Adjectives in layers

Размещено на http://www.allbest.ru/

Adjectives in layers

А. Pereltsvaig, O. Kagan

* Santa Clara University

** Ben-Gurion University of the Negev



This paper investigates the syntax and semantics of modifying/attributive adjectives in Russian, a language lacking articles but having complex patterns of case marking and agreement within a noun phrase. It has been claimed in the literature that due to its lack of articles, Russian has a completely different internal structure for noun phrases than in languages with articles. In this paper we argue against that claim and propose that there are six layers of functional structure within a noun phrase which modifying adjectives can occupy.

Key words: noun phrases, attributive adjectives, modification, case marking, agreement, semantics.



А.М. Перельцвайг, О.А. Каган

* Университет Санта-Клары

** Университет имени Давида Бен-Гуриона в Негев

Прилагательные в иерархической структуре именной группы

В данной статье исследуются синтаксис и семантика аттрибутивных (модифицирующих) прилагательных в русском языке, в котором отсутствуют артикли, но имеются сложные модели падежного маркирования и согласования между элементами именной группы. В синтаксической литературе утверждается, что из-за отсутствия артиклей у именных групп в русском языке совершенно иная структура, чем у их аналогов в языках с артиклями. В статье показывается, что данное утверждение неверно, а модифицирующие прилагательные могут занимать один из шести уровней в функциональной структуре именной группы.

Ключевые слова: именные группы, атрибутивные прилагательные, модификация, падежное маркирование, согласование, семантика.



Introduction

The main aim of this paper is to investigate the distribution, ordering and interpretation of prenominal modifying adjectives. The syntax of modifying adjectives has been hotly debated for more than two decades, with proposals including analyzing adjectives as heads that take NP as a complement [Abney, 1987], as heads left-adjoined to N [Sadler, Arnold, 1994], as heads that take NP as a rightward specifier [Delsing, 1993], as adjoined to NP [Svenonius, 1993], or as specifiers of dedicated functional projections [Cinque, 1994; Scott, 2002]. In this paper, we will probe into the syntax and semantics of modifying adjectives by closely examining the relevant data from Russian, an article-less language with intricate patterns of intra-nominal case marking and agreement. Based on this investigation, we argue that there are several (six, to be precise) slots that prenominal modifying adjectives can occupy, sprinkled throughout the extended noun phrase (eNP).

The idea is not completely new: it has been explored in the “cartographic approach” going back to Cinque's study of adverbial modification in clauses [Cinque, 1999]. For a detailed analysis of adjectival modification along these lines, see Scott (2002); for the application of the cartographic approach to Russian adjectives, see Pereltsvaig (2007). However, we depart from these earlier studies in several significant ways.

First, unlike Scott (2002) and Pereltsvaig (2007), we base our argument not only on the ordering of various semantic classes of adjectives (e.g., Size, Age, Color, Material, etc.) with respect to each other, but on the ordering and interpretation of adjectives with respect to other elements of the eNP. In this, our approach is closer to that of Svenonius (2008): like him, we argue that the interpretation of a modifying adjective depends on the adjective's position with respect to functional projections inside the eNP. We propose that adjectives are generated in functional projections we call aPs which are sandwiched between other, independently-motivated functional projections, discussed in detail in Section 2 of this paper.

Another departure from Scott's (2002) and Pereltsvaig's (2007) approach is that we argue for a less fine-grained hierarchy of adjective classes than that proposed by these authors. Based on a reanalysis of the data from Pereltsvaig (2007) and some additional new data, we show that the less fine-grained hierarchy is supported by stronger judgments about adjective ordering. This issue is addressed in Section 3 of this paper.

Furthermore, in addition to adjective classes considered by Svenonius (2008), all of which occur after/below numerals (i.e., projections aP-3 through aP-6 in the tree in (1) below), we also consider adjectives that can appear before/above numerals, that is above NumP (i.e., projections aP-1 and aP-2 in the tree in (1) below). The projection aP-1 - the highest projection for adjectives - hosts such items as the poslednie-type adjectives of Babby (1987), adjectival possessors (e.g., masin `Masha's') and demonstratives such as etot `that'. In Section 4 of this paper, we show that these adjectives can appear either before or after numerals, with significant differences in meaning: when occurring above numerals, they are associated with referential, quantificational and/or exhaustive meaning, absent if the same adjective occurs lower in the structure, where property-modifying interpretation emerges. We propose that these differences in interpretation derive from the functional projection DP, and not from the adjectives themselves. Thus, these data (and our analysis of it) provide additional argument in favor of postulating a DP for Russian (contrary to [Boskovic, 2005, 2008, 2009, 2012]). Following Pereltsvaig (2006a), we maintain that some noun phrases in Russian are DPs, while others are Small Nominals. We examine several types of Small Nominals in Russian (cf. [Kagan, Pereltsvaig, 2011; Pereltsvaig, 2011]) and show that these nominals lack the upper layers of eNP and hence have no room for the highest types of adjectives; as expected these adjectives are impossible in such nominals. Thus, our evidence for distinct classes of adjectives is based on syntactic facts and the syntax-semantics interface and cannot be explained away by appealing to purely semantic or cognitive factors (cf. [Sproat, Shih, 1988, 1991], among others).

