorphy, the research so far seems to point to the implausibility of such hypothesis. We will consider three conspicuous problems observed in the previous research.

2.2. Problems

At first sight, MIH can deal with the allomorph alternation in (2a), (2b) and (2c) in a simple and straightforward way. This is especially true when the surface allomorph satisfies most of the relevant markedness constraints. Consider the following tableaux slightly modified from Lapointe (1999) and Sung (2005):

(4) Evaluation of nominative allomorphy

a. mul (water) + NOM mul-{i, ka} ONSET *[ σCC NOCODA ALIGN-STEM

☞ i) mu.ri * ii) mul.i *! *

iii) mul.ga *! iv) mu.lga *! *

72 Y. LEE

b. so (cow) + NOM so-{i, ka} ONSET *[ σCC NOCODA ALIGN-STEM

i) so.i *! ☞ ii) so.ga

iii) sog.a *! * * (*[σCC = NOCOMPLEX/ONSET ALIGN-STEM = ALIGN (STEM, RIGHT, SYLLABLE , RIGHT)) All we have to do is to put both of the suppletive allomorphs in the input and leave the selection to the interaction of constraints. It seems to work perfectly. But it crucially fails to deal with ŋ-final stems.3 In Korean, /ŋ/ is not allowed in the onset position. Call it *ONSET/ŋ (=*[ σŋ: No [ŋ] in the onset). This constraint is undominated, since Korean does not allow any syllable-initial /ŋ/. Consider the following unfortunate evaluation.

(5) Wrong evaluation for an ŋ-final stem (king-NOM)

waŋ-{i, ka} *[ σŋ ONSET *[ σCC NOCODA ALIGN-STEM

� i) waŋ.i *! * ii) wa.ŋi *! *

?☞ iii) waŋ.ga * The actual surface form is (5i). But the evaluation wrongly chooses (5iii) as optimal. This is a classic case of “harmonic bounding,” as in McCarthy (2002a: 23). A candidate, A, is harmonically bounded by another candidate, B, if it contains all the violations made by B and at least one more. In this sense, (5i) is harmonically bounded by (5iii); (5i) violates all the constraints that (5iii) violates (NOCODA) and has an additional violation (ONSET). In this situation, (5i) cannot win over (5iii) under any ranking, unless there is a third constraint, which penalizes (5iii) but not (5i).

One possible way to explain the appearance of /i/ after ŋ-final stems is to make (5ii) optimal by lowering *[σŋ to allow /ŋ/ to come in the onset position. But such a move lacks any independent motivation, given the overall distribution of /ŋ/ in Korean. As such, the MIH apparently fails to explain the allomorph alternations in Korean noun particles.

3 Interestingly, the previous OT analyses, including Lapointe (1999), Sung (2005), Kim (2006) and Bonet et al. (2007), among others, have not noted the problems in ŋ-final stems in the allomorph selection.


Another problem comes from the conjunctive marker alternation. As noted earlier, the wa/kwa alternation differs from the i/ka alternation in that a more complex allomorph is found after a C-final stem. Lapointe (1999: 271-272) observes that “the problem here is that we would otherwise expect that the C-initial long form would attach to a V-final noun stem and the glide initial form should attach to a C-final form, but in fact what we find is just the opposite.” Lapointe’s (1999: 273) solution to the surprising deviation was to introduce a language-particular condition that allows stems to select an allomorph before entering into the evaluation. This is quite similar to Kim’s (2006) affix selection except that, for Lapointe, the selection is made by the stem, not by the affix.

Sung (2005) sees that the crucial constraint that motivates the wa/kwa alternation is SYLLCON as in Davis and Shin (1999: 270).

(6) SYLLCON: Avoid rising sonority over a syllable boundary.

The explanation goes as follows. As for V-final stems, the /wa/ form might be added since there is no SYLLCON involved here. But if the stem ends in a consonant, then adding /wa/ may cause a SYLLCON violation, since [w] is more sonorous than any coda consonant. This may be a right move, but there is an unexpected problem in the actual evaluation as shown in (7):

(7) wa/kwa alternation after C-final stems

a. after a V-final stem (cow-CONJ)

so-{wa, kwa} SYLLCON ONSET *[ σCC NO

CODA ALIGN- STEM

☞ i) so.wa ii) so.gwa *!

b. after a C-final stem (house-CONJ)

cip-{wa, kwa} SYLLCON ONSET *[ σCC NO CODA

ALIGN- STEM

i) cip.wa *! * ?☞ ii) ci.bwa * * � iii) cip.k’wa * *!

The constraints and their ranking work OK with V-final stems as shown in (7a). But they fail in (7b). Sung (2005) and Bonet et al. (2007) observe that one way to make (7biii) optimal is to place ALIGN-STEM anywhere over NOCODA, but the ranking, ⟦ALIGN-STEM » NOCODA⟧, conflicts with the evaluation given in (4a) as shown in Sung (2005). For the correct evaluation of (4a) and all the cases

74 Y. LEE

where the vowel-initial suffixes are chosen, we need the ranking, ⟦NOCODA »

ALIGN-STEM⟧. Sung (2005) highlights this ranking paradox and draws the conclusion that the wa/kwa alternation cannot be explained with MIH and that OT, in general, may have problems in dealing with allomorph alternations.

