1 Introduction

Numeral classifiers and plural morphemes have been claimed to be in complementary distribution, either between or within languages (e.g., Chierchia 1998; Borer 2005). For instance, Borer (2005) claims that nouns enter the derivation with an undivided mass interpretation, which is incompatible with counting. She proposes that, in order to be counted, the mass must be divided (i.e., individuated) into countable units. This individuating function is performed by the syntactic category Div(ision), which immediately dominates nP in the nominal spine as illustrated in (1) for the English plural marker –s, which realizes Div. In (1), the n head realized by the noun stone undergoes head movement to Div, merging with plural suffix –s. The noun can be counted by the numeral four in #P, as Div turns the noun into countable individuated units, i.e., it divides; thus, we observe the contrast between four stones and *four stone. Under this view, a plural marker in languages like English divides nouns so that they can be counted by a numeral denoting cardinality greater than one.1

    1. (1)

In a classifier language, the classifier realizes Div, and plays the role of dividing the noun into individuated units in order to be counted. This explains the supposed complementary distribution: plural markers and classifiers compete for the same syntactic head, and so if a classifier is present, a plural marker is impossible, and vice versa. Borer (2005: 94) gives the following examples from Armenian as evidence:2

    1. (2)
    1. Armenian
    1.  
    1. a.
    1.  
    1. Yergu
    2. two
    1. had
    2. CL
    1. hovanoc
    2. umbrella
    1. uni-m.
    2. have-1SG
    1. ‘I have two umbrellas.’
    1.  
    1. b.
    1.  
    1. Yergu
    2. two
    1. hovanoc-ner
    2. umbrella-PL
    1. uni-m.
    2. have-1SG
    1. ‘I have two umbrellas.’
    1.  
    1. c.
    1. *
    1. Yergu
    2. two
    1. had
    2. CL
    1. hovanoc-ner
    2. umbrella-PL
    1. uni-m.
    2. have-1SG
    1. ‘I have two umbrellas.’

These examples show that, while it is possible in Armenian to have either a classifier (2a) or a plural marker (2b) in a quantified noun phrase, it is not possible to have both (2c). The structure of (2b) would be analogous to the English structure in (1), with the plural marker realizing Div; the structure of (2a) would be as in (3):

    1. (3)

In essence, Div can be morphologically realized with either a plural marker such as English –s or a classifier such as Armenian had. Consequently, Borer predicts that classifiers and plural markers are in complementary distribution, as there can only be one Div head in a single nominal projection.

Korean is also known as a classifier language (e.g., Kang 1994; Sohn 1999), but it also has a plural marker, –tul (e.g., Lee Han-Gyu 1992; Kang 1994; Im 2000; Baek 2002; Kwak 2003; Jun 2004; Kim Chong-Hyuck 2005). Considering Borer’s (2005) proposal that plural markers and classifiers are in complementary distribution, it is predicted that the plural marker –tul should not co-occur with a classifier in a single nominal phrase in Korean. However, this prediction is not borne out, as shown in (4).3 In (4), –tul appears on a common noun, which is followed by a numeral and classifier (henceforth, a classifier phrase), which is the main data type that this paper will discuss.4,5

    1. (4)
    1. a.
    1. salam(-tul)
    2. human-PL
    1. ney
    2. four
    1. myeng
    2. CL
    1. ‘four people’
    1.  
    1. b.
    1. ai(-tul)
    2. child-PL
    1. sey
    2. three
    1. myeng
    2. CL
    1. ‘three children’

If a classifier in Korean instantiates the Div head as in other classifier languages, the data in (4) raises questions as to the syntactic status of the plural marker –tul in the language: Why is the plural marker able to co-occur with a classifier; does it instantiate a different head from Div? What is the status of –tul in a classifier language like Korean where it can co-occur with a classifier?

This paper addresses these questions, and proposes that the Korean plural marker –tul is not an instantiation of Div, and therefore is not a canonical plural marker, unlike English –s. It is argued that –tul is instead a modifier of the nP projection as shown in (5). We assume that affixal –tul is left-adjoined following the fact that modifiers in the language are generally left-adjoined.6

    1. (5)

Plural –tul in Korean seems to show similarities to as well as some differences from plurals in other East Asian languages such as Chinese or Japanese, as will be noted in this paper where relevant.7 However, it is not the concern of this paper to compare the distribution of –tul to plural markers of other East Asian languages and to propose a comparative analysis of the plural markers (or classifiers) in the East Asian languages. Although we recognize that this task is important in the field, such a task cannot be completed without establishing the syntax of –tul. There are only a few studies that focus on (the syntax of) Korean –tul (e.g., Kang 1994; Im 2000; Baek 2002; Kwak 2003; Jun 2004; Kim Chong-Hyuck 2005; Park 2008). More importantly, currently there is no study that examines the syntax of –tul in the perspective of current syntactic theories on plurals such as Borer (2005), Wiltschko (2008) or Kramer (2015). Thus, our main goals in this paper are to examine the distribution of plural –tul in Korean with respect to the prediction of Borer (2005), and to propose a syntax of –tul that can capture the observed co-occurring distribution with classifiers in the language, mainly adopting the analysis of Wiltschko (2008).

It has been claimed that plural markers can have different syntactic realizations across languages (e.g., Wiltschko 2008; Kramer 2009; 2015). Wiltschko (2008) proposes that, in addition to the familiar plurals that realize heads in the nominal extended projection which take a nominal complement (i.e., Num or Div), there are plurals that enter the derivation as adjuncts to different projections in the nominal spine in different languages, and that the properties of plural markers in a given language vary depending on their realization in that language. The former kind of plurals is referred to by Wiltschko (2008) as head plurals, and the latter as modifying plurals; we employ this terminology in the rest of this paper. Kramer (2009; 2015) provides an analysis of plurality in Amharic which involves two different plurals, one in nP and one in Num(ber)P, with different characteristics. We show that the Korean –tul fits Wiltschko’s criteria for a modifying plural, and Kramer’s properties of a nP plural. Building on this empirical result, we claim that –tul is a modifier of nP, and that it does not realize a head in the nominal spine dedicated to the function of individuation in syntax.8 This consequence suggests that a plural marker that does not instantiate a syntactic head of individuation, such as –tul in Korean, is predicted not to be in complementary distribution with classifiers, which are commonly viewed as instantiating an individuating head (e.g., Cheng & Sybesma 1998; 1999; Li 1999; Borer 2005; Cowper & Hall 2012). Thus, the current paper contributes to clarifying how the complementary distribution of plurals and classifiers noted in numerous studies should be understood: the complementary distribution in this prediction is syntactic, not semantic.

The paper proceeds as follows: in Section 2 we argue that –tul is a modifier. In the following sections we discuss the projection in which this modifier is situated: first, that it is not in RootP (Section 3), and second, that it is situated in nP (Section 4). In Section 5 we show that, contrary to a common claim in the literature, –tul does not mark definiteness, and so it is not in DP. In Section 6 we discuss the consequences of this paper. Section 7 concludes.

2 –Tul as a modifying plural

In this section we show that Korean –tul is a modifying plural, rather than a head plural, in the sense of Wiltschko (2008). The different properties of modifying and head plurals in Wiltschko (2008) are summarized in Table 1.

