Count adjectives: position and interpretation

In this section I return to specific classifier > count adjective > N orders. Adjectives that express dimensions presuppose a division of stuff (dimensions arise with bounded units, bounded units arise by division). How is it possible, then, to arrive at a classifier > dimensional adjective order? I suggest that the answer depends on whether an order like this has a type or a token interpretation.

3.4.1 Token readings

I propose that when the classifier > dimensional adjective order gives rise to a reading in which units or tokens are modified (i.e. wheretwo classifier big apple means two apples which are big), then the adjective is merged above the Cl position, as D´ek´any and Csirmaz (2010) suggest. This follows from the syntax-semantics mapping I assume: ifbig modifies units, then compositionality dictates that it must be merged after units are created, viz. above Cl. Then the classifier >big order arises because on the surface, the classifier sits in the highest classifier-related head, Num.

This is precisely what we have seen with the general classifierdarab.

On a token interpretation, then the classifier > dimensional adjective arises if i) the classifier is a pure Num classifier without a divisor function, and is merged in Num in the first place, ii) the

classifier is a Cl classifier that moves to Num, or iii) in a spanning framework it is also a possibility that the classifier spells out both Cl and Num, and linearizes in the higher Num position.

Either the second or the third option materializes for some speakers of Mandarin, for instance.

Mandarin classifiers are prototypical divisors, therefore they must be an exponent of the Cl position.

It has been claimed in the literature in various places that adjectives must generally follow Mandarin classifiers (c.f. Cheng and Sybesma, 1998, 1999 and Borer (2005), though they all acknowledge that certain exceptions do exist, and not all speakers find adj > Cl sequences equally bad). An example from Cheng and Sybesma (1999) is reproduced in (78).

(78) a. yi

(Cheng and Sybesma, 1999, p. 516., ex, 15. b.) Mandarin The examples in (78) are reminiscent of the distribution of Hungariandarab: the interpretation is a token interpretation, and the order is classifier > dimensional adjective. The token interpretation of (78) means that the merge-in order is dimensional adjective > Cl: first dog stuff is partitioned into dog units, and then these dog units are modified by the adjectivebig. Given that Mandarin classifiers are true divisors, they must be merged in Cl. The surface order then arises by classi-fier movement to Num (alternatively, Mandarin classiclassi-fiers span Cl and Num are are spelled out in the latter position). The merge-in position of Mandarin classifiers is transparently visible in those examples which admit the dimensional adjective > classifier order (according to Cheng and Sybesma, 1998, 1999, this is the exception rather than the rule).

(79) a. yi

‘a long film’ (Cheng and Sybesma, 1998, fn. 4.) Mandarin

Zhang (2011) (and the ample references she cites on p. 39) claims that the order in (79) is, in fact, widely accessible, and that classifiers can be generally preceded by adjectives (her examples feature dimensional adjectives).

‘one big good guy’

b. san

(Zhang, 2011, p. 34, exx. 104 c-e) Mandarin

She claims that (dimensional) adjective > classifier and classifier > (dimensional) adjective orders are interchangeable without a concomitant difference in meaning. Observe the flexible position of the dimensional adjectives in (81).

(81) a. yi

c. yi

‘one small square photo’

(Zhang, 2011, p. 45, ex. 117.) Mandarin

As both orders in (81) have the same interpretation with the dimensional adjective modifying units, these adjectives must be merged above Cl in both orders. This is forced by the syntax-semantics mapping. The left hand side members of the pairs then show the base-generated order. The right hand side members of the pairs, on the other hand, feature a classifier that ends up in Num on the surface.²¹

3.4.2 Type interpretations

If the classifier > dimensional adjective order gives rise to a type interpretation, then the dimen-sional adjective cannot be merged above the classifier. By compositionality, adjectives always modify their sister node. If a divisor is merged before the adjective, then units are created below the adjective, and the adjective must be interpreted as modifying units. An adjective that modifies types does not directly say anything about units, so in this case no units can be established inside the sister node of the adjective, and no divisor can be present below the adjective. Therefore the claim is that the classifier > dimensional adjective order with a type interpretation is base gen-erated, with the adjective being merged under the Cl node. That is, in this case the dimensional adjective is merged to a mass structure. As masses are unbounded and have no dimensions, the only feasible interpretation of this structure is to coerce the mass structure into a type interpreta-tion. Thusbig apple with no classifier between the adjective and the noun is interpreted as a big type of apple. If a classifier is merged on top of a [dimensional adjective; noun] constituent, then it creates units from a particular type of noun. In our case, two classifier big apple creates two units from a big type of apple.

Suggestive evidence for the correctness of this proposal comes from Thai. Thai classifiers are used in a different, wider variety of contexts than Hungarian classifiers. In certain cases it is even possible for the same classifier to appear multiple times in the same DP (c.f. Hundius and K¨olver, 1983; Singhapreecha, 2001; Jenks, 2006 and Piriyawiboon, 2010, among others). A full understanding of the data I review here requires an understanding of the previously mentioned factors. Therefore without pursuing in detail any particular theoretical mechanism that will ensure the correct distribution of classifiers in Thai, these data remain suggestive. Nonetheless I include them here because they provide potential support for my proposal in an interesting way.

Observe the contrast between (82-a) and (82-b). In (82-a) the dimensional adjective directly modifies the noun animal, without a classifier present in the DP. The interpretation is a type interpretation. In (82-b) minimally differs from (82-a) in the presence of a classifier. Crucially, the syntactic difference translates into a semantic difference: (82-b) has a token interpretation.