Finally we see a similar problem in the lo/ilo alternation. /lo/ is used after vowel-final stems AND liquid-final stems, and /ilo/ occurs elsewhere. One immediate problem here is how to group vowels and liquids together and separate them out from the rest of the consonants and how to translate that into a proper constraint. Without such implementations, which are highly unlikely, constraint interaction may fail in accounting for the alternation. Consider the following evaluation result.

(8) li/i lo alternation after a l-final stem (mul+{lo, ilo} (water-GOL/INS))

mul-{lo, ilo} ONSET *[ σCC NOCODA ALIGN-STEM � i) mul.lo *!

?☞ ii) mu.ri.ro * The ranking, ⟦NOCODA » ALIGN-STEM⟧, does not work for (8). The ranking permutation to save (8) causes problems as already discussed in connection with (7). We see the duplication of the ranking paradox that we observed in (7). The evaluation in (8) seems to prove the inadequacy of MIH and to support the need for a separate allomorph selection stage. Viewed from this perspective, MIH fails to present a comprehensive analysis for Korean noun particle allomorphy. So far, we have identified three problematic areas.

(9) Summary of problems

a. ŋ-final stems Since /ŋ/ cannot occur in the onset position, ŋ-final stems followed by V-initial affixes are harmonically bounded by those followed by C-initial suffixes.

b. wa/kwa alternation A more marked form, /kwa/, appears after C-final stems. SYLLCON does not help much, since it leads to a ranking conflict with the evaluation of the nominative, topic and accusative marker allomorphy.

c. lo/ilo alternation Liquid-final stems behave like vowel-final stems in that they choose /lo/. It is extremely difficult to come up with a constraint that can explain the alternation. Otherwise, we face the same ranking paradox as


shown in (9b). The previous analyses fail to deal with the choice of V-initial allomorphs after ŋ-final stems. The constraint ranking, posited for the nominative, accusative and topic marker alternations, causes problems in dealing with the conjunctive marker alternation. Further, the lo/ilo alternation after l-final stems duplicates the same ranking paradox. Having identified the problems, let us see how they are resolved in the pre-selection model.

2.3 Pre-evaluation Allomorph Selection

The inconsistent behavior of liquid-final stems and other problems noted in this paper lead Kim (2006: 279) to conclude that MIH cannot deal with the allomorph alternation, since it is “descriptively inadequate, making no satisfactory description of the data.” Instead, Kim (2006: 280) proposes the following steps for the allomorph selection.

(10) Three steps for suppletive allomorphy (Modified from Kim 2006: 280)4

a. Step 1: stem-final segment identification b. Step 2: suffix identification and allomorph selection c. Step 3: OT-based analysis: candidate evaluation

Note that there are two stages before the actual evaluation in her analysis. Stage 1 and Stage 2 are for the allomorph selection based on the subcategorization regulations specified in each and every suffixal allomorph. The relevant sub-categorizations of suffixes are given in (11):

(11) Exemplary subcategorization

a. Nominative5 Affix /i/ to [+consonantal] Affix /ka/ to [-consonantal]

b. Conjunctive Affix /wa/ to [-consonantal] Affix /kwa/ to [+consonantal]

4 Kim’s schema is simplified just for the OT analysis. Kim argues that even for a rule-based ap-proach the first two stages are needed for the correct result. Therefore, Stage 3 in Kim’s model in-cludes rule based derivation as another option for Stage 3. 5 The topic markers and the accusative markers also show the same pattern that the nominative markers have. The shorter forms (the vowel-initial allomorphs) appear only after stem-final conso-nants ([+consonantal]), while the more complicated forms are reserved for post-vowel positions ([-consonantal])

76 Y. LEE

c. Goal/ Instrument Affix /lo/ to [+vocalic] Affix /i lo/ to [-vocalic]

The key to the explanation lies in the distinction of [±vocalic] and [±consonantal]. Liquids are [+cons, +voc], other non-liquid consonants are [+cons, -voc], and finally vowels are both [-cons, +voc]. Liquids and vowels share [+vocalic] feature, while liquids and consonants are [+consonantal]. Once such distinction is motivated, we may simply say that /lo/ is subcategorized to take a [+vocalic] segment, a vowel or a liquid, as a base for affixation. The pre-selection device allows only one form in the input and the rest naturally falls out from the constraint interaction. The restriction imposed on each and every allomorph is part of the allomorph’s idiosyncratic nature.

Theoretically, the subcategorization and the pre-selection stage can be independent of each other. We can take subcategorization constraints, put them in the hierarchy and select the proper allomorph from multiple inputs. Consider the following simplified evaluation:

(12) Subcategorization constraint and multiple input

a. after an ŋ-final stem (king-NOM)

waŋ-{i, ka} AFFIX-i-

TO-[+CON] AFFIX-ka- TO-[-CON] *[ σŋ ONSET NOCODA

☞ i) waŋ.i * * ii) wa.ŋi *!

iii) waŋ.ga *! *

b. after a l-final stem (mul-GOL/INS)

mul-{lo, i lo} AFFIX-lo-

TO-[+VOC] AFFIX-ilo- TO-[-VOC]

*[ σŋ ONSET NOCODA

☞ i) mul.lo * ii) mu.ri.ro *!