Modifying plural Head plural

(i) Status Optional Obligatory
(ii) Interpretation Number-neutral vs. plural Singular vs. plural
(iii) Feature value Monovalent [plural] Bivalent [±plural]
(iv) Position RootP, nP, etc. Num

Table 1

Modifying vs. Head plurals (Wiltschko 2008).

Wiltschko (2008) claims that not all pluralizing morphemes are instances of a head in the nominal spine. Instead, they can enter the derivation in two ways: either as a modifying plural or a head plural. These two types of plurals vary in certain properties. The first kind of plural, a head plural, is a head of a category such as Num. Num is the locus of grammatical number, and is thus associated with a number feature that participates syntactic operations such as Agree (Chomsky 2000; 2001), which is detailed in the next section. In that section, building on the proposal of Wiltschko (2008), we express such a number feature on the head, Num, as the bivalent number feature [±plural], as illustrated in (6a). When a nominal is specified for [+plural], it is interpreted as plural; when it is specified for [–plural], it is interpreted as singular. A modifying plural, on the other hand, is not associated with this head, and is claimed to merge as an adjunct modifying some projection in the nominal spine, namely Root(P), n(P), or D(P). To illustrate, consider the root-modifying plural depicted in (6b), as proposed by Wiltschko (2008) for Halkomelem. Unlike a head plural, we express a modifying plural as a monovalent feature value, [plural], based on Wiltschko’s (2008) proposal that such modifying features do not participate in Agree (see Section 2.1). The presence of such a modifier gives a plural reading, but its absence does not necessarily lead to a singular reading. In the absence of other elements (e.g., a numeral) that indicate plurality, a nominal without a modifying plural is interpreted as number-neutral (see Section 2.1).9

(6) a. b.

We illustrate the contrast between head plurals and modifying plurals using Wiltschko’s (2008) example of the modifying plural in Halkomelem, as compared to the head plural in English, and show that Korean –tul shows properties of a modifying plural.

2.1 Halkomelem modifying plural

Being a (root; see Section 3) modifier, the Halkomelem plural marker does not instantiate a grammatical category, which results in certain properties distinguishing it from a head plural. Here we will focus on three properties that result from its status as a modifying plural: it does not take part in obligatory agreement processes; its absence leads to a number-neutral reading, rather than singular; and the language has no pluralia tantum.

As a modifying plural that adjoins to the relevant phrase with which it appears, the Halkomelem plural is not capable of participating in syntactic operations such as Agree (Chomsky 2000; 2001). The main evidence for this comes from the behavior of the plural in the number agreement pattern in the language, which is in contrast to that in English (see (8) and the discussion below). Consider the Halkomelem pattern illustrated in (7). In this example, a plural noun such as s-í:wí:qe ‘men’ is compatible with either a plural determiner ye (7a) or a singular determiner te (7b). A singular noun in the language shows a similar pattern being able to appear with a plural determiner ye (7c) or a singular determiner te (7d). In Halkomelem, thus, the co-occurrence of number morphemes on a determiner and a noun is optional: a plural noun does not require a plural determiner shown by the grammaticality of both (7a) and (7b), and a singular noun does not require a singular determiner as shown by the grammaticality of both (7c) and (7d).

    1. (7)
    1. Halkomelem (Wiltschko 2008: 643)
    1.  
    1. a.
    1. T’ílém
    2. sing
    1. ye
    2. DET.PL
    1. s-í:wí:qe.
    2. man.PL
    1. ‘The men are singing.’
    1.  
    1. b.
    1. T’ílém
    2. sing
    1. te
    2. DET
    1. s-í:wí:qe.
    2. man.PL
    1. ‘The men are singing.’
    1.  
    1. c.
    1. T’ílém
    2. sing
    1. ye
    2. DET.PL
    1. swíyeqe.
    2. man
    1. ‘The men are singing.’
    1.  
    1. d.
    1. T’ílém
    2. sing
    1. te
    2. DET
    1. swíyeqe.
    2. man
    1. ‘The man is singing.’

The pattern shown in Halkomelem (7) is different from English as illustrated in (8). In English, co-occurrence of number morphemes on the determiner and the noun is obligatory, as shown in (8). For instance, only the plural form of the determiner these is available with a plural noun boys (8a) but not with a singular noun boy (8c). The contrast between (8b) and (8d) points to the same conclusion.

(8) a.   These boys can sing.
  b. *This boys can sing.
  c. *These boy can sing.
  d.   This boy can sing.

To account for the obligatory co-occurrence of number morphemes shown in English (8), Wiltschko (2008) proposes the syntactic operation Agree, in the sense of Chomsky (2000; 2001). For example, in English (8), D is a controller of number agreement, and bears unvalued number features that have to be valued by those matching features. Matching features are provided by a head such as Num (see (6)); for example, we propose that those features can be either [+plural] or [–plural].10 Subsequent to this matching and valuation, spell out applies and the relevant exponent that is most compatible with the feature bundles resulting from Agree will realize the relevant head. Otherwise, the derivation will crash. Thus, in English (8), in order to be grammatical, the unvalued number feature on D must be obligatorily valued by the matching feature on Num. For instance, as the noun is plural boys in (8a), Num is associated with feature [+plural], and thus the unvalued feature on D is valued as [+plural] and D is spelled out as these. Consequently, in (8b), the singular determiner this cannot realize D when it appears with the plural noun boys. In this case, match and value cannot take place, and the result is ungrammatical, as shown in (8b). Under this view of Agree, as Wiltschko (2008) argues, the obligatory agreement in English follows from the obligatoriness of the operation Agree.11

Turning to Halkomelem, the data in (7) shows that there is no obligatory co-occurrence of the number morphemes on a determiner and its noun, which must be the result of a different operation than Agree (Wiltschko 2008). Note that neither this paper nor Wiltschko’s suggests that Agree is optional. What is proposed in Wiltschko (2008), which we assume in this paper, is that the optional presence of number morphemes shown in data such as in Halkomelem (7) suggests that Agree is not playing a role in the appearance of number morphology in the language, contrary to English (see (8)). For example, the plural determiner ye cannot be the result of spell out of valued number feature, as suggested by the fact that it is compatible with either a singular (7a) or plural noun (7c). Moreover, the singular determiner te is compatible with a plural noun (7b) or a singular noun (7d), which suggests that te cannot be also the result of feature valuation in Agree. Unlike in English (8), then, the relevant number feature cannot be associated with a head such as Num that values unvalued features on another head such as D. Upon the consequence that Agree does not play a role in the domain of number in Halkomelem, it is proposed that the relevant number feature in the language is a modifying feature that is not capable of valuing features on other heads, such as determiners. Following this insight of Wiltschko (2008), we indicate a modifying plural feature as a monovalent feature, [plural].

As a modifying feature which cannot value an unvalued feature on a head, the feature [plural] optionally adjoins to a relevant phrase; in informal terms, it can appear either on a determiner or a noun or both.12 Nothing in the system forces or prohibits its presence or absence. For instance, a modifying feature [plural] can appear on a noun, giving rise to a plural noun such as s-í:wí:qe ‘men’, and it is compatible with a singular determiner. The same plural noun is also compatible with a plural determiner, when the [plural] feature appears on the determiner yielding a plural determiner. Thus, the analysis of the plural in the language as a modifier accounts for the optional number morphemes on nouns.