21Zhang (2011) proposes deriving the flexibility in (81) by base-generating both orders. Specifically, she proposes that the phrase introducing the classifier and the phrase introducing the adjective can be merged in either order because the functional sequence does not order them. As the interpretation is the same, Zhang’s proposal does not make for a compositional semantics of her structures. Given the syntax-semantics mapping I am assuming here, I must reject her analysis. In the end will end up agreeing with Zhang in that both merge-in orders are possible, but I will argue that the different orders correlate with different meanings. Since both orders in (81) lead to the same interpretation, one of them must be a derived order.

(82) a. sat animal

yai big

‘big animals’ (type of animal: elephants, buffaloos, rhinoceroses, etc. ) b. sat

animal tua cl

yai big

‘animals that happen to be big’ (e.g. a dog that is big for a dog)

(Cheng and Sybesma, 2009, ex. 17., citing Kookiattikoon, 2001) Thai Example (83) makes the same point: the classifier ensures the token reading, and without it a kind interpretation arises.

(83) a. khaw he

suu buy

baan house

yay big

‘He bought a big house. (a mansion, a big kind of house)’

b. khaw he

suu buy

baan house

laN-cl

yay big

‘He bought a big house. (a house which is big for a house)’

(Jenks, 2006, ex. 2.) Thai

In his discussion of the omissibility of Thai classifiers, Piriyawiboon (2010) makes the general-ization that NPs with a type reading cannot contain classifiers. These data support my hypothesis that without a classifier dividing stuff into portions, dimensional adjectives cannot modify tokens.²² Cheng and Sybesma (2009) do not examine the relative orders of classifiers and adjectives, but they comment on (82) (their (17)) in the following way. "In (17a), the size has already been wired in before we start singling out the individuals or units in which they exist. In (17b), we single out the individual first, and then say that this particular individual is particularly big." This ties in with my analysis perfectly.

To conclude this section, I proposed that classifier > dimensional adjective orders with a token modifying interpretation are produced by merging the adjective above Cl and linearizing the clas-sifier in Num. Clasclas-sifier > dimensional adjective orders with a type interpretation, on the other hand, arise by merging the adjective to a mass structure.²³

3.4.3 Representing the flexibility

The proposal outlined above means that classifiers and dimensional adjectives can be merged in either order. This requires a certain flexibility in the order of functional heads. If this proposal is on the right track, then Cl and the functional heads that introduce adjectives are not ordered by UG. Their ordering is free, but due to the assumed syntax-semantics mapping, the different base-generated orders have different interpretations. This is not entirely new. It is well known that intensional operator adjectives like former and alleged do not have a fixed position in the adjective sequence. They can be merged in multiple positions, but with a change in interpretation.

Specifically, intensional adjectives take scope over their c-command domain, or in other words, over their sister node and anything that is inside it. The different merge-in positions correlate with different c-command domains, and consequently with different interpretations.

22In the spirit of full disclosure, it must be acknowledged that not every noun displays the kind of alternation shown in (82) and (83). Jenks (2006) shows that only a subset of Thai nouns admit direct modification by an adjective, and others cannot receive any interpretation in this case. (Jenks, 2006, p. 7.) further claims that the instances of direct modification by a dimensional adjective are "often lexicalized into a true compound, indicating a likely tight syntactic and semantic relationship between the two". Compound formation in the case of direct modification is perfectly compatible with my suggestion. Everybody agrees that there is a tight local structural relationship between members of a compound. According to Harley (2009), for instance, the two parts of a compound noun must be in a first-modifier relationship. Therefore dimensional adjectives that form compounds with their head noun cannot be separated from the noun by a Cl projection, these adjectives must be merged below Cl.

23Dimensional adjectives cannot combine with the mass noun use of any noun, of course. Big wateris ruled out, even on a type interpretation. This, however, is real-word knowledge (big watercannot be assigned an interpretation that makes sense in our world), not a syntactic constraint. Dimensional adjectives can combine with mass structures on a type interpretation as long as the noun has the potential to be packaged into natural bounded units.

(84) a. a famous alleged murderer

(a person who is famous and allegedly a murderer) b. an alleged famous murderer

(a person who is allegedly a famous murderer)

Non-definite superlatives exhibit the same freedom of word order (Teodorescu, 2006).

(85) a. My class has a shortest Italian student.

(an Italian student who is shorter than other Italian students in the class) b. My class has an Italian shortest student.

(the shortest student in the class is Italian) (Teodorescu, 2006, p. 403., ex. 22.)

Thus the claim is that dimensional adjectives, like intensional adjectives and non-definite superla-tives, scope over their complement. If their complement has the size of a mass structure, they do not take scope over a divisor and receive a type-modifying interpretation. If their complement contains a divisor, then they take scope over the divisor and yield a token modifying interpretation.

This does not require dimensional adjectives to be operators, it follows from compositionality. Note that UG still plays a role in adjective ordering: it is responsible for ordering dimensional adjectives before, say, color adjectives, and size adjectives before shape adjectives. In languages that can or must make use of a covert divisor, the token vs. type modification may not or does not translate into a word order difference, due to the invisibility of the classifier. But the presence or absence of the divisor between the dimensional adjective and the noun is still detectable on the basis of the interpretation.²⁴