A morphosyntactic haplology of -i drop in the Korean vocatives

4.1 Proper names: intimate -i

As seen in (1a) and (2a), Korean displays a unique phenomenon concerning the names of people and nominal particles. As is well known, insertion of -i to human proper names is phonologically determined: only closed-syllable names allow the -i insertion.

The examples in (14) show that proper names such as Ce-A and Celin are followed by several structural Case markers, postpositions, and delimiters for non-vocative use (Lee 1985). In (14a), Ce-A contains no coda consonant – it is open-syllabic and thus fails to take -i. Conversely, in (14b), such an insertion is licit with the closed-syllable name, Celin. The -i insertion is optional, as shown in (14b) and (14c). Notably, closed-syllable proper names can take -i for non-vocative use.

Let us now consider vocative counterparts of (14b). The epenthesis of -i is illicit with the vocative marker.

The marked contrast between (14b) and (15) raises the question of why -i insertion into proper names is possible in non-vocatives, as in (14b), and impossible in vocatives, as in (15), as Table 3 summarizes. This observation leads to Generalization (3): in vocatives, the -i morpheme must be absent only when it is string-adjacent to -(y)a.

There is strong empirical evidence supporting that the -i drop in proper names is not solely attributable to phonological factors.

The proper names, Yengmi and Swuci, contain the vowel -i as their own part. Unlike the case of Celin in (14b), the -i vowel of Yengmi and Swuci never drops in the vocatives (16) (Song 2008). If the drop in question is completely determined phonologically, the ill-formed outputs, *Yengm-a and *Swuc-a, should be incorrectly expected to be acceptable. Nevertheless, such impossibility attests to the view that the dropping phenomenon in the vocative constructions has no relation with phonology – it only targets the mysterious -i morpheme, not part of proper name.

Let us turn our attention to the defining characteristics of the -i morpheme in proper names. What is the difference, if any, between proper names without the morpheme, as in (14c), and those with it, as in (14b)? To the best of my knowledge, although sporadically noted in traditional grammar, this dropping phenomenon has received little attention in the generative literature, and there is no consensus in the traditional literature on the exact status of this epenthetic vowel in proper names. Choi (1975) regards the morpheme in question as an “euphonic vowel” (as cited in Ko 1968: 18) because it is inserted for smooth pronunciations. Similarly, Heo (1975: 38–40) labels -i a “pseudo-affix.” He posits that it has “no grammatical or lexical meaning” and only has the (non-semantico-syntactic) function of “rendering the sound smooth” [translations, author]. An (1977: 70) offers a similar analysis to Heo’s, stating that -i has no lexical meaning and simply attaches to proper names. Notably, Lee (1985) proposes that -i is a “suffix of phyengching [plain reference] for human proper names.” He focuses on the contrast between conching (honorific reference) and phyengching (non-honorific reference). Among various Korean speech styles, the non-honorific style features an intimate, affective, informal, or colloquial function (Brown 2015; Sohn 1999). Thus, the alternation between (14b) and (14c) with respect to -i insertion is far from free. As Lee (1985: 78) demonstrates, the following examples corroborate this line of analysis of ‘-i as an intimacy/affection morpheme’ (see also Sohn 1999).

Let us consider the following examples. Contrary to the case of first name alone, as in (14b), full name is usually considered more formal and honorific.

The degraded grammaticality of (17b) is straightforwardly explained by the mismatch between the strict formality of full names and the intimate and non-honorific nature of the -i morpheme. We can now fill in the blank from Table 2 in Section 1, as shown in Table 4 below. Note that -i on proper names has an intimate, informal, and/or non-honorific effect; thus, it is most natural and acceptable when it occurs with first names.

Let us recall from Section 3 that Middle Korean employed honorific and non-honorific vocative markers, and the latter is involved with the level of intimacy between the interlocutors.

