2.1 Short-form negation as syntactic negation

One way of negating a clause in Korean, commonly called short-form negation,Footnote ² is to place the negator an(i) before the predicate, as in an(i) ca-ss-ta ‘not sleep-past-decl’, an(i) mak-nun-ta ‘not block-pres-decl’ and an(i) cak-ass-ta ‘not small-past-decl’. Without the negator, the remaining verbal expression is affirmative. I argue, following J.-Y. Yoon (Reference Yoon1990), H.-D. Ahn (Reference Ahn1991), Y.-T. Hong (Reference Hong1992) and Han, Lidz & Musolino (Reference Han, Jeffrey and Musolino2007) among others, in favor of a syntactic analysis of short-form negation. More specifically, the negator an(i) is the head of the functional category NegP.

Short-form negation with an(i) is allowed before a negative prefix such as pu(l)-, pi- and mi-, as in an(i) pul-kanungha- ‘impossible’, an(i) pi-kwahakcek-i- ‘unscientific’ and an(i) mi-wanseng-i/toy- ‘incomplete’, but not before another instance of an(i), as in *an(i) an(i) kanungha-, *an(i) an(i) kwahakcek-i- and *an(i) an(i) wanseng-i/toy-. The ungrammaticality of stacking the negator before a predicate stands in contrast to the restriction on multiple occurrences of the negative prefixes. The latter restriction is that a predicate can have only one of these prefixes, making forms such as *pul-pul-A, *pi-pi-A, *pi-pul-A, *pul-mi-A, *mi-pi-A ungrammatical. The similarity and the contrast between the negator an(i) and the prefixes show that they are different syntactically and semantically. If an(i) were a prefix comparable to pu(l)-, pi- and mi-, its cooccurrence with the latter could not be explained without a special stipulation on the negator an(i). By contrast, if an(i) is not such a prefix, there will be no problem in having this negative element in addition to a negative prefix. This shows that an(i) is not a negative prefix.

One characteristic of syntactic negation (as opposed to negative prefixes) concerns its scope interaction with respect to quantifiers. The negator an(i) shows a clear scope ambiguity effect with a quantifier, as in the following examples, which both have two readings. (See I. Chung Reference Chung2007a for extensive discussion supporting the ‘hard-to-get’ Neg>∀ reading.)

1. (1)

   1. (a) motun haksayng-i an(i) ka-ss-ta. all student-nomneg go-past-decl ∀>Neg: ‘No student went.’ Neg>∀: ‘It is not the case that all the students went.’

   2. (b) wuli-ka motun chinkwu-lul an(i) manna-ss-ta. we-nom all friend-accneg meet-past-decl ∀>Neg: ‘We met no friend.’ Neg>∀: ‘It is not the case that we met all the friends.’

By contrast, the above scope ambiguity does not arise with predicates having a negative prefix, as shown below.

1. (2)

   1. (a) motun haksayng-i i an-ey pulchansengha-y-ess-ta. all student-nom this plan-dat disapprove-eg-past-decl ∀> pul-: ‘All the students disapproved this plan.’ *pul- > ∀

   2. (b) wuli-ka motun an-ey pulchansengha-y-ess-ta. we-nom all plan-dat disapprove-eg-past-decl ∀ >pul-: ‘We disapproved all the plans.’ *pul- >∀

The contrast between an(i) and the negative prefixes regarding quantifiers is a clear indication that an(i) is not a negative prefix but a syntactic negator.

Another piece of evidence in favor of the syntactic view of short-form negation comes from licensing of negative polarity items. One typical environment where a negative polarity item is licensed is a negative clause, which contains a syntactic (and semantic) negation element. In the following examples, the negator an(i) serves as such a licensor.

1. (3)

   1. (a) amu-to an(i) sengsilha-ta. any-npineg sincere-decl ‘Nobody is sincere.’

   2. (b) na-nun amu-to an(i) manna-ss-ta. I-top any-npineg meet-past-decl ‘I didn't meet anybody.’

   3. (c) amu-to na-lul an(i) manna-ss-ta. any-npi I-accneg meet-past-decl ‘Nobody met me.’

The crucial point is that a sentence without an(i) or with just a negative prefix cannot license a negative polarity item:

1. (4)

   1. (a) *amu-to (pul-)sengsilha-ta. any-npi in-sincere-decl

   2. (b) *con-un amu pep-(ey)-to (pul-)pokcongha-y-ess-ta. John-top any law-dat-npi dis-obey-eg-past-decl

   3. (c) *amu-to i pep-ey (pul-)pokcongha-y-ess-ta. any-npi this law-dat dis-obey-eg-past-decl

If an(i) were a prefix like pul-, it should not be able to license a negative polarity item (see I. Chung Reference Chung2007a for further supporting arguments). This contrast between an(i) and negative prefixes is another indication that an(i) is not a prefix but rather a syntactic negator which licenses a negative polarity item. From the above discussion, I conclude that short-form negation is a syntactic construction, not an instance of ‘lexical’ prefix attachment.

2.2 Suppletive negation: molu- ‘not.know’ and eps- ‘not.exist’

This section looks into the two suppletive negative predicates. The affirmative roots in question are al- ‘know’ and iss- ‘exist’, and their negative counterparts are molu- ‘not.know’ and eps- ‘not.exist’. These suppletive forms behave like short-form syntactic negation with respect to the distribution of the negator, negative polarity item licensing and scope ambiguity.

The suppletive negative roots molu- and eps- pattern as if a syntactic negator were present in them, in that they cannot be preceded by an(i): *an(i) molu- and *an(i) eps-. This suggests that these negative roots are the result of syntactic negation, which is usually done by an(i).

Negative polarity items further support the syntactic negation analysis of molu- and eps-. These predicates license a negative polarity item without the negator an(i), as in the following sentences.

1. (5)

   1. (a) na-nun amu tap-to moll-ass-ta. I-top any answer-npi not.know-past-decl ‘I didn't know any answer.’

   2. (b) amu-to tap-ul moll-ass-ta. any-npi answer-acc not.know-past-decl ‘Nobody knew the answer.’

1. (6) amu-to eps-ta. any-npi not.exist-decl ‘There is nobody.’

Quantifiers in sentences with molu- and eps- support these roots' status as syntactic negation. Consider the scope ambiguity between a quantifier and these suppletive negative predicates.

1. (7)

   1. (a) motun haksayng-i ku wuhwa-lul molu-n-ta. all student-nom the fable-acc not.know-pres-decl ∀>Neg: ‘No student knows the fable.’ Neg>∀: ‘Not all students know the fable.’

   2. (b) ku haksayng-i motun wuhwa-lul molu-n-ta. the student-nom all fable-acc not.know-pres-decl ∀>Neg: ‘The student knows no fable.’ Neg>∀: ‘The student doesn't know all the fables.’

1. (8) motun haksayng-i eps-ta. all student-nom not.exist-decl ∀>Neg: ‘There are no students.’ Neg>∀: ‘Not all students are present.’