As shown in (12), introducing subcategorization might actually obviate the need for a separate selection stage. Though Kim (2006) did not delve into this possibility, we may make a few observations that argue against this approach for the Korean noun allomorph selection. First, the allomorph selection, in general, improves the syllable structure in Korean. Notice that a V-initial suffix comes after a C-final stem, presumably to optimize the syllable by eliminating a coda and that a C-initial suffix comes after a V-final stem, this time to prevent vowel hiatus. The brute force of the subcategorization constraints fails to capture this generalization, leaving the choice to some morpho-phonological idiosyncrasy of


suffixes. Second, there is redundancy in the subcategorization constraints. We need both of the two subcategorization constraints, one for each allomorph. Suppose we do not have AFFIX-ka-TO-V in (12a), we see that (12aiii) may be wrongly chosen as optimal. Again eliminate AFFIX-ilo-TO-[-voc] in (12b), we see that the evaluation picks the non-surface form (12bii) as optimal, temporarily ignoring the role of ALIGN-STEM. To reduce the redundancy, we may eliminate subcategorization constraints for /i/ in (12a) and for /lo/ in (12b). But note that there is no a priori or principled way to select one subcategorization over the other.6 Finally, the proposed subcategorization constraints are language-particular and morpheme-specific in nature. Such an approach severely weakens one of the basic principles of OT: the universality of constraints. It is true that we cannot dispense with some language-specific constraints especially in morphology. But if there is a way to deal with them with general and universal constraints, such a solution is arguably preferred to the solution employing language-specific constraints.

3. ALTERNATIVE ANALYSIS WITH DEFAULT AND MIH

In this section, a universal constraint, DEFAULT, is proposed. This constraint regulates the default choice of allomorphs and it works together with MIH to present a simple and comprehensive analysis for the Korean noun particle allomorph alternations. If MIH can deal with the allomorph selection without additional devices, we may opt for such analysis on the basis of simplicity of explanation. As a matter of fact, we will see that the three most conspicuous

6 This is a peculiar situation. When subcategorization is at work, we usually see that there is an else-where relation between two allomorphs. One allomorph is used in a very specific environment (“only-when” effect) and the other is used elsewhere. Some specific examples are given below: a. In Dyirbal, the ergative allomorph /ŋgu/ is used only after a head foot and /gu/ is used elsewhere. (see McCarthy & Prince 2001:117) b. For perfective prefix in Kwamera, /uv-/form is used before /i/ and /u/ and /in-/ elsewhere. (Lind-strom and Lynch 1994: 12) c. In Udihe, the perfective is marked by a suffix /ge/ after a stem-final high vowel. Elsewhere it is realized by laryngealization of stem final vowels. (see Bye 2007:10-11) d. In Tahitian, the causative/factitive is marked by one of the two prefix allomorphs. /ha’a/ is used when the stem-initial segment is [+labial]. Otherwise /fa’a/ is selected. (see Paster 2006:11) The general observation here is that an allomorph which is subject to some specific subcategoriza-tion requirement is chosen only in the subcategorized environment and the general form wins else-where. But we cannot find the parallel elsewhere relation in the Korean allomorphy.

78 Y. LEE

problems noted in 2.2 are not real problems but simply the unfortunate results stemming from the improper choice of constraints.

Researchers have struggled to come up with a solution to explain the default choice of the allomorph. It so happens that when constraint interaction fails to select a proper allomorph, there is a certain tendency to choose the phonologically simple allomorph. Consider what Bye observes:

(13) Observation on default allomorph choice

Suppose that A is more complex than B in some sense, e.g. by being longer. Other things being equal, fB should be selected because fA will violate more structural well-formedness constraints. (Bye 2007: 70)

Bye illustrates his point with Djabugay genitive allomorphs: /n/ appears after V-final stems and /ŋun/ after C-final stems. Consider the following tableau to see his point:

(14) Choice in Djabugay genitive allomorphy (Bye 2007: 72)

guludu-{n, ŋun} *CXCODA MAX NOCODA *ŋ *n

☞ i) guludun * * ii) guludu ŋun * * *

There is no problem with C-final stems, since *CXCODA (NOCOMPLEX/CODA) eliminates the /n/ form. But after V-final stems as in (14), both /n/ and /ŋun/ can come, as they both satisfy *CXCODa. Then the final choice is in the hand of segmental markedness constraints such as *ŋ or *n which will penalize a longer form. Bye’s observation on the preference to simple forms may be correct, but his explanation with segmental markedness constraints may not be correct for the partially-optimizing allomorphy. If there is any other markedness constraint that dominates the segmental markedness constraints, *ŋ and *n, then it may lead to a different result. Korean allomorphy shows such partial-optimization effect as we will see later. We will take a more straightforward way to capture the observation in (13) and posit a universal constraint for allomorphy as in (15):

