RICHARD SPROAT AND CHILIN SHIH THE CROSS-LINGUISTIC DISTRIBUTION OF ADJECTIVE ORDERING RESTRICTIONS* 1, INTRODUCTION English displays well-known restrictions on the ordering of multiple prenominal adjectival modifiers (see Bloomfield, 1933; Whorf, 1945; Lance, 1968; Vendler, 1968; Quirk et al, 1972 among numerous others). Most descriptions include a hierarchy such as the following: QUALITY > SIZE > SHAPE > COLOR > PROVENANCE: (1) shows that the preferred orderings are in line with this hierarchy: (1a. SIZE > COLOR > PROVENANCE: small green Chinese vvase (‘green small Chinese vase, *green Chinese small vase, ‘72small Chinese green vase...) b. QUALITY > SHAPE: nice round plate (“round nice plate) SIZE > SHAPE: small square table (*square small able) While it has been noted that such hierarchies are not absolutely rigid, speakers of English have a strong intuition that the above orders are basic? If we turn now to another language, Mandarin, we might be led to the conclusion that there is no evidence for such restrictions in that language Inthe following examples, all of the indicated orderings are fine: Qa. SIZE,COLOR: xiad-de lide _huaping small-DE. green-DE vase small green vase lide xifo-de hudping small green vase b. QUALITY, SHAPE: hio-de yudn-de —_panzi good-DE round-DE plate nice round plate yudn-de hio-de panzi nice round plate 965Now, RICHARD SPROAT AND CHILIN SHIH in (2) is marked with the particle di so used to mark h of the adject it is also possible to use monosyllabic adjectives in Mandarin to modify noun phrases without the use of de (Li and Thompson, 1981, pp. ich cases ordering restrictions such as those found We note here that such modificat two de-less adjectives; we will provide an expl So, in (4) we see that SIZE > COLOR, and SIZE > SHAPE restrictions do hold in small green vase small green vaseADJECTIVE ORDERING RESTRICTIONS (4c SIZE > SHAPE: xido fing zhudzi small square table small square table “fing xiio zhudzi In this paper, we shall show that the different behavior of (1) and (4) on the one hand, and (2) on the other correlates with a difference in the manner of @-role assignment (or discharge) in the first two cases versus the latter. We shall argue that adjectival modification cross-linguistically breaks down into two kinds, both of which are exhibited in Mandarin. In the first kind, which we call ‘direct’ modification and which is exemplified in (1) and (4), the adjective assigns its 8-role(s) directly to its sister, which will be a projection of N, as indicated below:* role(s) In the second kind, which is exemplified in (2), the adjective’s @-role(s) are associated with that of its modifiee indirectly by coindexation. In the case of the de-modifiers in Mandarin, we shall argue that the modifier clause is a relative clause; thus the adjective assigns its 8-role(s) to a phonologically empty variable within the modifier clause, this variable being bound by an operator which is coindexed with the head of the entire noun phrase. We indicate this structure below (note that CP = $” and IP = Sin current GB usage; see (Chomsky, 1986)) ©) NY “\ ® Ct mae @-role(s)568 RICHARD SPROAT AND CHILIN SHIM We style this ‘indirect’ modification, The crucial point here is that unlike the structure in (5), the adjective is not assigning its @-role(s) directly to the N* which it modifies; we shall argue below that Arabic has indirect ‘modification in this sense, though an adjunction rather than relative clause Structure is involved in this language. So, we suggest that it is wrong to iow adjectival modification as a unitary phenomenon cross-linguistically ‘Any given language may exhibit one or both of the kinds of modification discussed, "Among instances of direct modification in particular it is also necessary to distinguish parallel and hierarchical modification. Hierarchical modif- ‘cation is diagramed in (7a). So, each adjective assigns its O-role(s) directly to ite sister and the whole structure is hierarchical Parallel modification is diagramed in (7b), using the notation for parallel structures developed in Goodall (1987). Here, the adjectives assign their @-role(s) directly to the head nominal independently of one another: Ms. We shall argue below that while Mandarin and (generally) English direct modification ate instances of hierarchical direct modification, there are Janguages such as French which exhibit parallel