That the interpretation of singular nouns in Halkomelem is different from that in an English-type language provides further support to the view that the plural in Halkomelem is a modifier, rather than being a head plural as in English. Nouns in Halkomelem that lack plural marking receive an unmarked interpretation, i.e., a number-neutral interpretation, rather than a singular reading as in languages like English, as shown in (9). In (9a), a plural-marked noun appears giving rise to plural interpretation. In (9b), a singular form of the same noun appears; however, it is unmarked in number, giving rise to either singular or plural interpretation. As for the plural noun in (9a), it is interpreted as plural.13

    1. (9)
    1. Halkomelem (Wiltschko 2008: 462)
    1.  
    1. a.
    1. swóweles
    2. boy.PL
    1. ‘boys’
    1.  
    1. b.
    1. swíweles
    2. boy
    1. ‘a boy or boys’

If Halkomelem were a head plural language, then every DP would include a head such as Num, which would have a head feature, e.g., either [+plural] or [–plural]. In the presence of [+plural] in Num, the noun would be interpreted as plural; on the other hand, in the presence of [–plural] in Num, the noun would be interpreted as singular. However, this is not the case in Halkomelem, as shown in (9): the noun has either a plural or number-neutral interpretation. Rather, the facts in (9) suggest that as in the case of (7) discussed above, the relevant number feature in the language is a modifier. If a modifying plural feature such as [plural] is present, a plural interpretation will result as shown with a plural noun as in (9a). In the absence of number marking, the plural feature is absent entirely, and an unmarked interpretation will be observed, either plural or singular, as in (9b).

Another property that follows from the analysis of the Halkomelem plural as a modifier, rather than a feature on a head, is the absence of form-meaning mismatches. For instance, in languages with grammatical gender, sometimes the gender of a noun does not match the gender of the entity it denotes (e.g., the German noun Mädchen ‘girl’ is neuter, rather than feminine; Wiltschko 2008). This type of mismatch can be observed because gender as a grammatical (i.e., inflectional) category does not have to encode natural sex-based gender, although it has to encode grammatical gender. Given that a grammatical category is represented as a head in syntax, Wiltschko (2008) predicts that a head can show a similar type of mismatch but a modifier cannot. Thus, a head plural such as English –s is predicted to show a similar type of mismatch, while a modifying plural as in Halkomelem is predicted not to show such type of mismatch. In the category of number, a similar kind of form-meaning mismatch to gender is the case of pluralia tantum, that is, nouns which appear with plural marking, even when they refer to a singular entity, as in English pants and scissors. As English –s instantiates a head plural, the presence of pluralia tantum in the language is expected. In contrast, Halkomelem has no pluralia tantum, which is also well predicted from its having a modifying plural (Wiltschko 2008); the Halkomelem plural marker is used only when denoting a plural referent. In other words, there is no mismatch in number between a plural marker and a noun with which it occurs: a plural-marked form always matches with a plural meaning, and is never used with a singular referent. This lack of pluralia tantum, along with the set of data in (7)–(9), supports the view that the Halkomelem plural is a modifier, in contrast to the English head plural on a Num head.14

2.2 Korean modifying plural

Turning to Korean, we demonstrate that the Korean plural marker –tul should be classified as a modifying plural, not a head plural, as summarized in Table 2.15 In this section, we demonstrate properties of modifying plural (i)–(iii) for –tul as indicated in Table 2, and in Section 4, we show property (iv) for –tul, namely that it modifies the nP projection.

Modifying plural Head plural Korean plural -tul

(i) Status Optional Obligatory Optional
(ii) Interpretation Number-neutral vs. plural Singular vs. plural Number-neutral vs. plural
(iii) Feature value Monovalent [plural] Bivalent [±plural] Monovalent [plural]
(iv) Position RootP, nP, etc. Num nP

Table 2

Korean plural –tul as a modifying plural.

Like bare nouns in Halkomelem, a bare noun in Korean is unmarked in number (Kang 1994), as shown in (10). In (10a), for instance, the bare noun salam ‘person’ can be interpreted as either singular or plural.

    1. (10)
    1. a.
    1. salam
    2. person
    1. ‘a person or persons’
    1.  
    1. b.
    1. sakwa
    2. apple
    1. ‘an apple or apples’

The unmarked status of a bare noun is further supported by the fact that it can appear in a plural context as shown in (11) (Kang 1994):

    1. (11)
    1. Kang (1994: 5)
    1.  
    1. a.
    1. sakwa
    2. apple
    1. twu
    2. two
    1. kay
    2. CL
    1. ‘two apples’
    1.  
    1. b.
    1. haksayng
    2. student
    1. twu
    2. two
    1. myeng
    2. CL
    1. ‘two students’

In (11), nouns in a bare form sakwa ‘apple’ or salam ‘student’ are grammatical in a plural context indicated by a numeral and a classifier, e.g., twu myeng ‘two CL’. As with Kang (1994), we conclude that the bare form of a noun (i.e., not marked as plural) in Korean semantically includes both singular and plural individuals. In other words, in the terms of Wiltschko (2008), the bare form of a noun in Korean is unmarked in number, giving rise to a number-neutral interpretation (either a singular or plural interpretation), just like the singular form of a noun in Halkomelem.

In contrast, a noun is interpreted only as plural when it appears with the plural marker –tul, as shown in (12). The data in (11) together with that in (12) shows that the plural marker –tul is optional for a plural interpretation, similar to the plural in Halkomelem, suggesting that the Korean one is also a modifying plural.

    1. (12)
    1. a.
    1. salam-tul
    2. man-PL
    1. ‘men/*a man’
    1.  
    1. b.
    1. sakwa-tul
    2. apple-PL
    1. ‘apples/*an apple’
    1.  
    1. c.
    1. haksayng-tul
    2. student-PL
    1. ‘students/*a student’

As discussed for Halkomelem in the previous section, the absence of –tul as shown in (10), in contrast to its presence (12), would result in a singular interpretation if the feature indicated by the plural –tul were a head feature such as on Num. However, this is not the case, as suggested by the number-neutral interpretation of bare nouns, shown in (10). Thus, we conclude that the feature associated with –tul is a modifying feature, i.e., a monovalent feature [plural] that can adjoin to a relevant phrase in the nominal domain.

More support for –tul as a modifying plural comes from the optional presence of number morphemes, similar to that in Halkomelem discussed in the previous section. The pattern in Korean is presented in (13)–(14) (adapted from Kang 1994).16

    1. (13)
    1. a.
    1. i/ce/ku
    2. this/that/the
    1. sakwa
    2. apple
    1. ‘this/that/the apple’
    1.  
    1. b.
    1. i/ce/ku
    2. this/that/the
    1. sakwa-tul
    2. apple-PL
    1. ‘these/those/the apples’
    1. (14)
    1. a.
    1. i-tul/ce-tul/ku-tul
    2. this-PL/that-PL/the-PL
    1. sakwa
    2. apple
    1. ‘these/those/the apples’
    1.  
    1. b.
    1. i-tul/ce-tul/ku-tul
    2. this-PL/that-PL/the-PL
    1. sakwa-tul
    2. apple-PL
    1. ‘these/those/the apples’

Just like in Halkomelem, in Korean, either a singular form of a noun (13a)17 or a plural form of a noun (13b) is compatible with the singular form of a determiner.18 Either form of a noun is also compatible with the plural form of a determiner in the language, as shown in (14). The observed pattern in Korean can be understood as the workings of a modifying plural that cannot value an unvalued feature on a head, not as the workings of a head plural. For example, the plural form of the noun sakwa-tul ‘apple-PL’ is the result of an optional modifying feature [plural] on the noun. As such, it is compatible with a singular determiner i/ce/ku (13b) or with a plural determiner i-tul/ce-tul/ku-tul as in (14b). There is no need of matching and valuing, as there is no such formal feature to be matched and valued.19

Korean, like Halkomelem, has no examples of pluralia tantum, providing further support for the proposed account in this paper that –tul is a modifying plural. For example, in Korean, nouns that are pluralia tantum in English are all marked as singular, unlike in English and many related languages, as shown in (15). All nouns in (15) are singular, that is, lacking the plural marker.