(15) DEFAULT ALLOMORPH (DEFAULT)

A phonologically simple allomorph is preferred.7

7 The need for a preferential relation among allomorphs was discussed in Mascaró (2007) and McCarthy and Wolf (2005) in the form of a constraint, PRIORITY (Respect the lexical ordering of


Phonological simplicity means shortness in the segmental length or unmarked nature in the phonological feature composition.8 In this sense, shorter forms are less marked. For example, in i/ka alternation (nominative particles), /i/ is the default choice. Unlike subcategorization constraints proposed in Kim (2006), DEFAULT is not a language-specific constraint. It should be applied across all the allomorph sets in a given language as well as across different languages. The constraint, DEFAULT, is for the choice of the default form out of multiple suppletive allomorphs. It does not interfere with the deletion from or the insertion into the derivative allomorphs. Further, DEFAULT must be crucially dominated by some constraint(s). Otherwise, it will ultimately have the effect of eliminating allomorphs all together, because a non-default allomorph, as it violates DEFAULT, has no chance to show up on the surface. The general ranking schema is in (16):

(16) Ranking schema for partially optimizing allomorphy

M1 » DEFAULT » M2 Suppose M1 and M2 are markedness constraints that govern the surface well-formedness of allomorphs. Some markedness (M1) dominates DEFAULT, therefore the surface form shows some markedness improvement as to these well-formedness conditions. But at the same time, some other markedness constraints (M2) are dominated by DEFAULT, leading to the violation of these markedness constraints to satisfy DEFAULT. This time, the surface form is not optimized with respect to M2. The result is the partial-optimization effect. Now

allomorphs). However, DEFAULT differs from PRIORITY in that no language-specific preference is involved. Though further research is in order, we see that the allomorphs with priority are generally the phonologically simple ones in Mascaró’s examples. Also note that the proposal made here is based on the comment by Kager (1996: 156, footnote 2) that we might posit “a universal constraint requiring that morphological categories are marked by minimal means (e.g. the ‘phonologically shortest’ morpheme).” 8 Note that the syllable structure markedness is not an issue here. In any case, syllable structures are not present in the underlying representation of an allomorph and there is no way to refer to the sylla-ble structure. For example, in i/ka alternation, /i/ is the default allomorph, even if the ka form is syllabically less marked. Further, in the analysis of Italian article lo/il allomorphy, Boyd (2006) employed *lo , meaning that il is the default. Note here that VC form is chosen over CV form in total disregard to the syllable structure markedness. Another interesting example comes from Bonet et. al.’s (2007) analysis of the Haitian Creole a/la alternation. Here again they gave the priority to /a/, which is syllabically more marked, but segmentally shorter than /la/. Special thanks go to one of the anonymous reviewers who brought this point to my attention.

80 Y. LEE

with such ranking schema, we see that the non-default form is chosen only when the default form violates high ranking M1 constraints AND when the non-default form satisfies them. Otherwise, the default form will surface. This exactly is the pattern that we observe in the Korean allomorph selection.9

3.1 Default Allomorph and NOHIATUS

Let us first re-examine the problem that involves ŋ-final stems. Remember that the ranking and the constraints for explaining the nominative allomorph alternation for C-final stems and V-final stems have caused an unexpected problem for ŋ-final stems in (5). The problem comes from the undominated nature of *[σŋ in Korean. This constraint eliminates any candidate that has [ŋ] in the onset. In (5), we see that [waŋ.i] and *[waŋ.ga] compete with each other. But since [waŋ.i] is harmonically bounded by *[waŋ.ga], it cannot be optimal. We need something else to break off the harmonic bounding to choose [waŋ.i] as optimal. ONSET penalizes both [Vŋ.V] and [V.V] without discrimination. One way to differentiate these two is to introduce an anti-hiatus constraint, which allows a Vŋ.V sequence and still reject a V.V sequence. The observation simply confirms that avoidance of hiatus cannot be reduced to the constraint ONSET as observed in Boroff (2003). Accordingly, we posit the following NOHIATUS constraints:

(17) NOHIATUS (=*VV): No two adjacent vowels across a syllable boundary NOHIATUS, not ONSET, is the responsible constraint in selecting the proper allomorph. NOHIATUS may be viewed as a specific form of ONSET. Given the specific-general relationship of ranking schema as expressed in the Panini theorem (Prince and Smolensky 2002: 89-90), ONSET may be anywhere lower than NOHIATUS in Korean. And we place DEFAULT between them to produce a partial ranking, ⟦NOHIATUS » DEFAULT » ONSET⟧. Now consider the following evaluation tableaux: 10

9 This does not mean that all the suppletive allomorph alternation can be explained with the ranking schema given in (16). We see the need for language-particular and morpheme-specific information or constraints to explain “only when” effect of allomorph alternations: A is used only in X. (Some examples are given in the footnote (6).) This is schematically different from “except when” distribu-tion of the default allomorph: B except in Y. The related issue is open to further research. 10 This paper does not make a claim that *VV is undominated in Korean. The tolerance of vowel clash in the underived words, [ma.im] (mind), [na.i] (age) and the derived forms like [i.i .ni] (as … link) shows that *VV is not inviolable. The dotted line between *[σŋ and *VV simply means that the ranking between them is irrelevant for the present discussion.