This scope ambiguity is unique to syntactic negation and thus supports the syntactic negation status of the predicates molu- and eps-.Footnote ³

Negation of iss-, however, does not always result in eps-. In other cases, the usual negation an(i) iss- is found as well. Consider the following:

1. (9)

   1. (a) eysute-ka yeki-ey iss-ess-ta. Esther-nom here-loc exist/stay-past-decl ‘Esther was here.’

   2. (b) eysute-ka yeki-ey eps-ess-ta. Esther-nom here-loc not.exist-past-decl ‘Esther was not here.’

   3. (c) eysute-ka yeki-ey an(i) iss-ess-ta. Esther-nom here-locneg be-past-decl ‘Esther was not/did not stay here.’

There is a semantic difference between the two (negated) predicates. While eps- means nonexistence ‘not exist’ or simple absence ‘be not present, be lacking’, an(i) iss- involves intention and means ‘intentionally not stay’.Footnote ⁴ Also, the cases where an(i) iss- is used must have an animate subject which is an agent. In (9), both the eps- negation and the an(i) iss- negation are possible, because iss- is ambiguous between the existential meaning and the agentive meaning ‘stay intentionally’.

2.3 A postsyntactic fusion analysis of suppletive negation

Sells (Reference Sells2001), in the same vein as Kim (Reference Kim, Kuno, Lee, Whitman, Maling, Kang and Kim1999), presents Korean as having three different kinds of clausal negation: long-form negation, short-form negation and suppletive negation (‘lexically negative verbs’, in his terminology). However, this classification misses the fact that the affirmative predicates of suppletive negative predicates do not have the expected short-form negation construction, and that long-form negation constructions do not have such suppletive cases. Short-form negation and suppletive negation can be systematically related in the Distributed Morphology as proposed in I. Chung (Reference Chung2007a) Footnote ⁵ and summarized below.

The following diagram represents the (partial) structure of C⁰ in Korean, as a result of overt movement of the verbal/adjectival root (in syntax or in morphology). This word structure reflects the phrase structure of a clause.Footnote ⁶

1. (10)

   

The commonalities between the regular syntactic short-form negation construction and suppletive negative predicates suggest that there is a single short-form negation construction in syntax (overt and covert) encompassing both types of negation, and that the difference between the two arises in PF (i.e., in the postsyntactic component). The two roots ‘know’ and ‘exist’ (or their negative exponents) show the same configurations and properties in syntax and semantics as do other predicates. The difference is that suppletive negative forms do not have a separate negator and that these negative predicates are not morphophonologically related to their affirmative root.

The following is the structure of the verbal complex molu-n-ta ‘not know’ at Spell-Out, lacking phonological features.

1. (11) Structure for molu-n-ta at the end of overt syntax

   

This structure enters PF and undergoes fusion. Fusion takes the V node containing [know] and the Neg node with the [+neg] feature and turns them into a single terminal node. This resulting node contains all the original syntactico-semantic features, as shown in (12):Footnote ⁷

1. (12) Fusion of Neg and V with [know] in PF (partial)

   

The Neg node is fused with the V node only if its sister node is the verb [know] or [exist], and not any other verb. Subsequently, vocabulary insertion takes place. The following are the relevant vocabulary items:

1. (13)

   1. (a) [+neg, know]↔/molu/

   2. (b) [know]↔/al/

1. (14)

   1. (a) [+neg, exist]↔/eps/

   2. (b) [exist]↔/iss/

1. (15) [+neg]↔/an(i)/

Potentially, any of the vocabulary items (13a), (13b) and (15) can be inserted into the node (12b). In such a situation, however, the most highly specified item compatible with a given terminal node wins the competition. Hence, (13a) /molu/ is chosen for the fused node, resulting in:

1. (16) Vocabulary insertion of /molu/ (into the fused node)

   

By contrast, fusion does not occur in a negative clause with verbs showing the regular short-form negation, and the original V and Neg nodes proceed to vocabulary insertion independently.

The sisterhood relation in (11) (and likewise in (12)) is crucial for the fusion operation. Therefore, fusion will not take place if another element intervenes between Neg and V in the input structure. Consider the derived causative of ‘know’ in the negative context. The causative feature is the head of v(P), which intervenes structurally between V(P) and Neg(P). After head movement of V→v→Neg→T→C, the following structure is obtained as part of the C⁰ complex of the causative predicate.

1. (17) Structure for an(i) al-li-ess- ‘neg know-caus-past-’ at the end of overt syntax

   

Here, Neg and V, though linearly adjacent, are not sisters, and therefore fusion is inapplicable. Then, each terminal node in the above structure separately undergoes vocabulary insertion, correctly yielding [an(i) [al-li]] with the vocabulary item (18).

1. (18) [+caus] ↔ /li/

The case of the existential iss- ‘exist’ and its negative counterpart eps- is treated in the same manner. With the root [exist], the output of overt syntax (19a) is turned into the fused structure (19b) in morphology with the vocabulary items in (20).

1. (19) Fusion of Neg and the V node of [exist]

   

1. (20)

   1. (a) [+neg, exist]↔/eps/

   2. (b) [exist]↔/iss/

Due to competition, (20a) is chosen over (20b) and (15) for the resulting fused node (19b).

The agentive iss- ‘stay intentionally’ (whose negative form is the non-suppletive an(i) iss-) behaves like any regular predicate. Agentive predicates are assumed to have an additional v(P) category intermediate between V(P) and Neg(P), associated with agentivity/volitionality and the agent theta role (see Hale & Keyser Reference Hale, Keyser, Hale and Keyser1993 and Chomsky Reference Chomsky1995a among others). The following structure illustrates this situation.

1. (21) Structure of the complex C for the agentive iss- in a negative clause (partial)

   

In this structure, unlike (19a), [+neg] and [exist] are not sister nodes and hence they cannot undergo fusion. Given that the [+agent] feature is phonologically null, vocabulary insertion applies to each node separately, yielding the correct form, an(i) iss-Ø-.

This section has presented the Distributed Morphology analysis of negative suppletion of ‘know’ and ‘exist’ in Korean. The morphological fusion operation is sensitive to the syntactico-semantic features of the terminal nodes being fused (i.e., [know] and [exist]). Another critical aspect of the fusion operation is its structure: only sister nodes can be fused and a hierarchically intervening node blocks this operation. Finally, fusion occurs after syntax. The analysis of fusion of the negation node and the V root node in PF explains why the two non-long-form negation cases exhibit the same syntactic and semantic behaviors. It also explains why there is no independent negator in the two suppletive negation cases and why a single vocabulary item appears instead of the negation-root sequence.Footnote ⁸

3.1 Honorific root suppletion and blocking of negative suppletion

Subject honorification or subject exaltation is used in Korean when the subject of the sentence is honorified or exalted by a speaker who is inferior to the subject (in terms of age, social status, etc.). In this case the predicate takes the honorific suffix -si- as shown below (with an epenthetic vowel -u- to break up two adjacent heteromorphemic consonants):Footnote ⁹

1. (22)

   1. (a) eysute-ka nichey-lul ilk-ess-ta. Esther-nom Nietzsche-acc read-past-decl ‘Esther read Nietzsche (non-honorific).’