    1. (15)
    1. a.
    1. paci
    2. pants
    1. ‘pants’
    1.  
    1. b.
    1. kawui
    2. scissors
    1. ‘scissors’
    1.  
    1. c.
    1. ankeyng
    2. glasses
    1. ‘glasses’

Marking these nouns with –tul results in an interpretation of plurality; for example, for the noun in (15a), paci-tul ‘pants-PL’ means multiple pairs of pants, suggesting that –tul, like the Halkomelem plural, is a modifier that indicates plurality, rather than a grammatical plural feature; since modifiers must be interpreted, this kind of form-meaning mismatch is impossible. The discussed data suggests that –tul shows the properties of a modifying plural, rather than a head plural. In what follows, we show that –tul modifies nP, rather than the RootP as in Halkomelem.

3 –Tul is not a Root modifier

We have demonstrated that the Korean plural marker –tul is a modifying plural, rather than a head plural. The next question is, what category in the nominal spine does it modify? For instance, consider a nominal spine as in (16).

    1. (16)

We claim that –tul is a nP modifier, showing properties of n as recognized in the literature (Marantz 2001; Acquaviva 2008; Lowenstamm 2008; and Kramer 2009; 2015), and demonstrate that it cannot be a modifier of RootP or DP in the nominal spine in (16).20 In this section, we show evidence that –tul does not modify RootP unlike a Halkomelem modifying plural, and in Section 4, we demonstrate that –tul shows the properties of a nP plural. In Section 5, we show that –tul cannot be a definiteness marker, and thus it cannot be a modifier of DP.

The modifying plural in Halkomelem is argued to modify the Root. As a root modifier, it is expected to occur closer to the root than any other affixes, including categorizing morphology. In Halkomelem, this is borne out: the plural marker does occur inside of derivational morphology. For instance, consider the examples in (17). The example in (17b) is a nominalised form of the word in (17a). When the nominalised form in (17b) is pluralized, the plural morphology, which is reduplication in this example, appears inside of the nominalizing prefix s–. In English, this is impossible (18):

    1. (17)
    1. Halkomelem (Galloway 1993: 379)
    1.  
    1. a.
    1. p’eq’
    2. white
    1. ‘white’
    1.  
    1. b.
    1. s-p’eq’
    2. NMLZ-white
    1. ‘white spot on skin’
    1.  
    1. c.
    1. s-p’eq’p’eq’ (*sp’eq’sp’eq’)
    2. NMLZ-white.PL
    1. ‘white spots on skin’
(18) *brother-s-hood, *tattoo-s-ist

Since nominal categorizing morphology is assumed to be of category n, the Halkomelem plural must modify a category lower than nP, leaving only the Root. A similar expectation applies for compounds. Compound nouns are generally analyzed as the combination of two roots (Wiltschko 2008);21 consequently, pluralization within compounds should be possible only if the plural marker occurs in RootP. This is the case in Halkomelem (19), but not in English (20):

    1. (19)
    1. Halkomelem (Wiltschko 2008: 644)
    1.  
    1. s-xexp’-í:tsel
    2. NMLZ-stripe.PL-back
    1. ‘chipmunk’
(20) *rats-infested, *teethbrush

The data (17)–(20) supports that the Halkomelem plural marker is a modifier at the Root level (see (6b)). In sum, the Halkomelem plural is a root modifier as it can occur inside derivational morphology, and inside of compounds, as demonstrated above in (17) and (19).

Neither of these is allowed with Korean –tul. It cannot attach closer to the root than a categorizing head. For example, (21) illustrates a compound noun kkoch-ip ‘petal’ consisting of two roots, kkoch ‘flower’ and ip ‘leaf’. As in (21b), –tul cannot appear between the roots:

    1. (21)
    1. a.
    1.  
    1. kkoch-ip
    2. flower-leaf
    1. ‘petal(s)’
    1.  
    1. b.
    1. *
    1. kkoch-tul-ip
    2. flower-PL-leaf
    1. (intended) ‘petals’

–Tul also cannot appear inside of derivational morphology. Consider a noun such as namu-kkun ‘lumberjack’ in (22). In (22a), the noun namu ‘tree’ is suffixed with a derivational morpheme –kkun ‘being a specialist of’. As shown in (22b), –tul cannot appear between the noun and this morpheme:

    1. (22)
    1. a.
    1.  
    1. namu-kkun
    2. tree-specialist
    1. ‘lumberjack(s)’
    1.  
    1. b.
    1. *
    1. namu-tul-kkun
    2. tree-PL-specialist
    1. (intended) ‘lumberjacks’

We conclude that –tul is not a modifier of RootP, unlike the Halkomelem plural. Next we show that nP is the category that –tul modifies.

4 –Tul is a nP modifier

It is often claimed that nP is the locus of idiosyncratic properties in the nominal domain; thus, a plural in nP or below may show idiosyncrasies, while a plural in higher projections should be regular (Acquaviva 2008; Lowenstamm 2008; Kramer 2009; 2015; see also a similar intuition in Corbett 2000). For instance, Amharic has plurality split into two types: a NumP plural and a nP plural (Kramer 2009; 2015). The NumP plural is the language’s regular plural marker: it can attach to any noun, resulting in bivalent feature interpretation on the noun (presence of the plural instantiates [+plural] feature on Num, absence of the plural instantiates [–plural] on Num), and does not vary with the gender of the noun. The idiosyncratic nP plural, on the other hand, is an irregular plural. It occurs with only a few nouns in the language, showing lexical gaps. It may give either a [plural] interpretation or an idiosyncratic reading; for example, the irregularly-pluralized ahzab ‘nation.PL’ in Amharic has two readings, either the predictable ‘nations’ or the idiosyncratic ‘barbarians’. The irregular plurals are also idiosyncratic in that their forms are different with different genders. Thus, while the NumP plural behaves as a regular grammatical plural marker, the nP plural shows various kinds of idiosyncrasy.