(18) Evaluation tableaux for Nominative marker allomorphy

a. after a V-final stem (cow-NOM)

so-{i, ka} *[ σŋ *VV D EFAULT ONSET NO

CODA ALIGN- STEM

i) so.i *! * ☞ ii) so.ga *

b. after an ŋ-final stem (king-NOM)

waŋ-{i, ka} *[ σŋ *VV D EFAULT ONSET NO CODA

ALIGN- STEM

☞ i) waŋ.i . * * ii) wa.ŋi *! * iii) waŋ.ga *! *

c. after a C-final stem (Inside-NOM)

sok-{i, ka} *[ σŋ *VV D EFAULT ONSET NO CODA

ALIGN- STEM

i) sog.i *! * ☞ ii) so.gi *

ii) sok.k’a *! * (18) shows the nominative marker allomorphy, but virtually the same explanation can be given to all five allomorph sets in regard to ŋ-final stems. There are two important observations to make in the tableaux. First, we cannot do away with DEFAULT. Consider the tableau (18b). Without DEFAULT, ONSET will favor (18biii), the one with an onset consonant. As can be seen here, DEFAULT has the effect of saving [waŋ.i] from being harmonically bounded by *[waŋ.ga]. The ranking, ⟦DEFAULT » ONSET⟧, leads to the violation of ONSET, to satisfy DEFAULT. As a result, the Korean allomorphy shows partial optimization, in the sense that some markedness constraints are not satisfied in the optimal form due to DEFAULT. Second, hiatus avoidance cannot be reduced to ONSET. Note that the candidate [waŋ.i] in (18bi) does not violate *VV (NOHIATUS), though it violates ONSET. Here without *VV and DEFAULT, EVAL will choose the non-surface form, *[waŋ.ga], as optimal.

The crucial difference from the previous analysis is witnessed in (18b), the case of ŋ-final stems. Previously, under ONSET-driven analysis, ŋ-final stems are left unexplained. We see that no ranking can possibly select the optimal form as long as there is the undominated constraint, *[σŋ. By shifting from ONSET to NOHIATUS, however, we see that the evaluation correctly picks out the optimal form. As such, we see that employing ONSET not NOHIATUS in the previous analyses was the cause of problem.

82 Y. LEE

3.2. Default Allomorph and SYLLCON

We now move on to tackle the problem with the lo/ilo alternation. As shown in (8) and (9c), the alternation seems to argue for specific subcategorization constraints. But what we actually observe here is the simple and straightforward result of evaluation. There is no special status of liquid-final stems. The need for grouping vowels with liquids is simply a pseudo-generalization. The actual evaluation does not depend on any special featural status of liquids. All we have to do is to put a markedness constraint, SYLLCON, as one of those anti-DEFAULT constraints (M1 in (16)). Consider the following evaluation tableaux.

(19) Evaluation tableaux of GOL/INS marker allomorphy11

a. after a V-final stem (cow-GOL/INS)

so-{lo, ilo} *[ σŋ *VV SYLL CON

DEFAULT ONS NO CODA

ALIGN- STEM

☞ i) so.ro ii) so.i.ro *! * *

b. after an ŋ-final stem (king-GOL/INS)

waŋ-{lo, i lo} *[ σŋ *VV SYLL CON

DEFAULT ONS NO

CODA ALIGN- STEM

i) waŋ.lo *! * ☞ ii) waŋ.i .ro * * *

iii) wa.ŋi .ro *! * *

c. after a liquid-final stem (water-GOL/INS)

mul-{lo, i lo} *[ σŋ *VV SYLL CON DEFAULT ONS

NO CODA

ALIGN- STEM

☞ i) mul.lo * ii) mu.ri.ro *! *

11 The tableau in (19d) is overly simplified. There may be at least two more logically possible candi-dates, *[soŋ.no] and *[so.glo] from /sok-lo/. The first one, *[soŋ.no], might be problematic, since we have the SYLL CON-complying /kl/�[ŋn] mapping in Korean as exemplified by /kiklak/�[kiŋ.nak] (paradise). But such mapping is not found in the noun-particle concatenation. We assume that a certain faithfulness constraint that penalizes the change in the noun stem due to the influence of suffixes, such as FAITH-STEM(obstruent) or IDENT(-son)/STEM is ranked higher than DEFAULT in Korean. Secondly *[so.glo] may be ruled out by an undominated onset cluster well-formedness constraint, since /gl/ is not a possible onset cluster in Korean. But here we may have a more general solution with the local constraint conjunction as discussed in connection with (21).


