Japanese Floating Classifiers

William McClure
Queens College/CUNY Graduate Denter

Japanese classifying expressions have the form [hon-3.satu]-o 3 books [lit: book-3.volume]-acc. Argument classifiers may float away from their NP giving hon-o 3.satu, while adjunct classifiers may not, and *mise-de san.ken at 3 stores [lit: store-at 3.buildings] is bad (Inoue 1978). Kitahara (1993) citing Kamio (1983) claims that the floated structure is basic and that it is a constituent DP. Kitahara tests for constituency using conjunction with to and, but it is not clear that to selects for DPs. Compare: [747-ga Narita-kara 4.ki] to [737-ga Kankuu-kara 5.ki] tobitatta Four 747s from Narita and five 737s from Kankuu took off. The case marking of the to-conjoined constituents contrasts with the genitive case usually found in complex nominals (cf. [np747-no Narita-kara-no 4.ki]). In contrast to this, Fukushima (1991, 1993) argues that 3.satu in both positions (and others) is simply an adverb. While this hypothesis more accurately captures the distribution of classifying expressions, how we derive the agreement between the noun and its classifier is not clear. Further, neither Kitahara nor Fukushima can explain why only argument classifiers float.
I begin with the syntax and semantics of Japanese case-marking. Let say that all nominal elements in Japanese have the syntax: [xp [qp NP [q Number] [q Classifier]] particle]. As Kitahara proposes, agreement between the NP and Classifier is determined by Head-Spec within QP. The specific label of XP depends on the semantic contribution of the particle. Arguments (ga, o) are DP, while adjuncts (de, kara) are PP. Both arguments and adjuncts allow the XP-internal [xp[qp ]-particle] structure, although it should be noted that the classifying element is actually optional. I will assume that all NP-external QPs (i.e. the floated ones) are selected by an adverbial PP with an empty head giving us the same [xp [qp NP [q Number] [q Classifier]] particle] structure. Paralleling Fukushima, I treat the floated classifiers as adverbs. Although these adverbial classifiers have a null head, for other kinds of adverbs, the head may be overtly realized as ni or other particles. Given that all XPs (argument, adjunct, adverbial) select for a QP, differences between the various XPs stem from the differing semantic contribution of the XP head.
What is the semantic contribution of the various XP heads? Adopting the version of event semantics used in Kratzer (1996) (where agents are introduced separately), I propose that the accusative case marker o has the semantics o*=lxlQle[Q(x)(e)] so that (hon-o)*= lQle [Q(book)(e)]. (Note: l: 'lambda'.) In contrast, particle de which indicates a location has the semantics de*=lxlQle [at(e)=x & Q(e)] and at the store is (mise-de)*=lQle[at(e)=store & Q(e)]. Particle o is a predicate of events and individuals, selecting a transitive VP in need of a direct object. Particle de is a predicate of events only. Adopting the semantics for classifiers found in Krifka (1995), [hon-3.satu]-o 3. books has the semantics (hon-3.satu-o)*=lQle [Q(book)(e) & RT(x, volume) & OU(volume)(book)=3] where a book is a volume and the number of volumes which are books is 3. Semantic agreement between the noun and its classifier is confirmed at this point. Likewise, an adjunct expression such as at 3 stores is (mise-3.ken-de)*=lQle[at(e) = store & Q(e) & RT(x, building) & OU(building)(store)=3]. In both cases, the resulting expression is a predicate of events selecting for a saturated predicate.
Recall that the floated quantifier is selected by an empty-headed adverbial. I propose the following semantics for this adverb: lQlxle[Q(x)(e)], so that (3.satu-)*=lQlxle[Q(x)(e) & RT(x, volume) & OU(volume)(book)=3] and (3.ken-)*=lQlxle[Q(x)(e) & RT(x, building) & OU(building)(store)=3]. Combining with a verb, say kau*= lxle[buy(x)(e)], results in (3.satu- kau)*=lxle[buy(x)(e) & RT(x, volume) & OU(volume)(book)=3] which may then combine with the direct object giving (hon-o 3.satu- kau)*=le[buy(book)(e) & RT(x, building) & OU(building)(store)=3] which is exactly right for buy three books. Crucially, the adverbial classifier is a predicate of events and individuals, and it is defined so that its QP must be construed via lx as counting an argument of the verb. Agreement between the classifer and the direct object is forced by the fact that the variable in the classifer and the variable in the predicate must be the same.
In contrast, (3.ken kau)*= lxle[buy(x)(e) & RT(x, building) & OU(building)(store)=3] is grammatical, but it can mean only buy 3 buildings. Again, the adverbial classifier is defined so that its QP must count an argument of the verb. Quite likely, there will be an agreement mismatch between the actual direct object and the adjunct classifer. If the verb is intransitive to begin with, as in (3.ken matu)*=le[wait(e) & RT(x, building) & OU(building)(store)=3], the variable x introduced by the classifier will never be bound. In either case, the derivation will be ill-formed. Classifiers floated off of adjuncts cannot be correctly interpreted. If by chance there is no mismatch between the adjunct classifer and the direct object, the classifer will simply be understood as counting the direct object and not the adjunct.
In short, there is one kind of QP which is predicated of individuals. This QP is selected by various kinds of XPs: PP, DP, and adverbial PP, all of which are of different semantic types. PP is predicated only of events, while DP is predicated of both events and individuals. Given the representations outlined here, DP classifiers can float (and PP classifiers cannot) because the adverbial PP is also defined as a predicate of both events and individuals. Such a representation forces semantic agreement between the classifer and an argument of the verb. The core of the analysis explains how an adverbial can agree with an argument noun. The agreement is built into the semantics and is not, strictly speaking, syntactic. This has been a problem in more than one language (French comes to mind), and to the degree possible other languages will be addressed. In addition, the analysis here might also be extended to a number of quantifying expressions in Japanese (existential, universal, etc.) all of which have the syntax of adverbials.


Fukushima, K. (1991) Phrase Structure Grammar, Montague semantics, and floating quantifiers in Japanese. Linguistics and Philosophy 14. 581-628.
Fukushima, K. (1993) Model theoretic semantics for Japanese floating quantifiers and their scope properties. Journal of East Asian Linguistics 2. 213-228.
Inoue, K. (1978) Nihongo no Bunpoo Kisoku [Grammatical Rules of Japanese] Taisyuukan, Tokyo.
Kamio, Akio (1983) Meishiku no kozo. In Kazuko Inoue (ed.), Nihongo no Kihon Kozo. Sanseido, Tokyo.
Kitahara, H. (1993) Numeral classifier phrases inside DP and the specificity effect. Japanese/Korean Linguistics 3. 171-186.
Kratzer, A. (1996) Severing the external argument from its verb. In: J. Rooryck and L. Zaring (eds.), Phrase Structure and the Lexicon. Kluwer, Dordrecth and Boston. 109-137.
Krifka, M. (1995) Common nouns: a contrastive analysis of Chinese and English. In: G. Carlson and F. J. Pelletier (eds.), The Generic Book. University of Chicago Press, Chicago. 398-411.

< Back to List of Abstracts | Back to PLC23 Home Page >

About the PLC23 Committee
Previously held Penn Linguistics Colloquium: PLC22 (1998), PLC21 (1997)

Penn Department of Linguistics
University of Pennsylvania