modification. ‘We now outline the primary claim we wish to make in this paper: (8) __ Restrictions on the ordering of multiple adjeetival modifiers — henceforth AOR — obtain iff the adjectives involved are hier- irchical direct modifiers We shall show that this claim is substantiated for a variety of languages from several language families "Although establishing the correctness of (8) is our primary goal in thisADJECTIVE ORDERING RESTRICTIONS 569 paper, there are some other issues which we shall address. First of all, we will observe that when AOR occur in a language, the ordering hierarchy is, at least to a first approximation, the one observed for English; we have already scen that this holds true for Mandarin in (4). So, the gross hierarchy given in the introduction at least seems to be universal One question which then arises is whether AOR are a statement about left-to-right ordering — henceforth the ‘linear ordering theory’, or whether they refer to structural distance from the head noun — henceforth the ‘head-proximity theory’.* We shall argue that the head-proximity theory is correct, although this necessitates an interesting but not unmotivated analysis of Celtic noun phrase structure. Finally, we will make some observations about the semantic and cognitive basis for AOR. We shall argue that the semantic property of absoluteness gives a clear first cut at predicting the ordering restrictions. We shall also show that absoluteness accounts for the property of Mandarin observed in passing above, namely that only two direct modi= fiers are possible. We shall point out that a crucial piece of the puzzle is ‘missing: why is it that AOR should be confined to cases of direct hierarchical modification cross-linguistically? In the next two sections we explicate the difference between direct and indirect modification, we argue that the primary claim in (8) holds cross- linguistically, and we show that AOR are cast in terms of head-proximity In the fourth section we discuss the role of absoluteness in determining AOR. 2. THE SYNTAX AND SEMANTICS OF ADJECTIVAL MODIFICATION 2.1. Direct Modification Higginbotham (1985) suggests that adjectival modification frequently involves a mode of @-role assignment which he calls é-identification, So, red apple has the structure in (9) where the @-position of the adjective and of the noun are identified: (9) (N’,)) al) «N.)) | X= red apple570 RICHARD SPROAT AND CHILIN SHIH “The semantics of identification is intersective, so that a red apple is something which is both red and an apple; this is of course the traditional treatment of such absolute adjectives (see Siegel, 1980). Adjectives such as big which refer to relative properties are slightly more complicated in that they involve both @-identfication and what Higginbotham calls ‘autonomous O-marking, as indicated in (10) (40) way) ae aS L—— (A,(,2)) aay fescue | | bie apple The meaning of this is something which is both an apple, and big for an apple. The interpretation ‘for an apple’ is the resull of autonomous & ‘murking from the second 8-position of A (and see again, Siegel, 1980). Now, specifiers such as the or John’s function to bind the O-role of a noun, thus discharging it and making it unavailable for further @-role assignments Ths O-binding is indicated by an asterisk as shown in (11:7 (ily (OP, (1*)) Y aa ~ ba (xP,(1)) \ wy | | the dog ‘One consequence of this analysis is that it should be impossible to have adjectival modification outside specifiers, since the @-role of the noun is hho longer available for 6-identification. Of course, there are adjectives which do not involve é-identification, so-called intensional adjectives such ns alleged; in Higginbotham’s analysis, such adjectives merely involve autonomous @-marking. However, the traditional semantic treatment of stich adjectives as functions from common noun meanings to common noun meanings (see Siegel, 1980) would lead one to suppose that they 100 must occur within the scope of specifier material. Specifiers are functionsADJECTIVE ORDERING RESTRICTIONS 371 from common nouns to referential expressions; therefore intensional adjectives cannot modify specified (referential) noun phrases. The upshot is that such adjectival modifiers as we have been discussing may combine with NPs, but not with DPs ‘That this prediction is corr de-less modifiers is easy to see: ct for both English adjectives