d. after a C-final stem (Inside-GOL/INS)

sok-{lo, ilo} *[ σŋ *VV SYLL CON DEFAULT ONS

NO CODA

ALIGN- STEM

i) sok.lo *! * ☞ ii) so.gi.ro * *

Before we discuss GOL/INS marker, it should be noted that positing SYLLCON over DEFAULT does not interfere with the explanation given in 3.1. The key to the explanation is that phonologically simple forms are favored unless such forms violate higher ranked markedness constraints such as *[σŋ or SYLLCON. Our main concern here is the evaluation given in (19c). Recall that the goal/instrument marker alternation was such a problem in Kim (2006) that she invokes the feature [+vocalic] to group liquids and vowels together. But with a universal constraint DEFAULT, we do not have to resort to the controversial feature [±vocalic] or the subcategorization-based selection stage as in Kim (2006). The irregular behavior of liquids is an epiphenomenon that comes from the segmental nature of the default allomorph, /lo/. Liquids are the most sonorous consonant in Korean (except for glides). Therefore, a consonant-liquid sequence over a syllable boundary leads to a SYLLCON violation except when the stem-final consonant is also a liquid. As such, stems with final liquids allow the default form /lo/, since there is no SYLLCON violation involved. Any other consonant in the stem final position will lead to the choice of the non-default form /ilo/ to avoid the SYLLCON violation. That is all there is to say about the goal/instrument marker allomorph selection. We do not see any special status of liquids.

Rather, a more complicated problem arises with the wa/kwa alternation. Consider the following evaluation tableaux:

(20) Evaluation tableaux for conjunctive marker allomorphy

a. after a V-final stem (cow-CONJ)

so-{wa, kwa} *[σŋ SYLL CON DEFAULT ONS *[ σCC NO

CODA ALIGN- STEM

☞ i) so.wa . ii) so.gwa *! *

b. after a C-final stem (Inside-CONJ)

sok-{wa, kwa} *[σŋ SYLL CON

DEFAULT ONS *[ σCC NO

CODA ALIGN- STEM

? ☞ i) so.gwa * * � ii) sok.k’wa *! * *

84 Y. LEE

We do not see any problem in (20a). But in (20b) even with DEFAULT, we end up with a wrong result. We need the ranking, ⟦ALIGN-STEM » DEFAULT⟧ for (20b), but it creates problems in (4a), (18b) and many other cases where V-initial suffixes are added to C-final stems. Sung (2005) notes the ranking paradox and concludes that OT fails to explain the wa/kwa alternation in Korean. This paper, however, presents a simple solution based on the local constraint conjunction (Alderete 1997). Consider the thick box in (20bi). If there is anything wrong with this candidate, it would be the joint violation of *[σCC and ALIGN-STEM at the same time within the domain of a prosodic word. Based on this observation, we may posit the constraint as in (21a):

(21) Local Constraint Conjunction and evaluation

a. Locally Conjoined Constraint (=LCC) [NOCOMPLEX/ONSET & ALIGN (STEM, RIGHT, σ, RIGHT)]PrWd

b. Evaluation of (20b)

sok-{wa, kwa} *[σŋ SYLL CON

LCC DEF ONS *[ σCC NO CODA

ALIGN- STEM

i) sog.wa *! * ii) so.gwa *! * * ☞ iii) sok.k’wa * * *

By placing LCC above DEFAULT as in (21b), we can obtain the desired optimal form. The constraint conjunction helps to eliminate the undesired form (21bii), and the constraint interaction selects the correct optimal form.12 Note also that LCC does not conflict with the previously explained data.

12 The same constraint, LCC, also helps to eliminate *[so.glo] in tableaux (19d) as noted in the foot-note (11). There may be another possibility to deal with (21). We may employ the comparative markedness proposal by McCarthy (2002b). Note that that [sok.k’wa] is a fully faithful candidate that contains NOCOMPLEX violation. However for /sok+wa/, the fully faithful candidate is *[sog.wa] which does not have NOCOMPLEX violation. In McCarthy’s term, *[so.gwa] in (20bi) has a new NOCOMPLEX violation, while [sok.k’wa] in (20biii) has an old NOCOMPLEX violation. We may posit NNOCOMPLEX (a new markedness constraint) over DEFAULT to eliminate the unwanted form, *[so.gwa], and choose the surface-true form, [sok,k’wa] in (20bii). However, given the far-reaching consequences of the markedness separation on such phonological phenomena as n-insertion and glide-formation in Korean, we will not consider this option here.


3.3. Some Residual Issues

So far, we have explained the three major problems witnessed in the previous analyses. In doing so, we might have caused another non-trivial problem. Sung (2005: 46) cites two more allomorph sets as in (22):

(22) Two more allomorph sets

a. Disjunctive marker na/ina alternation cha + disjunctive → [cha.na], *[cha.i.na] (a car or …) cip + disjunctive → [ci.bi.na], *[cib.na] (a house or …) mul + disjunctive → [mu.ri.na], *[mul.na] (water or …) son + disjunctive → [so.ni.na], *[son.na] (a hand or …)

b. Conditional maker myən/imyən alternation cha + conditional → [cha.myən], *[cha.i.myən] (if … a car) cip + conditional → [ci.bi.myən], *[cib.myən] (if … a house) mul + conditional → [mu.ri.myən], *[mul.myən] (if … water) son + conditional → [so.ni.myən], *[son.myən] (if … a hand)