As the non-honorificity of the marker under consideration is linked to intimacy, it is reasonable to extend this link to intimacy of the similar sort observed in proper names, seen in (14b). Considering the syntactic status of the -i morpheme in proper names, I contend that the morpheme at stake serves as an exponent of ϕ-features. I take ϕ-features – which are typically related to person, number, and gender – to be those involved in honorification. Alok and Baker (2022: 3) maintain that formal features of honorification, such as [±HON], have the “status of full-fledged phi-features” (Boeckx and Niinuma 2004; Choi 2010; Harbor et al. 2008; Kim 2007, 2012, 2017; Miyagawa 2010, 2017). Following this line of analysis and based on (14b) and (15), I posit that the ϕ-features of intimate -i are associated with non-honorificity or intimacy, notated as ϕ[−HON: intimate]. Furthermore, following Heim and Kratzer (1998: 244), in which ϕ-features are nodes of their own and each of them is adjoined to a DP, as illustrated in (18a), I propose that the ϕ[−HON: intimate] feature of -i projects in the syntax, as shown in (18b).

(18)

​

4.2 -i-derived animal nouns: animate -i

The second type of the -i morpheme can merge with onomatopoeic words (OWs) and attribute bound morphemes. It is a nominalizer because it derives non-nominal elements and roots into nouns.

In (19a), OWs are adverbs that imitate the sound they describe. In (19b), combining with -i, they are derived into nouns and come to refer to animals. Notably, without the -i, they simply remain their onomatopoetic interpretation. In a nutshell, -i converts inanimates into animates. I refer to it as an animator (Song 1992: 134–137 in which the morpheme is referred to as a yuceng myengsawhaso [animate nominalizer]). We can now update Table 4, as follows in Table 5. Notably, the -i morpheme is a nominalizer in morphosyntax and an animator in semantics (see also Choi 2011).

This line of analysis can carry over to attribute bound roots.

In semantics, as noted above, attribute bound morphemes denote abstract traits or characteristics of people or animals; thus, I gloss them in small capitals. For example, in (20a), ttolttol- on its own does not mean a smart person or anything. It only refers to a characteristic by which a person can be defined. In (20b), merging with the -i morpheme, they obtain the meaning of animates (people and animals). In morphosyntax, they are syntactic category-neutral or ‘acategorial’ roots in the sense of Distributed Morphology (Embick and Marantz 2008; Hale and Marantz 1993; Marantz 1997). Moreover, combining with the nominalizer -i, they become nouns of free morphemes. In the article, I restrict myself to OWs in (19), and the analysis for them can carry over to the attribute bound morphemes in (20).

Let us now consider the formal feature that the animator -i bears. Following Kim 2007, 2012, 2017 for Korean and Harley and Ritter 2002, in which animacy is a subcategory of gender (“class” in Harley and Ritter 2002: 486), I propose that animator -i bears the feature ϕ[animate], as illustrated in (21).

(21)

​

I assume that OWs are acategorial roots, and the animator resides in the n head position.

Let us now consider animacy and gender in more detail. By showing “how divisions into animate and inanimate, or human and non-human, function in language exactly as does the division into female and male,” Corbett (1991: 3–5) maintains that “the determining criterion of gender is agreement […] Since agreement is taken as the criterion for gender, there are no grounds for drawing a distinction between languages in which nouns are divided into groups according to sex, and those where human/non-human or animate/inanimate are the criteria.” According to this analysis, animacy is considered a subcategory of gender.

Furthermore, a diachronic fact in Korean empirically supports such a unified view of animacy and gender. Modern Korean contains three dative particles: -ey, -eykey, and -kkey. They are distinguished by animacy and honorificity toward goal DPs. In (22), -ey marks an inanimate goal such as ‘school,’ -eykey indicates a non-honorific animate goal such as ‘student,’ and -kkey specifies an honorific goal such as ‘teacher.’

Pak (1975) demonstrates that the Modern Korean datives, -eykey and -kkey, are originally derived, as shown in (23), from the combination of [(X)-genitive+‘that’+‘at’] – something like ‘at that of X.’ In his analysis of the diachronic origin of the honorific dative -kkey, An (1968: 344) contends that Middle Korean utilized two genitive markers, -ui and -s, with a division of labor based on animacy. -ui merges with animate possessors and -s combines with both inanimate and honorific possessors. In (23a), the animate dative -eykey originally contained the animate genitive -ui; however, in (23b), the honorific dative -kkey is derived from the inanimate genitive -s.