Korean –tul shows similar idiosyncratic properties to those of irregular nP plurals in Amharic, specifically in terms of gaps in its distribution, which provides support for the current proposal that –tul is adjoined to nP.22 Corbett (2000) notes the cross-linguistic generalization that the distribution of plural marking in a language depends on the animacy of the noun. A simplified version of the animacy hierarchy is shown in (23):

(23) Animacy hierarchy:
  human > non-human animate > inanimate

Corbett claims that, if a noun at some point in the hierarchy can be pluralized, then all nouns above that point can also be pluralized; thus, for example, if a language allows plural marking on non-human animate nouns, then it will also allow pluralization of human-denoting nouns, but not necessarily on inanimate nouns. However, Korean does not fit this generalization. The plural –tul can attach to almost any human noun (24a), and also to many inanimate nouns (24b), but less frequently with animals (25):23

    1. (24)
    1. a.
    1. salam-tul
    2. person-PL
    1. sey
    2. three
    1. myeng
    2. CL
    1. ‘three people’
    1.  
    1. b.
    1. chayk-tul
    2. book-PL
    1. sey
    2. three
    1. kwun
    2. CL
    1. ‘three books
    1. (25)
    1. ??
    1. kilin-tul
    2. giraffe-PL
    1. sey
    2. three
    1. mari
    2. CL
    1. ‘three giraffes’

In fact, Kang (2007) examines the number of nouns with –tul in the most frequently used 100 common nouns in the language in the SeJong corpus, and found that 77 of them denote humans, 22 denote inanimate things, and only one denotes an animal (non-human animate). Kang’s findings were presented in Korean; we present them in English as in Table 3.24

Semantics of nouns with –tul Number of nouns with –tul

Human (e.g., ‘person’, ‘child’, ‘student’ etc.) 77
Animal (e.g., ‘bird’, ‘rabbit’, and ‘animal’ etc.) 1
Thing (e.g., ‘work’, ‘book’, ‘tree’ etc.) 22

Table 3

Animacy of nouns with –tul (adapted from Kang 1997: 15).

Although the frequency of –tul on non-human animate nouns is lower,25 it is true that the occurrence of –tul on these nouns is well observed, as pointed out by reviewers. For example, it is mentioned to us that –tul can be easily observed with animate nouns in the corpus data (SeJong-RISK corpus). One of the reviewers also pointed out that the frequency of –tul is different from one non-human animate noun to another in the corpus data: it is found more frequently with some nouns denoting animals such as kay ‘dog’ or koyangi ‘cat’, while less frequently (and less natural) with others such as kilin ‘giraffe’, consistent with the data shown in (25). This fact seems to show that the occurrence or frequency of –tul is not regular across non-human animate nouns among the speakers or in the corpus data compared to its occurrence with human-denoting nouns or with inanimate nouns. We take this irregularity observed with –tul on animate nouns as evidence to indicate that there is no clear correlation between the animacy of a noun and the availability of pluralization with –tul, contrary to Corbett’s predictions. This idiosyncrasy in distribution is expected if –tul attaches in nP, the locus of idiosyncrasy, as proposed in this paper and shown in (5), repeated here.26

    1. (26)

Moreover, regarding the animacy hierarchy, a reviewer points out that the occurrence of –tul shows speaker variation. In addition to those speakers with the judgements reported above, there are some speakers who allow –tul consistently with the hierarchy, while others do not allow –tul with any non-human animate nouns. This variation provides further evidence that the overall occurrence of –tul seems to be irregular, unlike a usual regular plural marker such as the plural marker –s in English, which is consistent with the account of –tul proposed in this paper.

5 –Tul does not mark definiteness

In this section we present evidence that Korean –tul does not encode definiteness, which leads us to conclude that –tul is not a definiteness marker. Moreover, we argue that some of previous analyses that propose –tul as a D head are not well supported.

5.1 –Tul is not a definiteness marker

It has often been observed that plural markers in classifier languages tend to be associated with definite readings, or that they occur more often in definite than indefinite contexts (e.g., Iljic 1994; Li 1999; Ghomeshi 2003), and this is often interpreted as meaning that the plural marker itself encodes definiteness. Moreover, in Korean literature, there seems to be underlying assumption that –tul marks definiteness.27 For example, –tul is obligatory when a noun appears with demonstrative (ku) (Suh 2005; 2008, citing Sohn 1999) (see (27a)).28 However, this does not seem to be true in Korean as well as in other languages that have a plural marker similar to –tul.

For example, Ghomeshi (2003) shows that the Persian plural marker –ha does not itself encode definiteness. The Korean plural marker –tul behaves in a similar way, as illustrated in (27).

    1. (27)
    1. a.
    1. Inho-ka
    2. Inho-NOM
    1. ku
    2. DEM
    1. ai-tul-ul
    2. child-PL-ACC
    1. coha-ha-n-ta.
    2. like-do-PRES-DECL
    1. ‘Inho likes those/the children.’
    1.  
    1. b.
    1. Inho-ka
    2. Inho-NOM
    1. ku
    2. DEM
    1. ai-lul
    2. child-ACC
    1. coha-ha-n-ta.
    2. like-do-PRES-DECL
    1. ‘Inho likes that/the child.’

In (27), the object of the verb coha ‘like’ is definite regardless of the presence (27a) or absence (27b) of plural –tul. It is interpreted as being definite because of the presence of demonstrative ku, rather than the plural marker. The fact shown in (27) suggests that plural –tul per se does not mark definiteness, which is further supported by the fact that –tul can appear with an indefinite interpretation (see (29)–(30) below).

Nakanishi & Tomioka (2004), and Nomoto (2013) make similar claims about the Japanese plural marker, –tati.29 Here, we make a similar claim for Korean –tul and we present the evidence in (28), provided by the above authors to argue against a definite interpretation of a plural marker –tati. (The list of evidence in (28) is not exhaustive, and we do not discuss evidence irrelevant to Korean).

(28) tul-marked nouns can:
  a. receive an indefinite interpretation
  b. appear in existential constructions
  c. be a predicate
  d. combine with a wh-phrase
  e. be an antecedent of a sluiced wh-phrase

If –tul is a marker of definiteness, it should not be able to receive an indefinite reading. This prediction is not borne out, as illustrated in (29)–(30) (adapted from Kwon & Zribi-Hertz 2004):30

    1. (29)
    1. Mina-nun
    2. Minna-TOP
    1. chayk-tul-ul
    2. book-PL-ACC
    1. ilk-ci.an-ass-ta.
    2. read-NEG-PST-DECL
    1. ‘Minna didn’t read some books.’ (i.e., ‘There are some books that Minna didn’t read.’)
    1. (30)
    1. Chelswu-nun
    2. Chelsu-TOP
    1. cengchiin-tul-ul
    2. politician-PL-ACC
    1. manna-ko
    2. meet-COMP
    1. sipeha-n-ta.
    2. want-PRES-DECL
    1. ‘There are some politicians that Chelsu wants to meet.’31

In both examples, the –tul-marked nouns allow an indefinite reading, contrary to the prediction of an analysis of –tul as a definite marker, which would predict that the interpretation of (29) would be ‘Minna didn’t read the books’, and (30) would be interpreted as ‘Chelsu wants to meet the politicians’.

While the -tul-marked nouns in (29) and (30) take obligatory wide scope, there are also contexts in which a noun marked with –tul is obligatorily interpreted as taking narrow scope, which is a property typically associated with bare plural nouns (i.e., lacking any D), and impossible for definite noun phrases (Carlson 1977). Nakanishi & Tomioka (2004) demonstrate this for the Japanese plural –tati, and a relevant example is replicated for –tul in (31).

    1. (31)
    1. Ku
    2. DEM
    1. kongwen-eyse-nun
    2. park-at-TOP
    1. hangsang
    2. always
    1. ai-tul-i
    2. child-PL-NOM
    1. nol-ko-iss-ta.
    2. play-COMP-be-DECL
    1. (i) ‘In this park, there are always some/*the children playing.’
    2. (ii) #‘In this park, there are some children who are always playing.’