Finally, in Section 5, we examine an additional, often overlooked class of adjectives in Russian, which modify numerals rather than nouns in the eNP (i.e., the dobryx-class of [Babby, 1987]). We show that even though these adjectives are closely associated with numerals, they are not projected in the specifier in the functional projection of the numerals itself, NumP. Instead, we propose that these adjectives are projected in an aP of their own, above NumP but below the aP hosting poslednie-type adjectives. This strengthens our argument that the interpretation of adjectives depends (at least in part) on the independently motivated functional projections inside the eNP between which the relevant aP is merged.

It should be noted that for the purposes of this paper we focus on one- word modifying adjectives (e.g., bol'soj `big') leaving phrasal adjectival modifiers (e.g., gordyj svoimi uspexami `proud of own successes') aside.¹ While the question of whether such one-word adjectives are in the head or specifier of aPs is largely outside the scope of this paper, we adopt Pereltsvaig's (2006b) position that one-word adjectives are heads rather than phrases. Given this, we ignore the alternative possibility of phrasal adjunction of adjectives to the independently-motivated functional projections (e.g., poslednie-type adjectives being adjoined to DP, etc.); nothing in our analysis crucially depends on the choice between adjunction and separate aP functional projections for adjectives.



1. Independently motivated functional projections

In this section, we start our consideration of the structure of the extended noun phrase (eNP), in particular in Russian, by reviewing our views about the independently motivated functional projections that constitute the “skeleton” of the eNP, such as DP, NumP, UnitP, ClP, nP, etc. The functional architecture of the eNP that we ultimately adopt is shown in the tree in (1). The six positions for the six different types of adjectives are numbered, for ease of reference.

The highest independently motivated functional projection we assume is the DP, whose roots trace back to the work of Abney (1987). When it comes to article-less languages like Russian, a debate has been raging in the literature for some time as to whether such languages have the DP projection at all. Thus, some researchers argue that DP is found only in languages with articles ([Chierchia, 1998; Willim, 1998, 2000; Baker, 2003, p. 113; Trenkic, 2004] and most notably [Boskovic, 2005, 2008, 2009, 2012]). The opposing view, namely, that even in article-less languages a nominal may have the fully projected functional architecture, including the DP, is argued for in Rappaport (1992, 2001), Engelhardt & Trugman (1998), Progovac (1998), Leko (1999), Pereltsvaig (2001, 2006a, 2007), Rutkowski (2002a, b, 2006a, b, 2007), Franks & Pereltsvaig (2004), Rutkowski & Maliszewska (2007) and elsewhere. In this paper we provide an additional, albeit indirect, argument in favor of DP in article-less languages by showing that adjectives merged above and below this projection have a different morphosyntax and a different interpretation. Crucially, we show that the higher adjective position - aP-1 - is associated with referentiality and exhaustivity, properties that are known to characterize the DP projection in other languages.

(1) posledniecelyxpjat'stukbol'sixstaryx

lastwholefiveitemsbigold

belyxmedvedej

whitebears

plusevyx plush

`the last whole five items of big old plush polar bears'

Next, we assume the projection of NumP, which hosts numerals and other quantity expressions, including lower quantifiers like mnogo `many' and idiomatic quantity expressions like the PP do figa `lots' and even clausal quantity expressions like cert znajet skol'ko `devil knows how many'. Overall, this projection is uncontroversial (in some literature on Slavic it is called QP, but we adopt the label “NumP” so as to avoid a confusion of quantifiers like vse `all', which are merged higher). Moreover, we remain agnostic as to whether numerals like pjat' `five' are merged in the Num° or in Spec-NumP (see Bailyn 2004 for a discussion). As we discuss in Section 5 below, this choice is immaterial for the purposes of the present paper.