and Mandarin (12) *good the boy, *brown this rat, *round John’s table, *alleged that communist (13) *héng zhéige pin red this apple “yudn méige zhudzi round each table “qin nie zéngiéng former that president We shall henceforth call ‘direct’ modifiers any adjectival modifiers which directly assign Groles to the nouns they modify in one of the ways described above. So, direct modifiers must occur within the scope of specifier material, and hence cannot modify DPs; this necessity of occur ring within the scope of specifiers, since it is derivable as we have seen, may be taken as the hallmark of direct modification, In addition, as we saw in the introduction (cf. examples (1) and (4)), AOR apply to string: c direct modifiers in both English and Mandarin, ‘We note in passing that while direct modifiers cannot modify DPs, the level of projection of N selected by direct modifiers can in principle vary ‘across languages. In fact, Mandarin and English differ on this. In English, adjectives can modify NPs as i: (14) [sptapia [yp Russian [ypinvasion of Afghanistan] In Mandarin, on the other hand, direct adjectives would appear to be restricted to modifying N°. Modifiers with de, which as we shal see below can occur either inside or outside specifiers, must nonetheless occur outside direct modifiers: (15). héi-de sido shi black-DE small book small black book b.txiio héide shi small black-DE book We can suppose that this is due to a difference in the level of projection of the modifiee allowed by the two kinds of modifiers. Following Zhusv ap tm soana9 ‘ajqeuosear swoas 11 ‘Squ9a samisuenuT pur seanoatpe jaox9 ou 3 Sun Supa yung Sop roy ont no® myzigu “a. uy spaig af pag wD iq. snvsoup ep Bupysugy 2 ‘suadxo are sud aysuadsa wad “4 poog st soyeom oun poo8 smpvan “OD w se (cpr “d “1961 ‘uosdwouy, pue aieq SB 1n990 UED $9) aifpowy maapuy "T'C upunpunpy wu uonexaputoo ea Aq}oaurpUt yeunwou peoy aun yn parerposse Fuswos9q (s)>I0%9 1 paypal Jo aimyonns 2 jayres ing ApoaMIP uNOU Peay M7 OF (3}2I02 jou) UBIsse OU OP "HOW PA198q0 100 np andive 1x9u a4 SHOUIPOU wioas YOY ‘voHanpomt 2 p aa «suonoaford 1948 searoqy ‘N J0 ston2afoud jano-piom Kpour Kuo siamrpou 19auIp (961) HIHS NITIHO GNV LVOwas GUVHONE zsADJECTIVE ORDERING RESTRICTIONS 573 relative clauses with intransitive verbs in (19), as also assumed by Li and ‘Thompson (1981, p. 118) (18) a. hifo-de — tiingt ‘good-DE weather good weather b. guide bi expensive-DE. pen expensive pen « dirde _ kéngling big-DE dinosaur bigdinosaur (19)a. fide io ADE bird the birds which are flying b. nbizhi jido-de — gou that bark-DE dog the dog which is barking ©. ldide —chézi come-DE car the car which is coming Under standard assumptions, relative clauses are represented as having an operator in a non-argument position — usually COMP — binding a variable in an argument position; the operator would be coindexed with the head of the nominal. Assuming the same structure for de-adjectives would mean that the structure of (I8¢) is as diagramed schematically below (see also (6)): (20) fle}, 44] 0,) ‘That is, da ‘big’ assigns its @-role to an empty category [e] in the subject position of the de-ciause via predication, just as it assigns it to the overt noun phrase in (16c). This empty category is bound by an operator O, tng] which is further coindexed with the head of the noun phrase, Now, in a relative clause construction such as: 21) man, whofe], walks ‘man is identified via coindexation with the chain who . .. fe), which is assigned the external @-role of walk. The reference of such a phrase is574 RICHARD SPROAT AND CHILIN SHIH taken to be those things whieh are both men and walk. Similarly, in (20), coindexation between the chain O.... fe and the head kéngléng will yield the interpretation ‘those things which are dinosaurs and big (for such).’ Given this analysis, one would expect that adjectives which cannot occur as predicates can also not occur as de-modifiers. This prediction seems to be correct. So, as also observed by C. Huang (1987), the adjectives qidm ‘former’ and wai ‘fake’ cannot occur as de-modifiers (22: b), and this correlates with the impossibility of using them as predicates (2c, 4): a. tgién-de —_zdngt6ng former-DE_ president ce. tzhbige zingting gin this president former d.