The distribution is straightforward. V-initial forms are used after C-final stems and C-initial forms after V-final stems. These markers, however, if treated as cases of the noun suppletive allomorphy, pose a serious challenge to the present analysis. Consider the following unfortunate evaluation tableaux:

(23) Wrong evaluations of disjunctive marker allomorphy (22a)

a. after a stem-final liquid

mul-{na, ina} *[σŋ *VV SYLL CON

DEFAULT NO CODA

ALIGN- STEM

? ☞ i) mul.na * � ii) mu.ri.na *! *

b. after a stem-final nasal

son-{na, ina} *[σŋ *VV SYLL CON DEFAULT

NO CODA

ALIGN- STEM

? ☞ i) son.na * � ii) so.ni.na *! *

As shown in (23), the disjunctive marker allomorphy after sonorant-final stems may stand at odds with the present proposal. We may somehow posit a markedness constraint *LN (No lateral before nasal) for (23a) and put it over

86 Y. LEE

DEFAULT to fix the problems. But note that this is of no help to (23b). A sequence of two nasals is perfectly legitimate in Korean both in the underlying forms ([ən.ni] (sister)) and also in derived forms (/ka-(ə)s’-ni/ → [kan.ni] (Did he go?)).

This paper, however, argues that these are not true noun particles per se. The so-called disjunctive marker actually is the combination of existential marker {i} and verbal conjugation marker {na}. Also note that the verbal conjugation marker {na} has two variants: /na/ and /ina/, as exemplified by [sa.na.cu.gi.na] (live or die). The same is true also for the myən/imyən alternation in (22b). We have already seen that the conditional marker for verbs shows the myən/imyən alternation in (1).

Syntactically speaking, the noun-existential marker sequence forms a phrase, a verb phrase to be more specific, and the verbal suffixes, {na, ina}, follow the phrase. This explains the existence of many related alternations like lato/ilato (Concession I), ciman/iciman (Concession II), ni/ini (Resultative/Question), myənsə/imyənsə (Concomitant), to cite a few.

We suggest that the underlying form of [mu.ri.na] in (23a) is /mul-i-na/, temporarily ignoring the na/ina alternation. There is no allomorphy involved here. And as for the vowel final stems, for example [so.na], the input is /so-i-na/. Then the existential marker /i/ is deleted violating MAX -AFFIX to avoid the vowel hiatus under the ranking, ⟦NOHIATUs » MAX -AFFIX⟧. As such, what we see is the result of two independent processes: the concatenation of the verbal suffixes and the deletion of the existential marker /i/, by the demand of NOHIATUS. Consider the following tableaux:

(24) Evaluation of a noun-existential marker-suffix sequence

a. after a stem-final consonant

mul-i-na *[σŋ *VV SYLL CON

MAX- AFFIX DEFAULT

NO CODA

ALIGN- STEM

i) mul.na *! * ☞ ii) mu.ri.na *

b. after a stem-final vowel

cha-i-na *[σŋ *VV SYLL CON

MAX- AFFIX

DEFAULT NO CODA

ALIGN- STEM

i) cha.na * ☞ ii) cha.i.na *!

MAX-AFFIX may be placed anywhere under *VV, though the highest possible rank is given in (24). The tableaux in (24) clearly show that verbal conjugation


allomorphy should be treated separately.13 But since this paper is not committed to the analysis of verbal conjugation allomorphy, we simply conclude that the examples given in (22) should be dealt with in connection with verbal conjugation and therefore they do not constitute counter-examples for the proposal made in this paper.

4. CONCLUSION

This paper revisits allomorph alternations in the Korean particles. It first reviews problems in the previous analyses, which argue against MIH, and then shows that OT with MIH can account for them. Two major propositions are made in this paper. One is the need to recognize the role of the constraint, NOHIATUS, which cannot be replaced by ONSET. Second is the postulation of the universal constraint, DEFAULT, that works for choosing the optimal allomorph.

NOHIATUS, but not ONSET, can explain the emergence of vowel-initial suffix after ŋ-final stems. NOHIATUS selectively penalizes V.V sequence while allowing Vŋ.V. The problem with previous analyses with ONSET is that it assigns penalty marks indiscriminately both to V.V and Vŋ.V. It is also shown that DEFAULT can save [Vŋ.i] from being harmonically bounded by *[Vŋ.ga]. DEFAULT also explains the unexpected behavior of liquids in lo/ilo alternation. Viewed from this point, the superficial language-specificity is the natural result from the interaction of relevant universal constraints.

DEFAULT is introduced as a universal constraint that prefers a phonologically simpler allomorph. This constraint is crucially dominated by such markedness constraints as *ONSET/ŋ, NOHIATUS and SYLLCON. The surface form shows optimization with respect to these constraints. But since it dominates some markedness constraints like ONSET and NOCODA in Korean, the surface form may have violations of these constraints. This explains the partially optimizing effect in Korean noun particle allomorphy.