Furthermore, following Svenonius (2008), we adopt a three-way distinction between numeral classifiers merged in UnitP (cf. Borer's 2005 #P), sortal classifiers merged in SortP (cf. Borer's 2005 ClP) and noun classifiers merged in nP. Numeral classifiers make nominal referents countable or quantifiable, whereas sortal classifiers sort nominal referents by characteristics such as shape (see Svenonius 2008: 20). While in many languages, these two functions are performed by the same lexical items, there are a few other languages such as Akatek (as described by [Zavala, 2000]), where numeral and sortal classifiers co-exist.² In Akatek, numeral classifiers distinguish human, animal and inanimate nouns (here, only the inanimate one is shown). The sortal classifier distinguishes a dozen or more shapes (`smooth,' `long three-dimensional,' `erect,' `half-circle,' `round,' `wide flat,' `small spherical,' `separate,' etc.); note that the same noun can appear with different sortal classifiers, depending on how the referent is perceived. As shown in the examples below (from [Zavala, 2000, p. 117, 123]), numeral classifiers appear structurally outside sortal classifiers.

(2) a. kaa-b'sulan	aw-aan
two-INANsmoothA2-corncob `your two corncobs' ?os-eb'jilan'aan three-INANlong.3d corncob `three corncobs' kaa-b'b'ilan two-INANSMALL.ROUND	poon plum	yalixh-taj small-PL

`two small plums'

The third type of classifier is the noun classifier, which typically sorts nouns by material qualities or essences (cf. [Svenonius, 2008, p. 21]). Akatek has a set of fourteen noun classifiers (`man, `woman,' `animal,' `tree,' `corn,' `water,' `salt,' etc.) alongside the three numeral classifiers and the set of sor- tal classifiers, discussed above. All three types of classifiers are illustrated below (adapted from [Zavala, 2000, p. 126-127]).

(3) a. ?os-k'onkupan

three-ANIMHALF.CIRCLE

`three cows' (lying down)

b. ?os-eb'kupan

three-INANHALF.CIRCLE

`three (folded) tortillas'

c. ?os-eb'soyan

three-INANROUND

`three tortillas'

no' ANIMAL	wakas cow
?isim	paat
CORN	tortilla
?isim	paat
CORN	tortilla

The (simplified) structure for (3c) is shown below:

In Russian, only the first type of classifier is instantiated. As in Akatek and Squamish Salish, the set of numeral classifiers contains three items: stuk `items' for objects, golov `heads' for animals (especially livestock) and celovek `persons' for humans (there is also an archaic version for humans, dus `souls'). While Aikhenvald (2000, p. 115-116) claims that golov `heads' in Russian (as well as the corresponding items in English and Hungarian) are not numeral classifiers, we find her arguments rather weak or faulty. First, she maintains that these items “do not fill an obligatory slot in the numeral-noun construction”: while we agree on the facts (i.e., that numeral classifiers in Russian are optional), their position in the eNP is rigidly determined. Her second argument is that these items “often have a lexical meaning of their own”: while these items derive from lexical nouns meaning `item', `head' and `person', they do not contribute any lexical meaning when appearing as numeral classifiers. Her third argument that concerns the distinction between mass and count nouns does not apply to Russian (Aikhenvald illustrates it with Hungarian). Her forth argument is that golov `heads' and similar items occur with genitive complements; while this is true, we do not see it as an argument for the lexical nature of these items (cf. [Pereltsvaig, 2007] for arguments that numerals, which similarly take genitive complements, are not nouns either). Aikhenvald's (2000, p. 116) fifth and last argument is that “there is a restricted number of such words in a non-classifying language”; however, as we point out above, Russian mirrors Akatek and Squamish Salish in having three items in the set of purely numeral (i.e., non-sortal) classifiers: one each for objects, animals and humans.

There are three additional arguments for treating these items as numeral classifiers rather than nouns. The first argument has to do with the form of the classifier for humans, celovek: this special count form is peculiar for occurrences of `persons, people' together with a numeral. In the absence of a numeral (even if a quantifier like mnogo `many' is present), the plural form of celovek `person, man' is the suppletive ljudi `people', as shown below (cf. [Mel'cuk, 1985; Yadroff, 1999]):

More generally, stuk `items', golov `heads' and celovek `persons' cannot occur without a numeral being present; this is the hallmark of a numeral classifier (cf. [Yadroff, 1999, p. 152]):

(5) a. Javidelpjat' Isawfive `I saw five people.' Javidel(mnogo) Isawmany `I saw (many) people.' *Javidel(mnogo) Isawmany `I saw (many) people.'	celovek. person.PL.COUNT ljudej. person.PL celovek. person.PL.COUNT

(6) *Mynenaslistukjablok.

wenotfounditems.GENapples.GEN

Finally, stuk `items', golov `heads' and celovek `persons' used in this classifier function cannot take any modifiers (cf. [Yadroff, 1999, p. 151-152]):

(7) a. *desjat'	veselyx	celovek	nasix	oficerov
ten	happy	persons	our	officers
b. *pjat'	otdel'nyx	stuk	nexorosix	slov
five	separate	items	obscene	words

Therefore, contrary to Aikhenvald (2000), we take stuk `items', golov `heads' and celovek `persons' to be numeral classifiers; a similar position for corresponding items in Bulgarian is argued for by Cinque and Krapova (2007). According to our analysis, these numeral classifiers are merged in UnitP (cf. the structure we propose in (1) above). Note further that there is no adjective-hosting projection above UnitP, which explains why numeral classifiers cannot be modified directly (as shown in (7) above). The numeral classifiers are discussed further in Section 5 below.