   2. (b) apeci-kkeyse nichey-lul ilk-u-si-ess-ta. father-hon.nom Nietzsche-acc read-ev-hon-past-decl ‘(My) father read Nietzsche (honorific).’

1. (23)

   1. (a) eysute-ka pulwunsu-lul pulu-n-ta. Esther-nom Bruhns-acc sing-pres-decl ‘Esther is singing Bruhns (non-honorific).’

   2. (b) apeci-kkeyse pulwunsu-lul pulu-si-n-ta. father-hon.nom Bruhns-acc sing-hon-pres-decl ‘(My) father is singing Bruhns (honorific).’

1. (24)

   1. (a) eysute-ka khu-ta. Esther-nom big-decl ‘Esther is big (non-honorific).’

   2. (b) apeci-kkeyse khu-si-ta. father-hon.nom big-hon-decl ‘(My) father is big (honorific).’

There are a few predicates whose root form is morphophonologically totally different from the usual root form when the honorific suffix -si- is attached.

1. (25)

   1. (a) eysute-ka ttek-ul mek-ess-ta. Esther-nom rice cake-acc eat-past-decl ‘Esther ate rice cake (non-honorific).’

   2. (b) apeci-kkeyse ttek-ul capswu-si-ess-ta. father-hon.nom rice cake-acc eat.hon-hon-past-decl ‘Father ate rice cake (honorific).’

   3. (c) *apeci-kkeyse ttek-ul mek-u-si-ess-ta. father-hon.nom rice cake-acc eat-ev-hon-past-decl

1. (26)

   1. (a) eysute-ka pang-eyse ca-n-ta. Esther-nom room-loc sleep-pres-decl ‘Esther is sleeping in the room (non-honorific).’

   2. (b) apeci-kkeyse pang-eyse cwumu-si-n-ta. father-hon.nom room-loc sleep.hon-hon-pres-decl ‘Father is sleeping in the room (honorific).’

   3. (c) *apeci-kkeyse pang-eyse ca-si-n-ta. father-hon.nom room-loc sleep-hon-pres-decl

Such honorific root forms are not possible without the honorific suffix: *capswu-n-ta ‘eat.hon-pres-decl; *cwumu-ess-ta ‘sleep.hon-past-decl’.

An interesting situation arises when negation and honorification occur together with a predicate that has both a suppletive negative form and a suppletive honorific form. The predicate iss- ‘exist’ (with the negative form eps- ‘not.exist’ and the honorific form kyey-si- ‘exist-hon’) is realized as kyey- in this situation: an(i) kyey-si-.Footnote ¹⁰

1. (27)

   1. (a) apeci-kkeyse silhemsil-ey an(i) kyey-si-ta. father-hon.nom lab-locneg exist.hon-hon-decl ‘Father is not in the lab (honorific).’

   2. (b) *apeci-kkeyse silhemsil-ey eps-u-si-ta. father-hon.nom lab-loc not.exist-ev-hon-decl

   3. (c) *apeci-kkeyse silhemsil-ey an(i) iss-u-si-ta. father-hon.nom lab-locneg exist-ev-hon-decl

Since only one of the two competing suppletion processes (i.e., honorific suppletion) operates here, it can be utilized to identify the morphological structure of the morphologically complex predicate.

As assumed in Distributed Morphology (Halle & Marantz Reference Halle, Marantz, Hale and Keyser1993, Harley & Noyer Reference Harley and Noyer1999, Bobaljik Reference Bobaljik, Grohmann and Struijke2000 and Embick Reference Embick2007 among others), vocabulary insertion takes place starting from the most deeply embedded terminal node (i.e., from the root) and moves outward cyclically. When a terminal node is provided with the phonological features in a given cycle, this vocabulary insertion may be sensitive to morphosyntactic features present in an outer cycle. This aspect of ‘outwards sensitivity’ can be observed in the following structure for an(i) kyey-si-ess-ta ‘neg exist.hon-hon-past-decl’ produced by overt syntax.

1. (28) Structure of an(i) kyey-si-ess-ta at the end of overt syntax (partial)

   

When the root node is provided with phonological features in (28), the [+hon] feature in the next outer cycle chooses /kyey/ for [exist]. Formally, given the vocabulary items for [exist] and other relevant vocabulary items below, vocabulary insertion provides /kyey/ in (28) for [exist] in the environment of [+hon], which is the result of choosing (29b).

1. (29) Vocabulary items containing the feature[exist]

   1. (a) [+neg, exist]↔/eps/

   2. (b) [exist]↔/kyey/ / ____ [+hon]

   3. (c) [exist]↔/iss/

1. (30) Vocabulary items with a functional category feature

   1. (a) [+hon]↔/si/

   2. (b) [+neg]↔/an(i)/

   3. (c) [+past]↔/ess/

To ensure that the honorific-suppletive root /kyey/ is chosen in an environment where both negation and honorification occur, the structure of the inflected predicate would be expected to be (28), rather than (31) below, in that in (28) the [+hon] node is structurally closer to the root than the [+neg] node is.

1. (31) Possible alternative structure for [+neg]-[exist]-[+hon]-

   

In addition, because the root node and [+neg] are sisters in (31), fusion of these two nodes, (19), would wrongly take place in (31), resulting in eps-. The structure (28) effectively blocks the suppletive negative exponent. Vocabulary insertion converts (28) to [[[+neg] [/kyey/ [+hon]]] [+past]] and ultimately to [[an(i) [kyey si]] ess].

However, other evidence argues strongly that the correct structure is in fact (31) and not (28), as will be shown in the following section.

3.2 Honorific suppletion and negative suppletion: a paradox

A first problem arises when the structure for [+neg]-[know]-[+hon]- is considered along with the above structure (28) for [+neg]-[exist]-[+hon]-. In the following structure (32), identical to (28) except for the root morpheme, vocabulary insertion should take place separately, i.e., without fusion, and the result would be *an(i) al-si- (or *an(i) a-si-, due to a general phonological process deleting [l] before a coronal consonant).

1. (32) Structure for[+neg]-[know]-[+hon]- before vocabulary insertion

   

However, the correct form involves a fusion of [+neg] and [know], i.e., molu-si-. The problem is the failure of fusion because the description of the fusion rule is not met: the root and the [+neg] node are not sisters in (32). Accordingly, the suppletive behaviors of ‘exist’ and of ‘know’ result in a paradoxical situation regarding whether [+neg] or [+hon] is to be the node structured more closely to the root.

3.3 Predicate iteration constructions

An independent diagnostic phenomenon is called for in order to determine which of the two non-root nodes, i.e., negation or honorific, is structurally closer to the root. Predicate iteration constructions will show that negation is in fact more closely structured with the root than is honorification. I provide three such constructions: iterated rhetorical questions, the echoed verb construction, and the ha- focus construction.

The first predicate iteration construction is the iterated rhetorical question, where a portion of the inflected predicate is repeated and the second copy appears with a wh-phrase. This paper does not discuss which of the two instances of the predicate is the original and which is the ‘copy’, although it appears that the second is the original because of the canonical OV word order in Korean and because the first instance is nominalized and contains only a portion of the fully inflected predicate. When the predicate is copied, there is some freedom regarding the copied affixes. The honorific suffix may or may not appear in both instances of the predicate as shown in (33).