A salient reading of this type of example in (31) is (i), where the –tul-marked noun ai-tul takes a narrow scope with respect to the quantification adverb hangsang ‘always’. This is contrary to the prediction of an analysis of –tul as a definiteness marker. A wide scope reading of the –tul-marked noun as in (ii) is not favored. Moreover, the interpretation of ai-tul in (31) is indefinite, similar to (29) and (30), and definite reading is not allowed. The possible and impossible readings in (31) constitute strong evidence that –tul cannot be a marker of definiteness.

Another piece of evidence showing that –tul does not mark definiteness comes from the fact that –tul-marked nouns can appear in existential constructions. In English, only indefinite DPs can occur in existential sentences, as shown in (32) below (adapted from Suh 2005: 778):32

(32) a.   There’s a guy at the door.
  b.   There’s some guy at the door.33
  c. #There’s the guy at the door.
  d. #There’s that guy at the door.

If –tul indicated definiteness, it would be predicted to be disallowed in an existential construction as in the English examples (32c) and (32d); however, it can appear in existential sentences (33).

    1. (33)
    1. a.
    1. Kongwen-ey
    2. park-LOC
    1. ai-tul-i
    2. child-PL-NOM
    1. iss-ta.
    2. exist-DECL
    1. ‘There are children at the park.’
    1.  
    1. b.
    1. Salam-tul-i
    2. person-PL-NOM
    1. o-ko
    2. come-COMP
    1. iss-ta.
    2. PROG-DECL
    1. ‘People are coming.’

This is further evidence that –tul-marked nouns are not necessarily definite.

More evidence against –tul as a definite marker comes from copula sentences with predicate nominals. Predicate nominals are those which ascribe some property to the subject, as in (34).

(34) They are (#the) students.

In (34), without the definite article, students is a predicate nominal, ascribing the property of being a student to each of the individuals denoted by the subject pronoun. The sentence is also grammatical with the definite article, but the reading is different: the sentence is then an equative sentence, equating the group of individuals denoted by the subject with some definite, specific group of students (under discussion). Thus, while both definite and indefinite DPs can appear in copula sentences in English, only indefinite DPs allow a predicative interpretation. This is expected, as it is frequently claimed that bare N(P)s denote predicates, while higher projections (like DP) are needed to convert this predicate into an individual so it can act as an argument (e.g., Longobardi 1994; Chierchia 1998; Zamparelli 2000; Déchaine & Wiltschko 2002; Borer 2005, among others). Given that only an indefinite DP can be a predicate, Korean –tul-marked nouns should be disallowed as a predicate if they are definite. However, this prediction is not borne out; as shown in (35), –tul-marked nouns can act as predicates:34

    1. (35)
    1. Ku-tul-un
    2. he-PL-TOP
    1. haksayng-tul-i-ta.
    2. student-PL-be-DECL
    1. ‘They are students.’

Here, the plural-marked noun haksayng-tul is acting as a predicate nominal, which means that –tul is not marking it as definite.

The next two pieces of evidence for the ability of –tul-marked nouns to be indefinite come from similarities of –tul to Japanese –tati in two contexts identified by Nakanishi & Tomioka (2004) for Japanese: (i) being modified by a wh-demonstrative, and (ii) being an antecedent of a sluiced element. First, –tul-marked nouns can be modified by the wh-demonstrative etten ‘which/what kind’, like –tati. Nouns with this kind of modifier must be indefinite; what kind of (*the) students is grammatical only without the definite article. The fact that, in Korean, –tul-marked nouns may receive this modifier suggests that they must be indefinite in a context like (36):

    1. (36)
    1. Etten
    2. which
    1. haksayng-tul-i
    2. student-PL-NOM
    1. wa-ess-ni?
    2. come-PST-Q
    1. ‘What kind of (*the) students came?’

Another piece of evidence comes from sluicing. It has been shown by Chung et al. (1995) that, while weak indefinites may serve as the antecedent of sluiced elements, definite and strongly quantified DPs may not, as shown in (37) for English (Nakanishi & Tomioka 2004: 122), which has been shown to apply to –tati:

(37) a. John met a student, but Sue doesn’t know which one.
  b. #John met the student, but Sue doesn’t know which one.

Korean –tul-marked noun phrases can appear as the antecedent of sluicing, as shown in (38):

    1. (38)
    1. Kim
    2. Kim
    1. sensayngnim-uy
    2. teacher-GEN
    1. cip-ey
    2. house-LOC
    1. ai-tul-i
    2. child-PL-NOM
    1. moyeissta-ko
    2. gather-COMP
    1. tulessciman
    2. hear.but
    1. na-nun
    2. I-TOP
    1. etten
    2. which
    1. ai-tul-inci
    2. child-PL-whether
    1. molun-ta.
    2. don’t.know-DECL
    1. ‘I heard that children gathered at Professor Kim’s house, but I don’t know which children.’

The plural noun ai-tul ‘children’ in the first clause in (38) serves as the antecedent of the sluiced phrase etten ai-tul ‘which children’ in the second clause, which should be possible only if the antecedent is indefinite. The sentence is predicted to be ungrammatical if –tul marks definiteness, contrary to fact.

The final piece of evidence that –tul does not mark definiteness comes from its possible co-occurrence with the distributivity marker –ssik. Choe (1987) argues that –ssik falls into a class of distributivity markers that he refers to as anti-quantifiers, which include the so-called “shifted each35 in the English sentence (39):

(39) Choe (1987: 1)
  The children bought a balloon each.

In (39), the shifted each is analyzed as a distributivity marker situated in the DP headed by balloon, which indicates that there is one balloon per child – that is, the balloons are distributed among the children. Among the properties of these anti-quantifiers is that the accompanying noun phrase must be indefinite, as is the case with a balloon in (39). This is also true with anti-quantifier –ssik in Korean as illustrated in (40).

    1. (40)
    1. Ai-tul-i
    2. child-PL-NOM
    1. (*ku)
    2. (that)
    1. phwungsen-hana-ssik-ul
    2. balloon-one-SSIK-ACC
    1. sa-ess-ta.
    2. bought
    1. ‘The children bought one balloon each.’

As shown in (40), –ssik cannot co-occur with a demonstrative, which would force a definite reading. Another property that –ssik and shifted each share is that both are most natural with direct objects (in Korean, accusative-marked phrases), as in both (39) and (40).

In Korean, noun phrases marked by –ssik can include –tul, as shown in (41). This example is similar to the one provided in Choe (1987) except for the insertion of –tul on the accusative object ‘balloon’ (adapted from Choe 1987: 46).36

    1. (41)
    1. Ai-tul-i
    2. child-PL-NOM
    1. phwungsen-tul-hana-ssik-ul
    2. balloon-PL-one-SSIK-ACC
    1. sa-ess-ta.
    2. bought
    1. ‘The children bought one balloon each.’

Given that –ssik obligatorily appears with an indefinite noun, the fact that the co-occurrence of –tul and –ssik on a single noun is grammatical suggests that –tul does not mark definiteness.

The data discussed in this section suggests that –tul does not mark definiteness, and that –tul-marked nouns may receive a (weak) indefinite interpretation. We conclude that the assumption that –tul is a definiteness marker is incorrect.