The final, and perhaps the most controversial, functional projection that we assume in this paper is StageP, which turns entities to which Individual-level properties apply into entities to which Stage-level properties may apply. While the exact nature of the Individual-/Stage-level distinction remains illusive, at least some scholars analyze Stage-level properties/predicates as applying to the Event argument (or perhaps, a “Stage argument”) of the nominal (cf. [Kratzer, 1989, 1995]). It is our contention that StageP is the functional projection where this Event/Stage argument is added, similar to adding the Agent argument in vP (cf. [Kratzer, 1996]). This StageP projection becomes particularly relevant for structurally distinguishing between such Individual-level properties as material and origin, on the one hand, and such Stage-level properties as age and temperature, etc.; this is discussed in more detail in Section 3 below.

2. "Lower" adjectives

In this section, we focus on adjectives that modify the property denoted by the noun, such as bol'soj `big', staryj `old',pljusevyj `plush', belyj `white/ polar' in (1) above. Following Svenonius (2008), we reject the approach adopted by Scott (2002) whereby individual adjective classes (e.g., Size, Color, etc.) are associated with specific functional projections. As pointed out by Svenonius, “the categories [such as Size, Color, etc.] are not well-motivated outside of the adjectival ordering phenomenon that they are introduced to describe”. Furthermore, as we show below, the fine-grained hierarchy of adjective classes proposed by Scott (2002) is not supported by the actual adjective ordering facts; in particular, as we show below, ordering restrictions on some of Scott's categories, such as Age and Color, are far stricter than those on, say, Length and Height.

But first, we will briefly address the syntax of the so-called idiomatic adjectives, such as the ones in belyj medved' `polar [literally, white] bear', beloe vino `white wine', zeleznaja doroga `rail road [literally, ferrous] road'.

Cernoe more `Black Sea', Bol'saja Medvedica `Ursa Major'. First of all, note that adjectives appearing as idiomatic are not an exclusive lexical class: these adjectives can also appear as non-idiomatic, in combination with different nouns (e.g., belaja/cernaja kraska `white/black paint', zeleznyj zamok `iron lock', bolsaja kuca `big heap'). However, when occurring in such collocations, idiomatic adjectives do not make the usual, transparent contribution to the interpretation of the noun phrase (i.e., that of Color, Material, Size, etc.); for example, beloe vino `white wine' is not white but yellowish-green and zeleznaja doroga `rail road [literally, ferreous] road' of the toy kind may be made from plastic:

(8) a. zelenovatoebeloevino

greenishwhitewine

`a greenish white wine'

b. plastikovajazeleznajadoroga

plasticferrousroad

`a plastic railroad'

Note that the adjective with the idiomatic interpretation must be closest to the noun, hence the ungrammaticality of the following:

(9) a. *beloezelenovatoevino

whitegreenishwine

intended: `a greenish white wine'

(ok with comma intonation: `wine that is white in color with a greenish tint)

b. *zeleznajaplastikovajadoroga

ferrousplasticroad

intended: `a plastic railroad'

The same is true if we use two adjectives, each of which can potentially have an idiomatic meaning with the given noun: only the adjective that is closest to the noun will have idiomatic meaning in each case:

(10) a. belajabol'sajamedvedica

whitebigshe-bear

`a white Ursa Major'

b. bol'sajabelajamedvedica

bigwhiteshe-bear

`a big (female) polar bear'

This is in line with our general observation that the same lexical items can appear in distinct syntactic positions with distinct meaning (see also this section below and the following sections).

A further example is provided by the phrase belyj cernyj medved' `a white black bear', referring to a Kermode bear (or “spirit bear”, as the First Nations people call them). These bears are not albinos and not related to polar bears, but belong to the Ursus americanus kermodei subspecies of the American black bear. Thus, in belyj cernyj medved' `a white black bear' the adjective closest to the noun - cernyj `black' - is used idiomatically, to designate a species (and does not denote color per se), while the first adjective - belyj `white' - refers to the white (or cream-colored) coat of such a bear. Bruce Barcott describes the coats of the belyj cernyj medved' `a white black bear' thus: “more like a vanilla-colored carpet in need of a steam cleaning” [Barcott, 2011, p. 41]. Thus, in the structure we propose, belyj `white' in this phrase occupies the position aP-5, whereas cernyj `black' occupies the position A-6.