*nvifa yo wei that medicine fake These two adjectives can only occur as direct modifiers: (23) a. qin zéngténg former president former president b. wai yho fake medicine fake medicine Now, in English, relative clauses can be used either restrictively or non- restrictively (appositively) according to whether the relative clause is inside or outside the scope of the specifier (see Jackendoff (1977) for a similar analysis) (24) a. [[this man] [who is walking]} is speaking (= this man, who is walking, is speaking} b. [this [man [who is watking]]] is speaking ‘walking is speaking) this man who is ‘The interpretation of these would be roughly as follows: [this x1 man‘(x)] is speaking & he, is walking . [this x1 (man’(x) & walking’(x))] is speaking So, another prediction of the relative clause analysis of de-modifiers is thatADJECTIVE ORDERING RESTRICTIONS 575 such modifiers in general and adjectival de-modifiers in particular should be able to occur either inside or outside specifiers. This prediction is correct for Mandarin: 26) a. héng-de zhéibén shi zai zhudzi_ shang. red-DE this book at table on This red book is on the table. a’ ahéibén héng-de shi zhi zhudzi shang, this red-DE book at table on ‘This red books on the table. b. hui jito-de ndizht gou ziu-e. can bark-DE that dog leave-ASPECT ‘That dog which ean bark lett niizhi hui jido-de gou z6wle. that can bark-DE dog leave-ASPECT That dog which can bark left. ‘Assuming that the semantic translations of the two examples in (26a, a’) ‘are parallel to the non-restrictive and restrictive relatives in English, the interpretation would be as follows: (27). [this x1 book'(x)], is on the table & it; is red a, [this x1 (red’(x) & book’ (x))]is on the table The interpretation in (27a), which we have given for (26a), directly contradicts the claim of C. Huang that Mandarin does not have non~ restrictive relatives. In addition, he also denies something that we have tacitly assumed, namely that Mandarin has true specifiers, words which we have been glossing as ‘this’ or ‘that. He suggests that phrases like zhéibén shi ‘this book’ do not automatically denote individuals in the way that their English counterparts do: when such phrases occur within the scope of a de-modifier, they do not refer to a unique individual but rather have the type of common nouns. (When zhéibén shit occurs as an independent noun phrase, C. Huang assumes that there is a type-lowering operation which turns the noun phrase into a referential expression.) However, there seems to be evidence that Mandarin does have non-restrictive relatives, and that ‘this’ and ‘that’ are specifiers. With respect to the second point it is sufficient to note that specifiers cannot be doubled. Consider the contrast in: (28) a. *zheibén néixié wo mai-de shit this those I buy-DE book this one of those books that I bought576 RICHARD SPROAT AND CHILIN SHIH (28)b. yibln_néixié wS mii-de shi ‘one those I buy-DE book one of those books that I bought (28a) could in principle have a perfectly reasonable interpretation (note the English translation), in parallel with the well-formed (28b). The ‘ungrammaticality of (28a) suggests that in Mandarin, as in English, having two specifiers is simply structurally impossible. Note also that ‘this’ and ‘that’ which are specifiers in Mandarin contrast in behavior with posse: sives, which function as specifiers in English but not in Mandarin. Posses sives in English are in complementary distribution with all other specifier ‘material, whereas in Mandatin possessives may be stacked and may co ‘occur with one specifier: (29) a. “John’s Fitzgerald's this book b. Zhingsin-de Fitzgerald-de zheibén shi Z-DE F-DE this book (0Z, this book by F. belonging ‘The question of non-restrictive relative clauses or modifiers in Mandarin is more complex, and we only have space for a short discussion. We review three arguments that de-modifiers outside the scope of specifiers are plausibly analyzed as being non-restrictive relatives. First of all, our alysis of phrases like héng-de zhéibén shi with the accompanying interpretation in (27a) assumes that relative clauses can modify already referential expressions such as this book, Now, C. Huang observes that the Mandarin equivalents of John, who just arrived are ill-formed. While this is true, we observe that a large class of presumably identical constructions are perfectly well-formed so long as the non-restrictive relative refers 10 ‘an inalienable of salient property of the head:"” (80) qiniin shéng-de Zhiingsin last-year