5.2 –Tul does not instantiate D

The conclusion from the previous section in which –tul cannot be a definiteness marker is contrary to the analysis of Park (2008). Park provides an account of –tul set in the theoretical framework of Borer (2005), as illustrated in (42). In Borer’s framework, nP is dominated by a series of functional projections, each making a semantic contribution to the interpretation of the DP as a whole. The lowest functional projection is DivP; the Div head divides this mass to give a count interpretation. In English, this is realized by the plural marker, while in classifier languages, this is normally seen to be the role of the classifier (see Section 1). The next functional head is #, which counts this divided mass; this is realized by numerals and certain quantifiers. Finally, DP provides definiteness, strong quantification, and an argumental interpretation, following Longobardi (1994).37

    1. (42)

In Park’s analysis, –tul is base-generated as a Div head; N subsequently undergoes head movement to Div. –Tul moves along with N up to #, to provide a plural, counted interpretation, and finally to D, as shown in (42), providing an argumental, definite interpretation. According to Park (2008), –tul plays the roles of the heads Div, #, and D, as made possible via head movement.

The first objection to the analysis in Park (2008) is that –tul cannot be a head in the nominal spine, regardless of the type of head, as demonstrated in Section 2. If –tul were a head in the nominal spine as proposed in Park (2008), the relevant properties of –tul as a modifying plural shown in Section 2 could not be accounted for; moreover, the properties of –tul adjoining to nP shown in Section 4 cannot be accounted for either, as –tul in Park (2008) initially occupies Div, a head higher than nP. In the rest of this section, we discuss Park’s claim that –tul occupies the # and D heads, and conclude that the claim makes incorrect predictions.38 We discuss three aspects of Park’s proposal crucial to her analysis of –tul in D: (i) –tul has similar semantics to the English quantifier all; (ii) –tul cannot play the role of a predicate; and (iii) –tul gives a strong reading, such that N-tul scopes over all other modifiers in a sentence. As for the first piece of evidence, Park shows that –tul shows similar semantic properties to English all (following Brisson’s 2003 analysis of all), and takes this as evidence that N-tul must be a quantified DP. However, while this seems to capture the semantics of –tul, it does not imply a syntactic analysis in which N-tul must occupy the # and D nodes. For example, all itself does not necessarily occupy either of these nodes, as shown by examples like all the children, in which all is generated in spec-#P and moves to spec-DP, modifying both projections, while the occupies the # and D heads (Borer 2005). Therefore, Park’s semantic analysis of –tul does not force a syntactic analysis in which –tul occupies either # or D.

The second aspect of Park’s argument is the claim that her proposal correctly predicts that –tul-marked nouns cannot occur as predicates. This is predicted by Longobardi’s (1994) account of arguments and predicates, in which NPs are predicates, while D serves the function of saturating the predicate and allowing the phrase to function as an argument. Therefore, NPs are predicates, and DPs are arguments, so if N-tul is a DP, then it must not be able to function as a predicate. Park shows examples of N-tul in predicate position, and marks them as ungrammatical, which suggests that N-tul is necessarily an argument, and therefore DP. However, as shown above in (35), nouns marked with –tul can in fact occur as predicates for the speakers that we consulted.39 Given this variation, the analysis proposed in Park (2008) would leave the well-formed data unaccounted for, and cannot be strongly endorsed.40 Finally, Park’s analysis of N-tul as a quantified DP would predict that –tul-marked nouns obligatorily receive a strong reading (definite or specific indefinite), and so must receive wide scope with respect to other operators in the sentence. However, as shown in (31) in Section 5.1, in which ai-tul ‘child-PL’ receives narrow scope with respect to hangsang ‘always’, this is also not necessarily the case. Therefore, Park’s claim that –tul-marked nouns move to D is not supported.

Park also claims that her analysis correctly predicts the distribution of –tul-marked nouns with respect to numerals and classifiers. The structure for sakwa sey-kay ‘apple three-CL’ under Park’s analysis is shown in (43):41

    1. (43)

In (43), since the Div and # heads are occupied by sey-kay ‘three-CL’, there is no room for –tul to occupy these heads; therefore, under Park’s analysis of –tul realizing Div and #, –tul should be unable to co-occur with numerals and classifiers, or with numerals in the absence of classifiers. Park supports this prediction with multiple examples of –tul-marked nouns with numerals, some with a classifier (44) and some without a classifier (45), all marked as ungrammatical.

    1. (44)
    1. Park (2008: 189)
    1.  
    1. a.
    1. *
    1. sakwa-tul
    2. apple-TUL
    1. sey
    2. three
    1. kay
    2. CL
    1. ‘three apples’
    1.  
    1. b.
    1. *
    1. chayk-tul
    2. book-TUL
    1. tases
    2. five
    1. kwen
    2. CL
    1. ‘five books’
    1. (45)
    1. Park (2008: 190)
    1.  
    1. a.
    1. *
    1. cha-tul
    2. car-TUL
    1. seys
    2. three
    1. ‘three cars’
    1.  
    1. b.
    1. *
    1. sakwa-tul
    2. apple-TUL
    1. seys
    2. three
    1. ‘three apples’

As mentioned above, in Park’s (2008) analysis, –tul functions both to divide and count the denotation of the noun, and so it realizes both Div and #, as shown in (42). Park assumes that numerals in Korean also serve both to divide and count (i.e., they realize both Div and #), both when they occur with a classifier as a complex word, as in (43), and when they appear alone (see Footnote 41). Thus, Park argues that –tul cannot appear either with a numeral and classifier together, as in (44), or with a numeral alone, as in (45). However, Park’s evidence is not sufficient as it does not include the core distribution of –tul, namely its occurrence on human nouns: with human nouns, –tul can appear with classifiers or numerals, as shown in (4) in Section 1 and (46)–(47) below. The co-occurrence of –tul and a classifier or –tul and a numeral are acceptable when a noun is human, as shown in (46) and (47) respectively. Thus, Park’s argument that –tul realizes a Div or # head and so cannot co-occur with a classifier or numeral makes an incorrect prediction.

    1. (46)
    1. a.
    1. yeca(-tul)
    2. woman(-PL)
    1. sey
    2. three
    1. myeng
    2. CL
    1. ‘three women’
    1.  
    1. b.
    1. hyengcey(-tul)
    2. brother(-PL)
    1. ney
    2. four
    1. myeng
    2. CL
    1. ‘four brothers’
    1. (47)
    1. a.
    1. yeca(-tul)
    2. woman(-PL)
    1. seys
    2. three
    1. ‘three women’
    1.  
    1. b.
    1. hyengcey(-tul)
    2. brother(-PL)
    1. neys
    2. four
    1. ‘four brothers’

Moreover, with respect to Park’s data with animals or inanimates as in (44)–(45) which is shown to be ungrammatical, Park fails to note speaker variation. Some speakers reject the co-occurrence as in Park (44)–(45), but others accept such examples. Variation in judgement is well expected by the proposed account in this paper: –tul is a modifier of nP, where idiosyncratic properties are observed.

To summarize this section, we gave the following main reasons for rejecting Park’s analysis. First, as extensively demonstrated throughout the paper, plural –tul is a modifier adjoining to nP, rather than a head in the nominal spine such as Div, # or D, contrary to Park (2008). Second, while –tul appears to have similar semantics to English all, this does not constitute evidence that it occupies D, as all itself is not in complementary distribution with determiners. Third, as we show in Section 5.1, –tul-marked nouns do not necessarily receive wide scope with respect to other operators, and can appear as predicates. Finally, –tul is not in complementary distribution with numerals or classifiers, contrary to Park’s prediction. Therefore, we conclude that Park’s (2008) claim that –tul occupies the Div, #, and D heads is not supported, and thus, –tul cannot be a head plural, as proposed in this paper.