The analysis given in this paper shows that the dynamic variance and superficial complexity of the Korean suppletive allomorphy do not require any language-specific constraints or any separate allomorph selection stages. The inter-action of universal constraints properly selects the optimal allomorph from the multiple inputs. Though there are alternatives in the previous analyses that hinge on certain language-specific subcategorization constraints or any separate

13 One of the anonymous reviewers points out that the i/ø alternation shown in (24) may be viewed as a morphological truncation rather than a phonological deletion. It may be true, but still we see that the truncation is motivated by the presence of the high ranking *VV constraint and the result is the violation of MAX -AFFIX.

88 Y. LEE

stage other than EVAL, we see that the Multiple Input approach to allomorphy presents a simpler and more natural analysis for the Korean particle alternations.

Putting the analysis given in this paper in a wider context, we see that the OT framework provides the better means of explanation, as the concept of the default allomorph may not be easily captured in the rule-based approach. Further, we see that Korean noun particle allomorph supports the need for preferential allomorph that interacts with other constraints to choose the right allomorph among multiple inputs. And such preference can be encoded in the form of a universal constraint, not of a language-particular and morpheme-specific allomorph constraint.

REFERENCES

Alderete, John. 1997. Dissimilation as local conjunction. Kusumoto, Kiyomi, ed., Proceedings of North Eastern Linguistics Society 27: 17-32.

Bonet, Eulàlia, Maria-Rosa Lloret and Joan Mascaró. 2007. Allomorph selection and lexical preferences: Two case studies. Lingua 117(6): 903-927.

Boroff, Marianne L. 2003. Against an onset approach to hiatus resolution. Ms., State University of New York, Stoney Brook. [ROA#586]

Boyd, Jeremy. 2006. On the representational status of /s/-clusters. San Diego Linguistic Papers 2: 39-84.

Blaho, Synvia, Patrik Bye, and Martin Krämer, eds. 2007. Freedom of analysis? Berlin: Mouton de Gruyter.

Bye, Patrik. 2007. Allomorphy - selection, not optimization. Blaho, Sylvia et al. 1980: 63-91. Davis, Stuart and S.-H. Shin. 1999. The syllable contact constraint in Korean: An optimality-

theoretic analysis. Journal of East Asian Linguistics 8: 285–312. Hargus, Sharon and Ellen Kaisse, eds. 1993. Phonetics and phonology 4: Studies in lexical

phonology. San Diego: Academic Press, Inc. Hong, Sunghoon. 2001. Richness of the base, lexicon optimization, and suffix /i/~ø alternation in

Korean. Studies in Phonetics, Phonology and Morphology 7(1): 215-242. Kager, René 1996. On affix allomorphy and syllable counting. Kleinhenz, Ursula, ed. 1996: 155-171.

[Prepublication version at ROA-88] Kim, Gyung-Ran. 2006. Suffix-centered allomorphy in Korean. Studies in phonetics, phonology and

morphology 12.2: 265-281. Kleinhenz, Ursula, ed. 1996. Interfaces in phonology [Studia Grammatica 41] Berlin: Akademie

Verlag. Lapointe, Steven G. 1999. Stem selection in OT. In Booij, Geert & Jaap van Marle, eds., Yearbook

of Morphology 1999. Dordrecht: Kluwer Academic Publishers. 263-297. Lindstrom, Lamont and John Lynch. 1994. Kwamera. München: Lincom Europa. Mascaró, Joan. 2007. External allomorphy and lexical representation. Linguistic Inquiry 38.4: 715-

735. McCarthy, John J. 2002a. A thematic guide to optimality theory. Cambridge: Cambridge University

Press. ———. 2002b. Comparative markedness. Ms., University of Massachusetts. [ROA-489]


——— and Alan Prince. 2001. Prosodic morphology I. Ms., University of Massachusetts at Amherst and Rutgers University [ROA-482]

——— and Matthew Wolf. 2005. Less than zero: Correspondence and the null output. Ms., University of Massachusetts Amherst. [ROA-722]

Odden, David. 1993. Interaction between modules in lexical phonology. Hargus, Sharon and Ellen Kaisse, eds. 1993: 111-144.

Paster, Mary Elizabeth. 2006. Phonological conditions on affixation. Doctoral dissertation, University of California, Berkeley.

Prince, Alan and Paul Smolensky. 2002. Optimality theory: Constraint interaction in generative grammar. Ms., Rutgers University and The John’s Hopkins University. [ROA-537]

Russel, Kevin. 1996. Morphemes and candidates in optimality theory. Ms., University of Manitoba. [ROA 44]

Sung, Eun-Kyung. 2005. Allomorph selection in Korean. Studies in Phonetics, Phonology, and Morphology 11.1: 45-60.

[This paper has greatly benefited from the discussion with Stuart Davis and Minkyung Lee. And special thanks are due to the two anonymous reviewers who made this paper more consistent and readable. The author is solely responsible for all errors and misrepresentations in this paper.] Yongsung Lee <[email protected]>, <[email protected]> Department of English Pusan University of Foreign Studies Pusan 608-738, South Korea

[Received 20 February 2008; revision received 16 May 2008;

accepted 17 May 2008]

allomorphy in korean noun particles:acms.bufs.ac.kr/yslee/research/papers/53allomorhy... ·...

Documents