born-DE Z. Z.,who was born last year b. Céo Xuéqin side Hong Léu Meng CX write-DE red chamber dream The Dream of the Red Chamber, which was written by Cao Xuegin c, cSngming-le Lis smar-DE LL. Lisi, who is smartADJECTIVE ORDERING RESTRICTIONS S77 Since such constructions are quite freely formed, it seems ill-considered to rule them out in principle for Mandarin. Secondly, in line with the interpretations given in (27) for (26a, a"), one ‘would assign the interpretations in (31a’,b’) to the sentences in (31a, b): Bl) a, wo-de chézi you EDE car have big-DI My car has two doors, which are big. a’, my carhas [two x1 door’ (x)], & they, are big b. wide chézi you Tiingge divde mén, EDE car have wo big-DE door My car has two big doors. ’, my carhas [two x! (big’ (x) & door’ (x))] ture generated by the two examples is different and is identical to that of their English counterparts, as predicted by the interpretations wwe have given. The second example is felicitous if the speaker's car has two or more doors, two of which are big. The first example is only felicitous when the speaker's car has exactly two doors which furthermore are big; this implicature can be derived from Grice’s (1975) Maxim of Ouantity Finally we consider sentences such \ (32), discussed in Chao (1968); (32) dai ydnjing-de —_néiwai xidnshéng zi shudhua. wear eyeglasses-DE that gentleman is speak ‘That gentleman who is wearing eyeglasses is speaking, Chao implies that the interpretation of dai ylinjing-de ‘who is wearing eyeglasses’ is restrictive in this example. On the other hand in English that gentleman, who is wearing eyeglasses, is speaking, which we claim to have the same structure as Mandarin (32), the relative clause is clearly non- restrictive. We suggest that this discrepancy is due to the linear ordering difference between the two la Specifically, in English, given that who is wearing eyeglasses is outside the scope of the specifier, and since it comes after the phrase shat gentleman, it merely serves to further describe an already established referent. In Mandarin, however, the relative clause dai yanjing-de comes first, and so picks out the relevant set of gentlemen, Naiwei xidinshéng then merely further describes the established individual, Consistent with this is the observation that the most appropriate usage of (32) is when there are several gentlemen, only one of which is wearing eyeglasses; (32) is therefore equivalent to English the one wearing glasses, that guy, is speaking; this is along the lines of the analysis proposed by J. Huang (1982, pp. 68-70) for such constructions.578 RICHARD SPROAT AND CHILIN SHIH ‘To summarize what we have argued so far. Mandarin adjectives with de, indirect modifiers, are relative clauses and so their @-roles are not assigned directly to the nouns they modify. In this way they contrast with bare adjectives in Mandarin — and all adjectives in English, which we have called direct modifiers, We also showed that indirect modifiers should not be constrained to occur inside specifiers, and this we may take as a hallmark of indirect modification. Moreover, direct modifiers are constrained by AOR, whereas indirect modifiers are not, as (2) shows. Before we show that the domain of AOR in fact patterns as in English and Mandarin cross-linguistically, we need to briefly discuss parallel modification, 2.3. Parallel Modification We have been assuming that multiple modification structures are generally hierarchical. However, there are cases where the structure is plausibly parallel rather than hierarchical, As also noted by Nowicka-Schwartz (1980) for Polish, treating the prenominal modifiers as separate intona- tional phrases frees up the modifier order to some extent. So, for example, (33) seems felicitous with appropriate intonation, here indicated by Commas, despite the fact that the unmarked ordering is QUALITY > COLOR > PROVENANCE: (33) She loves all those Oriental, orange, wonderful ivories, Following Beckman and Pierrchumbert (1986) we will assume that in such cases the adjectives each constitute a separate (minimally) intermediate phrase. Furthermore, as Beckman and Pierrehumbert suggest, such an intonational phrasing is consistent with the analysis that such modifiers are interpreted in parallel. This is because clearly parallel structures such as coordinates are implemented intonationally as (minimally) intermediate phrases: (34) They gave watermelons, oranges, and berries. “The intonation of (33) is closer to the intonation of (34) than it is to that of (35), which can be said felicitously without comma intonation: (35) _ She loves all those wonderful orange Oriental ivories. If the structure of (33) is one of coordination, then, we may fairly assume that the adjectives involved are each modifying the head noun independently, or in parallel, rather than in sequence. Hence, one might speculate that AOR should not apply here since the adjectives are not cach modifying the entire structure consisting of the head and all adjee- tives closer to the head, but are rather each modifying the head in parallelADJECTIVE ORDERING RESTRICTIONS 319 ‘The structure that we assume would therefore be parallel, in the sense of Goodall (1987): (36) Oriental —_ orange wonderful 3. THE DOMAIN OF AOR ‘We now turn to the central claim of this paper, which we repeat below: (37) AOR obtain iff the adjectives involved are hierarchical direct modifiers. Put another way, AOR as exhibited in English and Mandarin are ‘universal, but they may fail to show up in some language simply because that language fails to have hierarchical direct modification. In general we expect multiple adjectival modification structures 10 exhibit the clusters of properties described in one of (a) or (by (38) ai. AORand ii, evidence that the adjectival modification involved is direct; in particular, the adjectives must be restricted 10 ‘occur within the scope of specifiers. b.i, NoAOR, and cither ii. evidence that the adjectival modification involved is indirect; in particular, the adjectives will not be restricted to occur within the scope of specifiers; furthermore, there may be overt evidence, as in Mandarin, for treating the ‘modifiers as relative clauses or appositives, or iii, evidence that the adjectival modification involved is parallel We shall show that adjectival modification across languages does in fact show one or other of these clusters of properties, as predicted by (37) and as we have seen holds in English and Mandarin direct modification (cohibiting (38a)), and in Mandarin indirect modification (exhibiting (38bi, biiy), Before turning to that we need to allay one possible misunder-580 RICHARD SPROAT AND CHILIN SHIH standing. It has been suggested to us that since one of the tests for direct- ‘modifier status is strict ordering within the scope of specifiers, a simpler statement of the universal would be the (uninteresting) claim that, on the assumption that specifiers are modifiers, modifiers are either ordered in a language of they are not. This suggestion is off the mark. As noted in the introduction, and as noted by most other work on AOR, the restrictions represent preferences which can nonetheless be relaxed in appropriate contexts. On the other hand, there is no way (o relax the restriction that modifiers must occur within specifiers in, say, English. This difference strongly suggests that AOR and restrictions on the ordering of specifiers and other modifiers cannot be collapsed, and that the ordering of specifiers outside direct modifiers is structurally required as we have argued, ‘We now turn to the cross-linguistic support for (37). In no case have we extensively examined the language in question; our purpose rather is to show that the properties noted in (38) appear to cluster as predicted. Also, it will be obvious that, at least to a first approximation, in all eases where AOR hold, they are the same ordering restrictions as in English and Mandarin, The only other question will be whether AOR are a statement about linear ordering or about head-proximity. We shall see that head- proximity y seems to be correct. 3.1. Languages with Direct Modifiers Some languages which exhibit the cluster of properties associated with direct modification are Dutch, given in (39), Greek in (40), Kannada in (41) and Mokilese in (42) (Harrison, 1976; note that DETERMINER is enclitic on the last word in the noun phrase), The orderings in the multiple modifier examples below are the preferred orderings, other orderings being much less acceptable: (39) a, SIZE > SHAPE > PROVENANCE: de grote ronde chinese vaas the big round Chinese vase the big round Chinese vase b.tprote de vans big the vase (40) a. SIZE > COLOR > PROVENANCE: to mikro kokkino kineziko vazo the small red Chinese vase the small red Chinese vase