Idiomatic adjectives also differ from non-idiomatic adjectives in that they cannot be modified or extended by a complement, such as ocen' belyj `very white', belyjpo cvetu `white in color':

(11) a. *ocen'belajamedvedica

verywhiteshe-bear

`a very (typical) polar bear'

b. belajapo cvetumedvedica

whitein colorshe-bear

`a white in color (female) bear', not `polar bear'

Given the lexical nature of idiomatic adjectives, we propose to analyze them as part of a lexical compound, specifically as an A° merging with a N°. This position is marked as adjective position 6 in the tree in (1) above. This proposal is similar to what Sadler and Arnold (1994) propose for all prenom- inal adjectives; we limit this structure to idiomatic adjectives only.

Let us now consider non-idiomatic noun-modifying adjectives. Here, we propose that these adjectives can be divided into three groups (rather than thirteen or more classes, as done by Scott 2002), and, as shown in the tree in (1), we believe that these three groups of adjectives are merged in three different structural layers.

The first, highest group includes adjectives that can modify only count nouns, not mass nouns. Such adjectives include those denoting various dimensions of size (overall size, length, height, depth, width), as well as speed and shape. Since substances do not have intrinsic boundaries, one cannot talk about the length, height or width of a substance. Examples of these adjectives are given in (12a) below, and the structural position of these adjectives - above SortP and below UnitP - on our tree in (1) is marked as aP-3.

The second group includes adjectives that denote potentially Stagelevel properties that can apply to mass nouns as well as count nouns, such as weight, wetness, age and temperature. Examples of adjectives from those two types are given in (12b) below, and their structural position - above StageP, but below SortP - on our tree in (1) is marked as aP-4.

The distinction between these two groups of adjectives has been drawn already by Muromatsu (2001), who divides adjectives into two classes: those which are sensitive to shape and merge above classifiers and those which are not sensitive to shape and merge below classifiers. Since we adopt Svenonius' (2008) distinction between three kinds of structurally distinguished classifiers, we modify Muromatsu's proposal as follows: adjectives that are sensitive to shape merge above sortal classifiers (in SortP) and adjectives that are not sensitive to shape merge below sortal classifiers.

Thus, a close semantic connection is established between elements to be merged in the functional projection, in this case SortP (i.e., classifiers that “sort nominal referents by characteristics such as shape”) and the adjectives to be merged in the functional projection immediately above SortP: these adjectives denote properties that are sensitive to shape (and hence can apply only to count, not mass nouns).

This is a general pattern that obtains with respect to other projections that host adjectives and functional projections that constitute the skeleton of the eNP: at each level, the interpretation of a given adjective depends crucially on the functional projection immediately below the projection where the adjective is merged.

The third and structurally the lowest group of (non-idiomatic) noun-modifying adjectives include those that denote properties of color, material and origin. Examples of adjectives from this group are given in (12c) below, and their structural position - immediately above nP and below SortP - on our tree in (1) is marked as aP-5.

(12) a. Adjectives sensitive to shape (aP3): malen'kij `small', dlinnyj `long', vysolkij `tall, high', glubokij `deep', sirokij `wide', bystryj `fast', polyj `hollow'

b. Adjectives with stage-level meanings not sensitive to shape (aP4): legkij `light', mokryj `wet', staryj `old' and teplyj `warm'

c. Individual-level adjectives (aP5): cernyj `black', russkij `Russian' and zeleznyj `ferrous'

2 It should be noted that the distinction between adjectives merged in aP-4 I and those merged in aP-5 is not a distinction of gradability, as has been moften suggested (cf. [Scott, 2002]), nor does it correspond to the distinction іbetween kacestvennye vs. otnositel'nye prilagatel'nye (“quality vs. relative adjectives”), drawn by Russian grammars (cf. [Academy Grammar, 1980]). For instance, the category of otnositel'nyeprilagatel'nye `relative adjectives' in Russian includes not only such low aP-5 adjectives as those denoting material or origin (e.g., derevjannyj jascik `a wooden box', gruzinskoe vino `Georgian wine'), but also possessive adjectives which are structurally high, as discussed in Section 4 below (e.g., olina kniga `Olga's book'). Furthermore, even though adjectives denoting color can form adverbs in -o/-e (e.g., belo), comparative and superlative forms (e.g., belee `whiter', belejsij `whitest') and adjectives with the suffix -ovat (e.g., belovatyj `whitish'), which can be attached only to gradable adjectives [Kagan, Alexeyenko, 2011], and are thus classified as both gradable and kacestvennye prilagatel 'nye “quality adjectives), we classify them as belonging to the lowest category, structurally in aP-5, for reasons discussed in detail below.