1. (33) Iterated rhetorical question: affirmative, honorific

   1. (a) apeci-kkeyse po-si-ki-nun nwukwu-lul po-si-ess-e? father-hon.nom see-hon-nmlz-foc who-acc see-hon-past-inf ‘Father didn't see anybody, indeed. (lit. Who on earth did father see?)’

   2. (b) ?apeci-kkeyse po-si-ki-nun nwukwu-lul po-ass-e?

   3. (c) apeci-kkeyse po-ki-nun nwukwu-lul po-si-ess-e?

Negation, however, is obligatorily included in both copies of the predicate.

1. (34) Iterated rhetorical question: negative, non-honorific

   1. (a) eysute-ka an(i) po-ki-nun nwukwu-lul an(i) po-ass-e? Esther-nomneg see-nmlz-foc who-accneg see-past-inf ‘Esther did see somebody, indeed. (lit. Who on earth didn't Esther see?)’

   2. (b) *eysute-ka an(i) po-ki-nun nwukwu-lul po-ass-e?

   3. (c) *eysute-ka po-ki-nun nwukwu-lul an(i) po-ass-e?

This asymmetry is confirmed in the following examples involving both negation and the honorific suffix.

1. (35) Iterated rhetorical question: negative, honorific

   1. (a) apeci-kkeyse an(i) po-si-ki-nun father-hon.nomneg see-hon-nmlz-foc

   2. (b) nwukwu-lul an(i) po-si-ess-e? who-accneg see-hon-past-inf ‘Father did see somebody, indeed. (lit. Who on earth didn't father see?)’

   3. (c) *apeci-kkeyse an(i) po-si-ki-nun nwukwu-lul po-si-ess-e?

   4. (d) *apeci-kkeyse po-si-ki-nun nwukwu-lul an(i) po-si-ess-e?

   5. (e) ?apeci-kkeyse an(i) po-si-ki-nun nwukwu-lul an(i) po-ass-e?

   6. (f) apeci-kkeyse an(i) po-ki-nun nwukwu-lul an(i) po-si-ess-e?

Because negation is an obligatory element in each copy and the honorific suffix is not, negation can be considered to form a smaller constituent with the root than the constituent including the honorific suffix.

Exactly the same point is made by the second predicate iteration construction: the echoed verb construction (see No Reference No and Baek1988, T. Chung Reference Chung1994, Lee Reference Lee1995, Cho, Kim & Sells 2004 and Aoyagi Reference Aoyagi, Kuno, Lee, Whitman, Maling, Kang, Sells and Sohn2006 among others for discussions of this construction). In (36), the negation appears obligatorily in both copies of the iterated predicate.

1. (36) Echoed verb construction: negative, non-honorific

   1. (a) ailin-i eysute-lul an(i) po-ki-nun an(i) po-ass-e. Irene-nom Esther-accneg see-nmlz-focneg see-past-inf ‘Irene certainly did not see Esther.’

   2. (b) *ailin-i eysute-lul an(i) po-ki-nun po-ass-e.

   3. (c) *ailin-i eysute-lul po-ki-nun an(i) po-ass-e.

However, the honorific suffix need not appear in both copies, as shown in (37).

1. (37) Echoed verb construction: affirmative, honorific

   1. (a) apeci-kkeyse eysute-lul po-si-ki-nun po-si-ess-e. father-hon.nom Esther-acc see-hon-nmlz-foc see-hon-past-inf ‘Father certainly saw Esther.’

   2. (b) ?apeci-kkeyse eysute-lul po-si-ki-nun po-ass-e.

   3. (c) apeci-kkeyse eysute-lul po-ki-nun po-si-ess-e.

The examples in (38), containing both the negator and the honorific suffix, show the same asymmetry between these two affixes.

1. (38) Echoed verb construction: negative, honorific

   1. (a) apeci-kkeyse eysute-lul an(i) po-si-ki-nun an(i) po-si-ess-e. father-hon.nom Esther-accneg see-hon-nmlz-focneg see-hon-past-inf ‘Father certainly did not see Esther.’

   2. (b) *apeci-kkeyse eysute-lul an(i) po-si-ki-nun po-si-ess-e.

   3. (c) *apeci-kkeyse eysute-lul po-si-ki-nun an(i) po-si-ess-e.

   4. (d) ?apeci-kkeyse eysute-lul an(i) po-si-ki-nun an(i) po-ass-e.

   5. (e) apeci-kkeyse eysute-lul an(i) po-ki-nun an(i) po-si-ess-e.

Again, the obligatory negation supports the claim that negation is closer to the root than the optional honorific suffix is.

The third type of predicate iteration construction is the ha- focus construction, where the second predicate is the pro-form ha- ‘do’ (see Kang Reference Kang1988, Cho, Kim & Sells Reference Cho, Kim and Sells2004 and Aoyagi Reference Aoyagi, Kuno, Lee, Whitman, Maling, Kang, Sells and Sohn2006 among others). When the original predicate contains negation, the pro-form replaces the portion minimally including both the root and the negation, as in (39).

1. (39) Ha- focus construction: negative, non-honorific

   1. (a) ailin-i eysute-lul an(i) po-ki-nun ha-y-ess-e. Irene-nom Esther-accneg see-nmlz-foc do-eg-past-inf ‘Irene certainly did not see Esther.’

   2. (b) *ailin-i eysute-lul an(i) po-ki-nun an(i) ha-y-ess-e.

   3. (c) *ailin-i eysute-lul po-ki-nun an(i) ha-y-ess-e.

The honorific suffix, on the other hand, may or may not be part of the pro-form, as shown in (40).

1. (40) Ha- focus construction: affirmative, honorific

   1. (a) apeci-kkeyse eysute-lul po-si-ki-nun ha-si-ess-e. father-hon.nom Esther-acc see-hon-nmlz-foc do-hon-past-inf ‘Father certainly saw Esther.’

   2. (b) apeci-kkeyse eysute-lul po-si-ki-nun ha-y-ess-e.

   3. (c) apeci-kkeyse eysute-lul po-ki-nun ha-si-ess-e.

Another difference between the negator and the honorific suffix is that the honorific suffix -si- can appear on either the lexical predicate or on ha- or both. The honorific suffix is not required to appear on the main predicate, unlike the negator.

The contrast between negation and the honorific suffix is confirmed in (41), where the predicate has both of these affixes.

1. (41) Ha- focus construction: negative, honorific

   1. (a) apeci-kkeyse eysute-lul an(i) po-si-ki-nun ha-si-ess-e. father-hon.nom Esther-accneg see-hon-nmlz-foc do-hon-past-inf ‘Father certainly did not see Esther.’

   2. (b) *apeci-kkeyse eysute-lul an(i) po-si-ki-nun an(i) ha-si-ess-e.

   3. (c) *apeci-kkeyse eysute-lul po-si-ki-nun an(i) ha-si-ess-e.

   4. (d) apeci-kkeyse eysute-lul an(i) po-si-ki-nun ha-y-ess-e.