Thus, the honorific dative -kkey can be characterized as the inanimate counterpart of animate dative -eykey. This leads Pak (1975: 90) to “regard honorific nouns as [grammatically] a subclass of inanimate nouns” and argue that the way to honorificate someone is the process of making them grammatically inanimate. Thus, honorification is considered to be “yucengmyengsaui mwucengmyengsahwa” [inanimate nominalization of animate nouns, translation author].

Acknowledging this line of analysis of ‘honorification as conversion of animates (humans) into inanimates,’ Kim (2007, 2012 offers a generative account for the alternation of the three datives and subject honorification, and he develops a theory of “animacy as a gender ϕ-feature.” Overall, there is a convincing reason (at least in Korean) to treat animacy as gender and subsume the honorificity associated with animacy under (morphosyntactic) agreement. Note that animacy is considered a gender feature in Korean, and the -i animator bears the ϕ[animate] feature.

Let us consider the contrast shown in (1b) and (2b). The -i animator displays specific behavior when merged with nominal particles, which behave differently from vocative and non-vocative markers.

When used non-vocatively in (24a) and (24b) – merging with Case particles, postpositions, delimiters, and the plural marker -tul – the OWs must have an animator to obtain the intended animal meaning. Conversely, when used vocatively in (24c) – combining with the vocative marker – the -i-derived animal nouns must drop the -i. Significantly, the animal denotation is available in the vocatives even without the animator, contrary to the non-vocative use. As Table 6 sums up, for the animal interpretation, the otherwise present -i must be absent only when followed by -(y)a.

Let us consider another significant set of data regarding the -i drop in vocatives, as shown in (25):

Unlike the singular vocatives in (24c), the animator must be present in their plural counterparts, as exemplified in (25). This observation implies that the drop of -i is not absolute in the vocatives. The distinction between singular vocatives (24c) and plural vocatives (25) lies in the linear arrangement of -i: -i is directly adjacent to -(y)a in the former but not in the latter. The plural marker -tul intervenes between the animator and the vocative marker, indicating an adjacency effect on the -i drop.

Another vocative context avouches that the adjacency effect is at play in the dropping phenomenon, as shown in (26):

In co-ordinate vocatives (26), each conjunct exhibits distinct behavior regarding the -i drop. It is correctly expected that, as in (24c), yaong in the second conjunct, denoting ‘cat,’ cannot carry the animator -i, because it is adjacent to -(y)a. Conversely, kaykwul in the first conjunct is obliged to retain -i, even in the vocative, because it is not adjacent to -(y)a, but to the conjunction particle -wa.

5.1 Explanandum and explanans

As a point of departure, let us first consider the intimate -i morpheme in proper names, as shown below.

As Table 7 depicts, the marked contrast lies in the fact that the intimate -i is licit with the non-vocatives, as in (27a), whereas it is illicit with the vocatives, as in (27b).

One key role of the -i morpheme in proper names is its semantico-pragmatic effect, indicating a close connection between discourse participants. In Section 4.1, I posit that this morpheme carries ϕ-features associated with non-honorificity, annotated as ϕ[−HON: intimate], as depicted in (28).

(28)

a.

​

b.

Vocabulary Items for ϕ[−HON: intimate]

ϕ[−HON: intimate] ⟷ -i elsewhere

ϕ[−HON: intimate] ⟷ ∅/___⌒-(y)a[4]

In (28), the identical ϕ-features, ϕ[−HON: intimate], of the intimate -i and vocative -(y)a match each other. Constraint (4a), as repeated here, now reveals Generalization (3).

Two crucial aspects in (4a) include (i) the matching ϕ[−HON: intimate] feature of -i and -(y)a and (ii) their linear adjacency. This specific featural configuration is reminiscent of the precise adjacency effect observed in haplology (Nevins 2012). According to Constraint (4a), the occurrence of -i is hindered when the morpheme is immediately adjacent to -(y)a due to the ϕ[−HON: intimate] feature in both morphemes, as illustrated in (28a). Matching is interpreted as being identical. Consequently, as given in the Vocabulary Items in (28b), the node for ϕ[−HON: intimate] is not spelled out.

Let us now move on to the animator -i. As observed in Section 4.2, the morpheme under discussion exhibits the same pattern as the intimate morpheme in terms of -i drop in vocatives, as shown in (29) (Table 8).