Instead of relying on gradability or the morpho-semantic distinctions like the one underlying the distinction between kacestvennye (quality) vs. otnosi- tel'nye (relative) adjectives, we propose that the relevant property that characterizes adjectives appearing in aP-5, as distinct from those appearing in aP-4, is their inherently Individual-level nature. As can be seen from (1), aP-5 appears below the StageP projection, the one in which the event argument of Stage-level predicates is introduced. As a result, adjectives that merge above this projection, for example in aP-4, are compatible with Stage-level interpretation (although this interpretation is not obligatory, since an event argument need not be introduced). In contrast, adjectives that merge in aP-5 appear too low in the structure to receive Stage-level readings: these adjectives apply to the property denoted by the NP irrespective to any particular stage, situation or event, and are thus interpreted as denoting permanent, inherent, essential, Individual-level properties.

One way to distinguish between Individual- and Stage-level properties/ predicates in English is their grammaticality in existential constructions: only Stage-level properties/predicates are said to be possible [Kratzer, 1995]:

(13) a. There are firemen available.

b. *There are firemen altruistic.

Another test involves their grammaticality as secondary predicates: once again, only Stage-level properties/predicates are grammatical in such structures (cf [Rapaport, 1991; McNally, 1993]):

(14) a. I bought the dog sick.

b. *I bought the dog intelligent.s

When it comes to Russian, two tests emerge as distinguishing Individual vs. § Stage-level properties/predicates. The first test involves the pronoun doubling in colloquial Russian. As has been shown by McCoy (1998), this construction is possible only with essential, Individual-level properties/predicates:

(15) a.Morozenoeonoxolodnoe.

ice-creamit.NOMcold

`As for ice-cream, it is cold.'

b.*MorozenoeegoAllaljubit.

ice-creamit.ACCAllaloves

`As for ice cream, Alla loves it.'

The second test concerns the availability and use of short vs. long forms of adjectival predicates. As argued by Kagan and Alexeyenko (in progress), following traditional grammars, short forms typically denote Stage-level properties in the sense that they “denote properties that hold of an individual in a particular event/situation”, whereas (nominative) long forms “attribute a property to an individual without a dependence on a particular situation/ event which typically results in an individual-level interpretation”. See Soschen (2001), Geist (2010) and references therein for a discussion of the relation between short and long forms of adjectives in Russian and the individual-/stage-level distinction. (But see e.g. Bailyn (1994) for a different approach and some counterexamples.) When it comes to instrumental long forms, they “are relativized to a temporal interval, [but] this interval can be equivalent to an individual's lifetime, in which case the resulting meaning is close to that of an individual-level (nominative) predicate”. Crucially for our present purposes, short forms generally denote Stage-level properties; hence, if a given adjective does not have a short form, it is likely to denote an Individual-level property (except for a few rare cases of morphological gap in the paradigm).

With these tests in mind, we can now consider the categories of adjectives that occur, according to our analysis, in aP-5. Adjectives that are most clearly linked to this position are adjectives of material and origin, such as zeleznyj `ferrous' and russkij `Russian'. Both groups relate to properties that are generally conceptualized as permanent and inherent and that are not expected to change from situation to situation. In English, these types of adjectives are impossible in existential constructions (e.g., *There are chairs wooden / *There are songs Russian) and as secondary predicates (e.g., *I bought the chairs wooden / *I heard the songs Russian).

In (colloquial) Russian adjectives denoting material or origin can occur in the pronoun doubling construction, even in non-generic sentences:

(16) a.Etotkljuconzeleznyj.

thiskeyhe.NOMferrous

`As for this key, it's iron-made.'

b.Vanjaonrusskij.

Vanyahe.NOM Russian

`As for Vanya, he's Russian.'

Furthermore, the Individual-level nature of such adjectives in Russian is supported by the fact that they generally lack short forms. For example, adjectives such as derevjannyj `wooden', stekljannyj `glass-made', mednyj `brass-made', russkij `Russian', amerikansky `American', svedskij `Swedish', brjussel'skij `of Brussels' all lack a short form (cf. [Sochen, 2001, p. 6; Timberlake, 2004, p. 290]), which points to their inherently Individual-level nature.