   5. (e) apeci-kkeyse eysute-lul an(i) po-ki-nun ha-si-ess-e.

   6. (f) *apeci-kkeyse eysute-lul an(i) po-ki-nun ha-y-ess-e.

While the pro-predicate ha- replaces the root and the negator obligatorily, the honorific suffix may or may not be part of the replacement. This contrast between negation and the honorific suffix supports the claim that the root and negation form a minimal constituent for ha- replacement, which may further include other suffixes such as honorifics.

Some linguists have suggested that honorification is optional (e.g., Niinuma Reference Niinuma2003). If so, it could be argued that the asymmetrical patterns between negation and honorification in predicate iteration constructions may be due to this optionality. In other words, it might be the optionality of honorification, not the structure advocated in this article, that is responsible for the optionality of the honorific suffix in the data in this section.Footnote ¹¹ However, there are many arguments against this view. First of all, the subject cannot be respected or honorified without the -si- suffix. Only with this suffix (and the honorific case suffix) is the subject honorified by the speaker. Second, when a subject DP contains the honorific nominative case suffix -kkeyse, the predicate must bear the agreeing honorific suffix -si- (see footnote 9). In the case of predicate iteration constructions, at least one of the two iterated predicate copies must bear -si- along with -kkeyse. Otherwise, the sentence is unacceptable. Third, this paper limits itself to those cases where honorification is in fact involved and present. If it is absent in a given expression, there can be no negative-honorific interaction. Therefore, the optionality would be irrelevant even if it were real. Finally, even if honorification were optional, it would be natural and expected to assume that a morphosyntactic element is represented in terms of hierarchical relationships when present. These considerations show that honorification itself is not optional, or that the optionality, if real, is irrelevant in the constructions under discussion.

The asymmetry between negation and the honorific suffix in different predicate iteration constructions suggests that the negator and the root form a smaller constituent in copying or replacing a certain part of the predicate. Honorification is concluded to be outside of this constituent because the honorific suffix is not part of the obligatory copy or replacement pro-predicate. This result accords with (31), and runs counter to the discussion in section 3.1. I will henceforth consider (31) to be the correct structure.

3.4 Root suppletion with small v and [+hon]

Now that the conjugated structure (31) is established, the mechanism of how suppletion/allomorphy works in this structure must be determined and explained. For this, consider suppletion of the roots ‘know’ and ‘exist’ with v and [+hon]. In the case of [[know]-[_v+caus]]-[+hon]-, the hierarchical structure is straightforward, and [know] does not show root allomorphy in the environment of the causative suffix, honorification or both.

Consider now the more complex case of [exist]. The honorific agentive kyey- has the [_v+agent] node, as does the non-honorific agentive iss- discussed in section 2 and in I. Chung (Reference Chung2007b). Significantly, the root [exist] is realized as the honorific suppletive allomorph kyey- even with v intervening between the root and the honorific node. If so, the v⁰ node (with the [+agent] feature) will have to be ignored (or transparent) for the purpose of vocabulary insertion of [exist] in the presence of [+hon]. That is, this intervening [_v+agent] behaves as if it were not present between the root and the [+hon] node. This transparency of v is confirmed by other agentive verbs showing honorific suppletion such as ‘eat’: [eat]-[_v+agent]-[+hon]- is realized as capswu-Ø-si- (with the null exponent for [+agent]), and not as mek-u-si- (with the phonologically conditioned epenthetic vowel -u-). The following structure shows the transparent behavior of the v node with respect to honorific allomorphy.

1. (42) Structure of[exist]-[_v+agent]-[+hon] and vocabulary insertion of the root

   

Finally, consider [+neg]-[exist]-[_v+agent]-[+hon]-[+pres]-. The entire sequence is realized as an(i) kyey-si-n-, and not as *an(i) iss-u-si-n-, as shown in the following example:

1. (43) Agentive kyey- occurring with an(i) and -si-

   1. (a) apeci-kkeyse ilyoil-ey-to cip-ey an(i) kyey-si-n-ta. father-hon.nom Sunday-temp-also home-locneg exist.hon-hon-pres-decl ‘Father does not stay home on Sundays (honorific).’

   2. (b) apeci-kkeyse pang-ey an(i) kyey-si-n-ta(-ko) ha-si-ess-ta. father-hon.nom room-locneg exist.hon-hon-pres-decl(-rprt) say-hon-past-decl ‘Father said that (he) would not stay in the room (honorific in both clauses).’

In the case of [+neg]-[exist]-[_v+agent]-[+pres]- without [+hon], the root is realized as the normal allomorph iss- without showing negative suppletion (i.e., an(i) iss-nun-). This fact has suggested, in section 2.3, that the v⁰ node is closer to the root than the Neg⁰ node is, i.e., [Neg [_v V v]]. When the [+hon] node is added, the following structure results:

1. (44) Morphosyntactic structure of[+neg]-[exist]-[_v+agent]-[+hon]-

   

The intervening v and [+neg] are both transparent with respect to honorific root suppletion in the environment of [+hon].

In order for the cyclic application of vocabulary insertion to yield the phonological form kyey-Ø-si- in (42) and an(i) kyey-Ø-si- in (44), the vocabulary item (29b) will have to be revised as follows, adopting the c-command view of contextual allomorphy (Halle & Marantz Reference Halle, Marantz, Hale and Keyser1993, Bobaljik Reference Bobaljik, Grohmann and Struijke2000).Footnote ¹²

1. (45) Vocabulary item for[exist] (revision of (29b))

   [exist]↔/kyey/ when c-commanded by [+hon]

In structures (42) and (44), the root is c-commanded by the [+hon] feature, and hence allomorphy of the root is possible even if the v node intervenes between the root and its allomorphy-conditioning [+hon] feature. Bobaljik (Reference Bobaljik, Grohmann and Struijke2000) notes that the conditioning factor for allomorphy of a certain morpheme can be non-local, i.e., farther away from that morpheme than the immediate next outer cycle in a given word. Honorification in Korean shows that, indeed, (outwards-sensitive) contextual allomorphy is not always strictly local (Halle & Marantz Reference Halle, Marantz, Hale and Keyser1993, Embick & Noyer Reference Embick and Noyer2001; contra Siegel Reference Siegel1977, Allen Reference Allen1978, Simpson & Withgott Reference Simpson, Withgott and Borer1986).

Section 4 will propose an integrated explanation of the two suppletion phenomena and blocking of the inner negative suppletion by the outer honorific suppletion. Before that, I will briefly discuss predicate iteration constructions a little more in section 3.5.

3.5 Excursus: predicate iteration constructions with suppletive roots

This section considers suppletive roots in two of the predicate iteration constructions. The negative, honorific and negative-honorific suppletives show the following behaviors:

1. (46) Echoed verb construction: suppletive negative, non-suppletive honorific

   1. (a) ailin-i eysute-lul molu-ki-nun moll-a. Irene-nom Esther-acc not.know-nmlz-foc not.know-inf ‘Irene certainly doesn't know Esther.’

   2. (b) *ailin-i eysute-lul al-ki-nun moll-a.

   3. (c) *ailin-i eysute-lul molu-ki-nun al-a.