6 Consequences

Recall the prediction of Borer (2005) that motivated this study: plural markers and classifiers cannot co-occur, as they are both viewed as elements of the category Div. In a language where both plural markers and classifiers are present, they are predicted not to co-occur in a single nominal projection, which is not the case for Korean as shown earlier in (4), repeated as (48), in which plural –tul can optionally co-occur with a classifier. Given this co-occurrence and other pieces of evidence provided throughout this paper, we argued that –tul in Korean is not instantiating a head plural which is realized by a classifier. Instead, –tul is proposed to be a modifier of the nP projection.

    1. (48)
    1. salam(-tul)
    2. human(-PL)
    1. ney
    2. four
    1. myeng
    2. CL
    1. ‘four people’

If the classifier realizes a Div head that appears above nP as Borer (2005) claims, then co-occurrence of –tul and a classifier is unproblematic, and perhaps expected. In fact, a similar prediction is made in Wiltschko (2008) for a modifying plural of RootP in Halkomelem. It has shown that in Halkomelem the root-modifying plural can co-occur with an element that plays a role of a classifier in the language. Thus, the proposed analysis of –tul as a modifying plural of nP provides strong support for both Borer (2005) and Witlschko (2008). That is, a plural marker that is not the realization of a head dedicated to individuation (e.g., Div or Num), such as –tul in Korean, is predicted to be able to co-occur with a classifier.

A question that arises at this point is what is the status of classifiers in Korean, e.g., do they realize a head similar to Div (so that it can co-occur with the plural –tul)? Although we do not address whether a classifier in Korean is an instance of Div, we provide some support for the view that a classifier in the language instantiates a head in the nominal spine, unlike modifying plural –tul. The status of a classifier as a head is consistent with the fact that plural –tul can co-occur with a classifier in the language. The classifier shows properties of a syntactic head, in the sense of Wiltschko (2008): it is obligatory in a counting context, rather than optional; it can appear with any noun; it is (at least partly) inflectional, in the sense that it enters into agreement relationships with certain features of the head noun; and it enters into a selectional relationship with the head noun, selecting it based on class features.42

A classifier is obligatory in a counting context, like head plurals (Wiltschko 2008), as shown in (49).

    1. (49)
    1. saca
    2. lion
    1. ney*
    2. four
    1. (mari)
    2. CL
    1. ‘four lions’

The classifier mari may not be omitted in a noun phrase like (49), unlike non-head plurals, such as –tul (as shown in Section 2.1). Furthermore, unlike –tul as illustrated in (50), it is required for any noun. Obligatoriness of the classifier suggests that it is a head in the nominal spine (Wiltschko 2008; Kramer 2009; 2015).

    1. (50)
    1. a.
    1. saca(??-tul)
    2. lion-PL
    1. ney*
    2. four
    1. (mari)
    2. CL
    1. ‘four lions’
    1.  
    1. b.
    1. chayk(??-tul)
    2. book-PL
    1. ney*
    2. four
    1. (kwen)
    2. CL
    1. ‘four books’

Classifiers also enter into agreement and selectional relationships with their head nouns, both of which are inflectional properties indicative of head status (Wiltschko 2008; Kramer 2009; 2015). For example, a classifier agrees with the honorific status of a head noun (Lee Chingmin 2000), as shown in (51):

    1. (51)
    1. sensayngnim
    2. teacher
    1. twu
    2. two
    1. pwun/?myeng
    2. CL.HON/?CL.person
    1. ‘two teachers (honorific)’

With the honorific head noun sensayngnim ‘teacher’ in (51), the classifier should appear in the honorific form pwun, rather than the non-honorific classifier myeng, which is normally used when counting people. Thus, a classifier agrees with the honorific feature on the head noun. Furthermore, a classifier selects its head noun based on properties of the noun, such as humanness, animacy, gender, shape, and so on (i.e., noun class; see Kihm 2005 for discussion), as shown in (52):

    1. (52)
    1. a.
    1. salam
    2. person
    1. ney
    2. four
    1. myeng/*mari
    2. CL.person/CL.animal
    1. ‘four people’
    1.  
    1. b.
    1. horangi
    2. tiger
    1. ney*
    2. four
    1. myeng/mari
    2. CL.person/CL.animal
    1. ‘four lions’

With the noun salam ‘person’, or any noun referring to a person (except for honorific nouns, as discussed above), the classifier myeng is required; the classifier mari, used for animals, is ungrammatical in this context. However, with horangi ‘tiger’, or other nouns denoting animals, mari is required, and myeng is ungrammatical. Thus, the classifiers select for head nouns with certain class features, which is also indicative of a syntactic head.43 Thus, in Korean, with a classifier being a head on the one hand, and a plural –tul being a modifier on the other hand, co-occurrence of the two is well predicted. Moreover, Borer’s (2005) claim that a classifier and a plural cannot co-occur in a language should be taken as syntactic complementary distribution. In a language where both elements are markers of individuation represented by a head such as Div, the lack of co-occurrence is a valid prediction, e.g., Armenian as cited in Borer (2005) (see (2) in Section 1). However, in a language such as Korean where classifiers and plurals are not exponents of the same syntactic head in the nominal spine but only appear to have similar semantic role such as individuation, co-occurrence can be observed. Thus, the consequences of the current paper suggest that the relation between a plural marker and a classifier can be represented as one of the complementary syntactic functions instantiated by a relevant head (e.g., Div), but having a similar semantic function such as individuation may not guarantee syntactic complementary distribution.

7 Conclusion

In this paper we claimed that the Korean –tul is a modifying plural in the sense of Wiltschko (2008), adjoining to nP, whose properties are similar to those of the irregular plurals in Kramer (2009; 2015). In particular, –tul is not a head in the nominal spine, unlike classifiers in the language. This explains the co-occurrence of –tul and a classifier in a single nominal phrase in Korean; whether a classifier is an instance of a functional head above nP in a nominal spine or not, the proposed analysis in this paper suggests that two morphemes appear at different points in the nominal extended projection.

One of the conclusions of this paper is contrary to the common assumption in Korean literature that –tul instantiates a head plural, e.g., a Num head (e.g., Kim Chong-Hyuck 2005; Suh 2008). However, the properties that we demonstrated in Section 2 constitute strong evidence against such an assumption. Moreover, our claim that –tul in Korean is a modifying plural is in parallel to other plurals in classifier languages such as Mandarin (–men) or Japanese (–tati). For both languages, the plural marker is shown to exhibit ranges of properties exceptional to a usual head plural marker (Li 1999 for Mandarin; Nakanishi & Tomioka 2004 for Japanese), as we have shown for Korean –tul. These languages are also similar in that they belong to similar types in terms of Chierchia’s (1998) Nominal Mapping Parameter. Chierchia proposes that nouns differ semantically across languages in terms of whether they are an argument or a predicate ([±argument, ±predicate]) and this difference manifests itself in a cluster of properties.44 Specifically, he identifies Chinese and Japanese as [+argument, –predicate] languages where bare nouns can function as arguments and manifest general number. This type of language has classifier systems, as in Japanese and Chinese. As this paper shows, Korean appears to belong to the [+argument, –predicate] type, as it is a classifier language, and bare nouns in the language can appear as arguments and show general number. Overall, the present paper contributes to the ongoing discussion of the role of pluralizers in classifier languages as well as in non-classifier languages.