A somewhat less obvious case is posed by the third group of adjectives that occupy aP-5, namely, adjectives denoting color. Whether these predicates are originally (unless coercion is involved) Individual-level or Stagelevel may be subject to debate. We believe, however, the basic meaning is indeed Individual-level, which is why the adjectives pattern together with those of material and origin. Even though the color of certain objects may be changed either by external factors (e.g., This chair was black but now it is red - we painted it) or by internal factors (e.g., This apple was green last week, but now it's red - it's ripened), unless forced by such special context, color is perceived as an inherent property. For example, when a recipe calls for green apples, it is understood as calling for Granny Smith apples, not unripe Red Delicious or Gala apples. The idea is that even though color may in principle be changed, we do not typically conceptualize this property as one that changes from situation to situation and is therefore linked to a particular stage or eventuality. Furthermore, adjectives of color pattern with other Individual-level predicates in the existential sentence test: *There are chairs green (cf. *There are chairs wooden, vs. There are chairs available). Similarly, color is not good (unless coerced) as a secondary predicate: *I bought the dog brown (cf. *I bought the dog intelligent, vs. I bought the dog sick). On coercion from Individual- to Stage-level property, see Chierchia (1995: 177).

Similarly, in Russian, adjectives denoting color can occur in the pronoun doubling construction, even in non-generic sentences (compare the sentence in (17a) with an adjective of color to the sentence in (17b) with an adjective of material):

(17) a.Etomorozenoeonozelenoe.

thisice-creamit.NOMgreen

`As for this ice-cream, it is green.'

b.Etomorozenoeonofistaskovoe.

thisice-creamit.NOMof.pistachios

`As for this ice-cream, it is made from pistachios.'

Furthermore, adjectives of color typically lack short forms. This is particularly true of descriptive color adjectives (i.e., those that are derived from the name of an object of that color); cf. [Sochen, 2001, p. 6-7]: there are no such short forms as *kremov (from kremovyi `cream-colored'), *kofeen (from kofeinyi `coffee-colored', *sokoladen (from sokoladnyi `chocolatecolored'), *persikov (from persikovyj `peach-colored'). Even when it comes to abstract color terms, their use in short forms is limited to archaic meanings (e.g., krasen from krasnyj with the archaic meaning `beautiful', not the modern meaning `red') and archaic/poetic style, as in the example below from Marina Tsvetaeva's poem:

(18) И	плащ	его -	был -	красен,
I	plasc	ego -	byl -	krasen,
and	cloak	his	was	red.SHORT
И	конь	его -	был -	бел.
I	kon'	ego -	byl -	bel.
and	horse	his	was	white.SHORT



3. And his cloak was red and his horse was white

Also, short forms of color adjectives may occur in the attributive position in idiomatic expressions, but these expressions are frozen relics of an earlier stages of Russian when short forms were not associated with Stage-level properties:

(19) a. krasnadevica

red/beautiful.SHORTmaiden

`a beautiful maiden'

b. sred'beladnja

amidstwhite.SHORTday

`in broad daylight'

Thus, we classify adjectives of color as belonging to the same category as those of material and origin, the category of Individual-level properties. These adjectives denote inherent characterizing properties that are Individuallevel by default, unless coercion is involved. Therefore, they appear especially close to the NP and below the StageP projection.

Our division of noun-modifying adjectives into three large groups instead of a more fine-grained hierarchy is supported by the data concerning the strength of judgments regarding adjective orderings, I as expressed by inter-speaker homogeneity of judgments (i.e., how little variation across speakers is found with respect to a given adjective pair). Having reanalyzed the data presented in Pereltsvaig (2007), we find that judgments regarding different adjective pairs vary in homogeneity. Some adjective pairs, such as in the phrase staryj belyj taburet `an old white stool', present no problem to the speakers who select the same adjective ordering unanimously. In contrast, judgments for other pairs, such as uzkij suxoj ovrag `a narrow dry ravine', are much more variable, with only 68% of speakers selecting the preferred order. Overall, pairs involving adjectives that would be merged in a3 and a5, according to our proposal in (1), were judged most unanimously, with the average figure of judgment homogeneity being 95.6%. For pairs that involve adjectives from a3 and a4, or from a4 and a5, this figure is 83.3%. For pairs that involve adjectives that belong to the same group, according to our proposal, that is adjectives that are both merged in a3 or in a5, the figure is merely 74% (there are no adjective pairs in Pereltsvaig's 2007 data that involve two a4-type adjectives). The unpublished follow-up study conducted by Pereltsvaig in 2008, enhanced by pictures (whose goal was to make sure that speakers interpret the adjectives in the most uniform fashion), came up with similar figures. The combined results from Pereltsvaig (2007) and the follow-up study (with 66 speakers total) are as follows: 91% for a3-a5 adjective pairs, 79% for a3-a4 and a4-a5 adjective pairs, and 76% for a3-a3 and a5-a5 adjective pairs.