   4. (d) apeci-kkeyse eysute-lul a-si-ki-nun a-si-e. father-hon.nom Esther-acc know-hon-nmlz-foc know-hon-inf ‘Father certainly knows Esther.’

   5. (e) apeci-kkeyse eysute-lul al-ki-nun a-si-e.

   6. (f) (?)apeci-kkeyse eysute-lul a-si-ki-nun al-a.

1. (47) Echoed verb construction: non-suppletive negative, suppletive honorific

   1. (a) ailin-i an(i) ca-ki-nun an(i) ca-a. Irene-nomneg sleep-nmlz-focneg sleep-inf ‘Irene certainly isn't sleeping.’

   2. (b) *ailin-i ca-ki-nun an(i) ca-a.

   3. (c) *ailin-i an(i) ca-ki-nun ca-a.

   4. (d) apeci-kkeyse cwumu-si-ki-nun cwumu-si-e. father-hon.nom sleep.hon-hon-nmlz-foc sleep.hon-hon-inf ‘Father is certainly sleeping.’

   5. (e) *apeci-kkeyse ca-ki-nun cwumu-si-e.

   6. (f) *apeci-kkeyse cwumu-si-ki-nun ca-a.

1. (48) Echoed verb construction: suppletive negative, suppletive honorific

   1. (a) apeci-kkeyse pang-ey an(i) kyey-si-ki-nun an(i) kyey-si-e. father-hon.nom room-locneg exist.hon-hon-nmlz-focneg exist.hon-hon-inf ‘Father certainly isn't (staying) in the room.’

   2. (b) *apeci-kkeyse pang-ey an(i) kyey-si-ki-nun kyey-si-e.

   3. (c) *apeci-kkeyse pang-ey kyey-si-ki-nun an(i) kyey-si-e.

   4. (d) *apeci-kkeyse pang-ey eps-ki-nun/an(i) iss-ki-nun an(i) kyey-si-e.

   5. (e) *apeci-kkeyse pang-ey an(i) kyey-si-ki-nun eps-e/an(i) iss-e.

In the echoed verb construction, as shown above, suppletive negation shows the same pattern as non-suppletive negation, while suppletive honorification does not pattern with non-suppletive honorification. With suppletive honorific roots, both copies of the predicate root must appear as the suppletive exponent, along with the honorific suffix -si-.

It appears that an independent identity condition on the two copies of an iterated predicate is needed. That is, the morphophonological form of the two copies of the root must be identical. In addition, the honorific suffix must appear on both copies because each copy needs the [+hon] feature within the word which motivates the suppletive exponent. That the identity condition is not a surface phonological condition is shown by sentences (46d–f). The general phonological process mentioned briefly in section 3.2 deletes a morpheme-final liquid consonant before a morpheme-initial coronal non-stop consonant in morphemes of a certain kind within a word, as in a-si-e < al-si-e ‘know-hon-inf’ (cf. al-a ‘know-inf’), a-n-ta < al-n-ta ‘know-pres-decl’ (cf. al-ki ‘know-nmlz’), tta-nim < ttal-nim ‘daughter-honorific.title’ (cf. ttal-man ‘daughter-only’) and so-namu <sol-namu ‘pine-tree’ (cf. sol-pangwul ‘pine-cone’). This phonological process and its results do not affect the identity condition, showing that this condition is not a purely phonological condition but a (somewhat deeper) morphological condition.

The identity condition, however, does not hold for the ha- focus construction. (The optionality of negation is not considered in the following examples.)

1. (49) Ha- focus construction: non-suppletive negative, suppletive honorific

   1. (a) apeci-kkeyse cwumu-si-ki-nun ha-si-ess-e. father-hon.nom sleep.hon-hon-nmlz-foc do-hon-past-inf ‘Father certainly slept.’

   2. (b) apeci-kkeyse cwumu-si-ki-nun ha-y-ess-e.

   3. (c) (?)apeci-kkeyse ca-ki-nun ha-si-ess-e.

1. (50) Ha- focus construction: suppletive negative, suppletive honorific

   1. (a) apeci-kkeyse pang-ey an(i) kyey-si-ki-nun ha-si-ess-e. father-hon.nom room-locneg exist.hon-hon-nmlz-foc exist.hon-hon-past-inf ‘Father certainly wasn't (staying) in the room.’

   2. (b) apeci-kkeyse pang-ey an(i) kyey-si-ki-nun ha-y-ess-e.

   3. (c) ?apeci-kkeyse pang-ey eps-ki-nun/an(i) iss-ki-nun ha-si-ess-e.

   4. (d) *apeci-kkeyse pang-ey eps-ki-nun/an(i) iss-ki-nun ha-y-ess-e.

By definition, the pro-predicate ha- does not and cannot have the same exponent as the original predicate (except if the original predicate happens to be the lexical predicate ha- ‘do’ or the light predicate ha- ‘do/be (so)’). Therefore, the ha- focus construction, in which the identity condition does not hold, shows the real optionality of the honorific suffix -si- in the predicate iteration constructions.

In brief, the discussion suggests that the output of the (morpho)syntactic predicate iteration constructions are subject to a morphological identity condition. But, abstracting away from this condition, the optionality of the honorific suffix is maintained in predicate iteration constructions in contrast to the obligatoriness of the negator.

This section has considered subject honorification and the paradoxical situation for ‘exist’ and ‘know’ in the environment of both negation and honorification. The predicate iteration constructions support the structure with negation closer to the root. The honorific suffix can cause root allomorphy even with negation intervening between the allomorphy-triggering feature and the allomorphy-exhibiting root. Furthermore, negation is not the unique intervening, transparent element. The agentive v node can also appear between the root and the honorific suffix triggering honorific allomorphy of the root. The identity of the intervening node(s) between the root and the honorific suffix is thus in principle arbitrary; it is because of this arbitrariness that a c-command-based allomorphy rule was employed for honorific root suppletion. Based on this allomorphy, section 4 will propose an integrated explanation of two cases of suppletion and suppletion blocking.

4.1 Blocking interaction of negative suppletion and honorific suppletion

As mentioned in the previous section, it has been argued in the Distributed Morphology literature (Halle & Marantz Reference Halle, Marantz, Hale and Keyser1993; Bobaljik Reference Bobaljik, Grohmann and Struijke2000, Reference Bobaljik2007 among others) that contextual allomorphy only requires head-internal c-command (see footnote 12). In all of the structures in question – (42) with [[V v] Hon], (44) with [[Neg [V v]] Hon], and (31) with [[Neg V] Hon] – the [+hon] feature c-commands the root whether or not anything intervenes and regardless of what the intervening nodes are. All that matters is that [+hon] c-commands the root; if so, the root can show suppletive root allomorphy for honorification. Hence, allomorphy of the root is possible even with the intervening [+neg] feature. The vocabulary item (45) formulated for [exist] in section 3.4 fits into this picture perfectly without any modification:

1. (1) Vocabulary item for[exist] in the environment of[+hon]

   [exist]↔/kyey/ when c-commanded by [+hon]

The c-command approach ensures this non-local property of contextual allomorphy without any additional machinery and stipulation of transparency of certain features.