To denote an animal, the morpheme at stake must be present in the non-vocatives, as in (29a). Conversely, the absence of -i is necessary in vocatives in (29b), and the intended interpretation of an animal is attainable. Therefore, the primary focus revolves around understanding how OWs achieve the denotation of animals even in vocatives with no overt animator. The present analysis accounts for the fact as follows:

(30)

a.

​

b.

Vocabulary Items for ϕ[−HON: animate]

ϕ[animate] ⟷ -i elsewhere

ϕ[animate] ⟷ ∅/___⌒-(y)a

^[5]In (30), the -i and -(y)a bear distinct types of ϕ-features. Specifically, the ϕ[animate] feature of the former pertains to gender (Harley and Ritter 2002; Kim 2007, 2012, 2017) whereas the ϕ[2p] feature of -(y)a corresponds to a person feature. Consequently, the ϕ-features of these morphemes differ in content and feature type, with -i representing gender and -(y)a representing person. Given this disparity in ϕ-feature types, one should expect lack of featural matching between ϕ[animate] and ϕ[2p].

To resolve this conundrum, I investigate ϕ-feature types and their interrelations. The ϕ[2p] feature of -(y)a is initially proposed to capture the addressee semantics of the vocative, specifically responsible for the anthropomorphic reading of inanimates (Yim 2021 for Korean). The first and second person represent authentic discourse participants inherently characterized as human and animate. Conversely, an inanimate entity, by default, cannot partake in discourse as it lacks human attributes. Hence, the speaker and addressee are invariably animate (see Janson 2013 for addressee semantics and animacy). As the [1p] and [2p] features imply that the elements containing them are animate and human, it logically follows that the ϕ[2p] feature of -(y)a entails the ϕ[animate] feature of -i: ϕ[2p] ⊨ ϕ[animate].

My conjecture of ϕ-feature types and their entailment relation makes a clear prediction regarding the element left unrealized in the context of haplology. Notably, the entailment relation is unidirectional, with the ϕ[2p] feature implying the ϕ[animate] feature but not the other way around (ϕ[animate] ⊭ ϕ[2p]). An animate entity can potentially belong to the first, second, or third person. Therefore, the one-way nature of the entailment relation may offer a plausible explanation for the avoidance strategy in the -i drop – wherein, within haplological contexts, the question arises whether to avoid the -i morpheme or the vocative -(y)a; consistently, -i is omitted, not -(y)a.

The rationale behind this dropping pattern is attributed to the unidirectionality of entailment. Specifically, the unidirectional entailment from the [2p] feature to the [animate] feature leads to a consistent drop of -i. This pattern can be understood in terms of recoverability. A feature entailed by another can be deleted because it can be recovered through the entailment relation between the two (see Ackema and Neeleman 2004; Martinović 2017; Nevins 2012 for impoverishment under recoverability). As ϕ[2p] ⊨ ϕ[animate], the latter becomes a recoverable feature from the former. Consequently, as in Vocabulary Items in (30b), the -i is not exponed. The consistent omission of -i is recoverable because the ϕ[animate] feature of -i can be predicted from the ϕ[2p] feature of -(y)a. Therefore, in (30), the entailment relation between the second person and gender allows them to be considered as “matching.” Thus, matching in Constraint (4a) includes identical ϕ-features and non-distinct ϕ-features under entailment.

Let us now consider the following example of plural vocatives like the ones from Section 4.2.

In plural vocatives (31), unlike singular vocatives, it is not possible to drop the animator morpheme. This discrepancy arises because, in the plural vocative, -i is not string-adjacent to -(y)a, a condition that sharply contrasts with the singular vocative depicted in (29b). The present analysis predicts the grammaticality of (31): the otherwise unacceptable -i-ya sequence becomes well-formed as the intervening plural marker separates these elements, as illustrated in (32).

(32)

​

In (32), the Korean plural marker, -tul, is a nP modifier (Kim and Melchin 2015; Kim and Park 2024; Wiltschko 2008). In this structure, the animator morpheme is not linearly adjacent to the vocative marker because of the intervening plural marker. Thus, the intervention thus bleeds the dropping. As such, Constraint (4a) does not apply, yielding the spell-out of -i in (31).