Furthermore, on this approach the problem of locality of the root and the Neg node does not arise. To allow the fusion process, the Neg-V-Hon sequence should be structured as [[Neg [V]] Hon], where the [+neg] feature is positioned structurally between the root and the [+hon] feature. This structure is not only necessary for fusion, but is also maintainable under the assumption that the outer feature [+hon] can trigger its root allomorphy under c-command.

To implement this idea, an important difference between suppletive negation and suppletive honorifics needs to be highlighted: all the honorific cases involve suppletion of a root in the environment of honorification, but the entire structure is preserved. Most importantly, the honorific morpheme per se continues to surface as the invariant -si-. Therefore, honorific suppletion is contextual allomorphy. On the other hand, negative suppletion involves fusion, i.e., a portmanteau form is created.

Negative suppletion is thus local, requiring a strict sisterhood configuration, while honorific suppletion merely requires a c-commanding head. However, negative suppletion is bled by honorific suppletion. This effect can be obtained without stipulation if contextual allomorphy and fusion are treated in distinct ways. The root can show suppletive allomorphy for honorification whether or not [+neg] or anything else intervenes, simply because [+hon] c-commands the root. On the other hand, fusion operates strictly locally. Also, it is triggered crucially by the higher head (in this case, the higher Neg node containing only [+neg] and the root). If fusion is allowed to make reference to post-insertion vocabulary items of the root node rather than to pre-insertion abstract ‘morphemes’, i.e., if fusion is interleaved with vocabulary insertion, the bleeding effect will follow automatically. The fusion rule is then to be reformulated as follows:Footnote ¹³

1. (51) Revised fusion of[+neg] and /iss/ to /eps/ (revision of (19))

   

If the rule is part of the cycle with the root and the [+neg] node, the root node must be stated as /iss/, not as [exist], because vocabulary insertion will already have applied in the inner cycle.

The reformulated fusion rule makes reference to a root node with a phonological feature and a node in the second cycle with a formal feature. Even though these two features are quite different kinds of objects, fusing these two nodes is a necessary consequence of resolving the paradoxical situation. Further, this very type of fusion has been proposed in other recent work. Dealing with verbal repetition constructions in Nupe, Kandybowicz (Reference Kandybowicz, Scheffler, Tauberer, Eilam and Mayol2007) proposes the post-insertion fusion of a low tone L and a verbal root √ as shown below, where Foc(P) represents focus (phrase).

1. (52) Fusion of a verbal root and a focus low tone in Nupe (Kandybowicz Reference Kandybowicz, Scheffler, Tauberer, Eilam and Mayol2007: 93)

   

One of the two fused nodes, √, involves category information which is a morphosyntactic feature, and the other node involves a prosodic element which is a phonological feature. The situation with fusion in Nupe is exactly parallel to fusion in Korean in that both cases fuse a phonological feature and an abstract nonphonological feature.

With the reformulated fusion rule (51), cyclic vocabulary insertion will apply to the morphosyntactic representation [[Neg [exist]] Hon] as follows:

1. (53) Vocabulary insertion of[+neg]-[exist]-[+hon]- (partial structure of the complex C ⁰)

   

Note that the morphosyntactic structure (53a) – more precisely, the entire C⁰ containing (53a) – enters PF as a whole after the overt syntactic component. Syntax constructs the entire structure, and this structure, still lacking phonological features, is sent to the morphological component (within the larger PF component). The relevant nonphonological features in the above structure are all present before vocabulary insertion. Therefore, the third cycle with the [+hon] feature is already there at the time of vocabulary insertion in the root cycle.

In (53b), the root is provided with /kyey/ due to the vocabulary item (45) because [+hon] c-commands this node. As a result, the choice of the root allomorph (i.e., honorific suppletion) must be determined in the root cycle. Therefore, the proper domain of vocabulary insertion for the root cycle is that very cycle itself, i.e., not including the allomorphy-triggering [+hon] feature. The c-commanding [+hon] feature simply serves as the context for the root allomorphy operating in the first cycle. Even though [+hon] is required for the choice of the suppletive exponent /kyey/, vocabulary insertion in the root cycle does not involve the (lowest) Hon node (see Halle & Marantz Reference Halle, Marantz, Hale and Keyser1993, Bobaljik Reference Bobaljik, Grohmann and Struijke2000). However, the triggering morphosyntactic feature can ‘penetrate’ into negation (and any other intervening nodes such as v, as discussed in section 3.4), because this is contextual allomorphy. Subsequently, the [+hon] node will be provided with /si/ in (53c) as usual, and vocabulary insertion in this step yields the correct form an(i) kyey-si-.

The negative suppletion, however, crucially takes place in the second cycle. This fusion operation makes reference to the output of vocabulary insertion in the first cycle. Therefore, if the output of the first cycle is /kyey/ as in (53b), the negation fusion rule will not apply. On the other hand, when the [+hon] feature is not involved in a given word, fusion applies to [+neg] in the next outer cycle along with the root item /iss/. This is because vocabulary insertion converts the root morpheme into /iss/ in the first cycle and the rule description is met only in the next cycle. This situation is illustrated in the following, where [+hon] is absent:

1. (54) Vocabulary insertion of [+neg]-[exist]-T- with negation fusion

   

With the reformulation of the fusion rule, the analysis presented in section 2.3 remains almost the same. In particular, if the fusion operation is restricted to sisters, a null v in a negative clause will still block the fusion rule, giving the agentive iss-. When the [_v+agent] feature is present, it is the sister of the root and causes the [+neg] feature to c-command the root asymmetrically. The only change in the new fusion rule is that the root is now represented as a vocabulary item instead of an abstract morpheme, i.e., the nonphonological feature bundle [exist]. Consequently, the fusion rule finds its place accordingly. It does not apply before vocabulary insertion, but applies cyclically, interleaved with vocabulary insertion.

There have been other alternative proposals to Halle & Marantz's (Reference Halle, Marantz, Hale and Keyser1993) original postsyntactic, pre-insertion fusion. As noted earlier in this section, Kandybowicz (Reference Kandybowicz, Scheffler, Tauberer, Eilam and Mayol2007) recognizes fusion operating after vocabulary insertion. Trommer (Reference Trommer1999) proposes that fusion (and other morphological operations) should be described as (part of) vocabulary insertion. These alternative analyses are in harmony with the present reformulation of fusion.Footnote ¹⁴ More broadly, other morphological operations have also been argued to take place after vocabulary insertion. Embick & Noyer (Reference Embick and Noyer2001) discuss post-insertion local dislocation, a morphological merger, in Lithuanian and Huave. Schütze (Reference Schütze and Carnie1994) argues for post-insertion prosodic inversion of certain clitics in Serbo-Croatian. These studies and the analysis in this paper support the possible occurrence of morphological operations after vocabulary insertion. Along with the c-command analysis of contextual allomorphy (for honorific suppletion) and the revised fusion analysis of a portmanteau suppletion (for negative suppletion), the discussion in this section confirms the conjugation structure (44) in Korean. Negation is structured closer to the root than the honorific suffix is.

4.2 Negative suppletion is not contextual allomorphy

It could be proposed that suppletive negation be treated as contextual allomorphy (e.g., Trommer Reference Trommer1999). Specifically, it might be that the root node and the [+neg] node undergo vocabulary insertion separately and that [+neg] is realized as a phonologically null form with a suppletive root allomorph for [exist] and [know]. This section considers this scenario and argues that suppletive negative cases in Korean must indeed be treated as fusion, as proposed in this paper, and not as contextual allomorphy.

The structure at issue is that seen in (31), [[[+neg] [exist]] [+hon]], which is realized as an(i) kyey-si-. The vocabulary items for [exist] and [+neg] would look like the following in the contextual allomorphy view:

1. (55) Vocabulary items for[exist]

   1. (a) [exist]↔/kyey/ when c-commanded by [+hon] (=(45))

   2. (b) [exist]↔/eps/ when c-commanded by [+neg]

   3. (c) [exist]↔/iss/ (=(29c))

1. (56) Vocabulary items for[+neg]

   1. (a) [+neg]↔Ø / ____ /eps/

   2. (b) [+neg]↔/an(i)/ (=(30b))

The contextual allomorphy view works for the case of [+neg [exist]] without [+hon] in the given word. The vocabulary item (55b) realizes the root [exist] as /eps/ in such a structure. Then in the next cycle, a zero form is inserted for [+neg] in the environment of the morphophonologically peculiar root form /eps/ due to the vocabulary item (56a). The resulting morphophonological form is the correct Ø-eps-.

Several difficulties arise with this contextual allomorphy analysis. A major problem is the indeterminacy between /kyey/ and /eps/ when both negation and honorification are present in a given conjugated form, as in (31). Both [+neg] and [+hon] c-command the root, according to (55a) and (55b). Furthermore, neither of the two vocabulary items' environments, i.e., ‘c-commanded by [+hon]’ and ‘c-commanded by [+neg]’, is a subset of the other, and hence they are not in a competition relationship. Therefore, an extrinsic ordering will be necessary. At the same time, it would be reasonable and plausible to assume that the allomorphy-triggering feature closer to the root, in this case [+neg], would win out over an outer allomorphy feature, i.e., [+hon]. However, the reverse is true. This raises a question about the validity of the contextual allomorphy approach to suppletive negation.

Another serious problem with the contextual allomorphy view of negative suppletion is that it wrongly predicts that any intervening node between the root node and the negation node will be ignored, just as in the case of honorific allomorphy. The vocabulary item (55b) wrongly ignores the intervening null v with [+agent] (as in [[+neg] [[exist] [_v+agent]]]) which, in fact, blocks negative suppletion of the root [exist]. This would result in the ungrammatical form *eps-nun-ta ‘not.exist-pres-decl’ for the negative agentive [[[Neg [[exist] v]] T] decl] with the vocabulary item (55b). The contextual allomorphy analysis of suppletive negation thus does not work either conceptually or empirically.

Hence, [+neg] does not serve as the environment of the suppletion phenomenon, but rather is the phenomenon itself. That is, the negative suppletion takes the root and the negation, and provides a single vocabulary item for them. To embrace the two sister nodes, a cycle bigger than the root cycle must be referred to, whereas contextual allomorphy refers only to the relevant cycle itself excluding the allomorphy-triggering feature from the vocabulary insertion operation. In this way, the apparent paradox noted in section 3.2 is resolved.Footnote ¹⁵

4.3 In support of late vocabulary insertion

The paradox between negative and honorific morphology can be resolved by employing two different mechanisms for negative suppletion and for honorific suppletion. However, a crucial assumption needs to be made. In order for the vocabulary item (45) to be inserted in structure (53) with [[[+neg] [exist]] [+hon]], the [+hon] feature should be present in the structure when vocabulary insertion applies to the root cycle. That is, vocabulary insertion of the root must wait at least until [+hon] is added. Otherwise, fusion will still operate: if vocabulary insertion took place in the structure [[+neg] [exist]] prior to the addition of [+hon], the root would be realized as /iss/ due to (55c) and then [[+neg] /iss/] would turn into /eps/ due to the new fusion rule (51). On the other hand, if all three morphemes are present in the given structure at the point of vocabulary insertion, the root will be directly realized as /kyey/ due to [+hon] in the given structure, as the vocabulary item (45) will be chosen in the environment. Therefore, the [+hon] feature must already be present at the time of vocabulary insertion of the root.

This situation provides another argument in favor of late insertion as assumed in Distributed Morphology. Suppose that phonological features and nonphonological features were to be provided at the same time for each morphological node, as in a lexicalist framework. In the case of [[[+neg] [exist]] [+hon]] which is realized as an(i) kyey-si-, morpheme concatenation will proceed in the order: root, negation, and honorifics. In the first cycle, the root node is provided with the nonphonological feature bundle [exist] and the corresponding phonological feature bundle. Critically, the root node by itself cannot determine the correct phonological form. The subseqent cycle containing the root node morpheme and the [+neg]-/an(i)/ pair cannot determine the correct root allomorph (i.e., /kyey/) either, because the [+hon] environment is not present. Instead, the negative suppletive form /eps/ would be yielded for the first two cycles containing negation and [exist]. If the [+hon]-/si/ cycle is added further, then (and only then) the entire structure will turn into an(i) kyey-si- (due to the honorific feature). This lexicalist derivation is illustrated in (57):

1. (57) Cyclic morpheme concatenation of[[[+neg] [exist]] [+hon]] in lexicalism

   

This needlessly complex and undesirable derivation is unavoidable if the phonological features are provided at the same time as the nonphonological features for each cycle.

On the other hand, if phonological features are available later in the derivation after nonphonological features are provided for all the relevant morphological nodes in a certain domain such as word,Footnote ¹⁶ then the entire morphophonological derivation will be coherent and consistent. The root node will make reference to the remote outer nonphonological feature (i.e., [+hon]) when it chooses the root allomorph (i.e., kyey-), as shown in the derivation (53). Determining the correct phonological form of the root in one fell swoop is possible when the environment [+hon] feature (c-commanding the root node) is present at the time of vocabulary insertion of the root node and the root cycle lacks its phonological information. Therefore, verbal suppletion in Korean provides strong support for separation and late insertion as assumed in Distributed Morphology, as opposed to lexicalism.

To summarize this section, the suppletion interaction in Korean can be explained by appealing to two different mechanisms for the different suppletion types. Negative suppletion is characterized as fusion in (51), replacing the post-insertion root and the sister negation feature. This revised fusion formalism is interleaved with vocabulary insertion and operates in the second cycle. Honorific suppletion is formalized as contextual allomorphy in the c-command configuration. With the c-command requirement, the honorific feature acts as the environment for honorific suppletion. Because honorific suppletion, i.e., choosing the honorific suppletive vocabulary item (45), operates crucially in the root cycle, it bleeds negative suppletion operating in the second cycle when the root [exist] has both the negative and the honorific features, resulting in an(i) kyey-si- for [[[+neg] [exist]] [+hon]]. The discussion and analysis presented toward the end of this section support the notion of late vocabulary insertion as assumed in Distributed Morphology.