Let us now consider coordination constructions, exemplified by the Korean one shown in (26). The constructions serve as a means to ascertain whether morphological dissimilation such as the -i drop is subject to linear adjacency rather than syntactic hierarchy such as c-command. Ortmann and Popescu (2000) present a clear haplology case from Romanian possessives in which deletion occurs only in a single conjunct rather than in the entire conjunct, as exemplified in (33).

In (33a), the possessive marker a occurs in both conjuncts when combined with the indefinite noun o casă. In (33b), however, it appears only in the conjunct not adjacent to the definite noun cas-a – specifically, the second conjunct. Such selective realization of the a marker in the coordinate structure, under a purely syntactic analysis equipped with c-command, would affect the entirety of the structure at stake rather than just a portion of it, contrary to fact. This fact from the typologically unrelated language (Romanian) clearly attests to the conclusion that the haplological dissimilation in question is conditioned by string adjacency rather than structural hierarchy.

Let us now examine an example of co-ordinate vocatives from Korean.

In co-ordinate vocatives, similar to the Romanian example in (33), there is a selective manifestation of a morpheme: the animator -i is licit in the preceding conjunct and is illicit in the following conjunct. This selective spell-out is correctly predicted under the present analysis because -i is not string-adjacent to -a in the initial conjunct, but so is it in the subsequent conjunct.

Furthermore, according to the parallelism condition of coordination structure, both conjuncts, α and β, must match in structural size, as illustrated in (35).

Despite one conjunct requiring the -i morpheme and the other rejecting it, the parallelism condition demands equivalence in structural size between α and β. This means that since the α constituent must include the -i, the β constituent must maintain the same structural size, despite apparently lacking the morpheme. As Constraint (4a) operates as a deletion rule in the post-syntactic component, the parallelism condition is met in the syntax proper. This indicates that the -i drop is not the removal of a syntactic node, but rather constitutes a case of surface non-realization (cf. Song 2008).

5.2 Nevins’s (2012) modular model for haplology

My analysis of the -i drop in Korean vocatives treated as an example of haplology finds its primary support in the defining characteristics outlined by Neeleman and Van de Koot (2008). The salient attributes of haplology are as follows. First, the phenomenon in question is accidental, not intentional like reduplication. Second, it is triggered by morphosyntactic features such as ϕ-features. Third, the -i drop refers to strict identity. Therefore, the drop phenomenon is a true example of haplology (see Neeleman and Van de Koot 2008 for more details).

Presenting various types of haplology from diverse languages, Nevins (2012: 116) claims that “[t]he architecture of haplology is thus best understood in terms of modular components,” further proposing a strict division of labor between syntax, prosody, and morphology, and a rigid ordering between them: syntax, followed by prosody, followed by morphology. As given in (36), Nevins hypothesizes four distinctive stages in the mapping from the syntax to phonology at which haplological constraints are at play. The stages are sequential from Stages 1 to 4.

Considering all the properties of the four stages described in (36), several questions remain. It is unclear why each stage exhibits its own properties. For instance, some stages are subject to adjacency and others are not. Likewise, some stages refer to individual features and some do not. Furthermore, it is unclear why there are four stages, not more or fewer.

Considering the stage to which the -i drop haplology can belong, in (36), Stage 1 is not relevant to the dropping phenomenon, because it refers to macro-level syntactic categories, such as DPs. The dropping phenomenon operates at the feature level, and the target is ϕ-features. Similarly, the haplological phenomenon does not belong to Stage 4, because this stage does not refer to individual (morphosyntactic) features under strict adjacent identity, which is a significant property of the dropping phenomenon. Considering Stages 2 and 3, the shaded shells in Table 9 indicate the properties of the -i drop, and there is a clear overlap between the two stages.

The dropping phenomenon is subject to strict adjacent identity, and Stage 2 exhibits this property. However, this is one and only one matching property between the phenomenon under discussion and the stage. As for Stage 3, its defining properties almost match those of the dropping phenomenon, except for the adjacency property, which is crucial for the Korean case of haplological dissimilation. All in all, the Korean case of haplology does not easily fall under one particular stage (Nevins (2012) mentions challenging cases that also do not accommodate his model), and, therefore, the Korean haplological dissimilation of -i drop undermines Nevins’s theory.