However, there are several problems with these two studies. First, some adjectives could be interpreted differently because of inherent lexical ambiguities; for example, zdorovyj (in zdorovyj ryzij kot `a huge red cat') could be interpreted alternatively as `huge' or `healthy'. The classification of certain other adjectives adopted by Pereltsvaig (2007) is questionable as well: for instance, she classifies blestjascij `shiny' as Color and drevnij `ancient' as Age (rather than Typing Attribute, the lowest category on her hierarchy). Many of these problematic examples are due to the fact that Pereltsvaig (2007) implements very strict measures to control for adjective frequencies. In particular, she selected adjectives with closely matching frequency: no more than 30 positions apart on [Sharoff, 2002] frequency ranking of 5,000 most frequent words, and no adjectives ranked in between the two adjectives selected as a pair to be tested. While controlling for frequency is necessary (e.g., [Scott, 2002] shows that adjective frequency in a corpus can affect its ordering), we felt that the measures implemented by Pereltsvaig (2007) were too strict and resulted in too many unnatural, ambiguous or difficult to process pairs.

In order to remedy these problems, we conducted a new follow-up study, using a different set of lexical materials. In particular, we implemented a different set of measures to control for the frequency of adjectives by using m adjectives that are the most frequent in each meaning subcategory (size, length, etc.), according to Sharoff (2002) frequency dictionary of Russian. These adjectives are listed below:

(20) a. aP3:

malen'kij `small', dlinnyj `long', vysokij `tall, high',

glubokij `deep', sirokij `wide', bystryj `fast', polyj `hollow'

b. aP4:

legkij `light', mokryj `wet', staryj `old' and teplyj `warm'

c. aP5:

cernyj `black', russkij `Russian' and zeleznyj `ferrous' Zolotoj `golden' is more frequent than zeleznyj `ferreous', but it is ambiguous between Color and Material interpretations.

Six types of pairs have been tested: a3-a4, a3-a5, a4-a5, a3-a3, a4-a4 and a5-a5, with two examples of each type of pair (thus, the total 12 items were tested). The actual test items (in the expected order) are listed in (21) below. The types of pairs and the order in which the two alternative orders - the expected and the reverse orders - were presented have been randomized. Each adjective has been used in no more than three test items; all test items involved masculine gender.

(21) a. vysokij staryj dom

b. cernyj zeleznyj sar

c. teplyj russkij platok

d. legkij mokryj plasc

e. dlinnyj cernyj xvost

f. staryj cernyj avtomobil'

g. glubokij sirokij rov

h. staryj teplyj sviter

i. dlinnyj teplyj sarf

j. malen'kij zeleznyj kljucik

k. zeleznyj russkij krest

l. dlinnyj vysokij zabor

`a tall old house'a3-a4

`a black ferrous [i.e., iron] ball' a5-a5 `a warm Russian kerchief'a4-a5

`a light wet cloak'a4-a4

`a long black tail'a3-a5

`an old black car'a4-a5

`a deep wide trench'a3-a3

`an old warm sweater'a4-a4

`a long warm scarf'a3-a4

`a small ferrous [i.e., iron] key' a3-a5 `a ferrous [i.e., iron] Russian cross' a5-a5 `a long tall fence'a3-a3

Overall, 70 speakers participated in the study, giving us a total of 840 responses. These responses were coded according to whether they were in agreement with the expected order. For each tested pair, we calculated the percentages of speakers who preferred the expected order. Then, average percentage for each pair type, as well as for inter-type pairs (i.e., a3-a4, a3-a5 and a4-a5) and intra-type pairs (i.e., a3-a3, a4-a4 and a5-a5) was calculated. The results are as follows: for the inter-type pairs the average percentage of expected orders is 78.8%, whereas for intra-type pairs the corresponding figure is 52.2%. In other words, speakers were much more in agreement regarding the pairs that involve adjectives from different types (i.e., according to our analysis, adjectives attaching at different levels of the structure) than regarding the pairs that involve adjectives of the same type. This is one measure of how strong and homogeneous the intuitions are.

Individual responses and speaker comments are also interesting. One thing to point out is the much higher frequency of “no preference” responses for intra-type pairs than for inter-type pairs. Such responses were particularly frequent for pairs (21g), (21h), (21k) and (21l). The pair in (21g) received 16 (23%) “no preference” responses (plus five additional speakers commented that their “preference is very slight”), and the pairs in (21h) and (21k) received 6 and 5 “no preference” responses, respectively. The pair in (21l) received only three “no preference” responses, but three additional speakers commented that their “preference is very slight”. Many speakers also commented that items (21g) and (21l) are better rephrased as coordinated phrases with either order of the adjectives (i.e., glubokij i sirokij rov `deep and wide trench' or sirokij i glubokij rov `wide and deep trench'; dlinnyj i vysokij zabor `long and tall fence' or vysokij i dlinnyj zabor `tall and long fence'). Of the inter-type pairs, only the item in (21i) received a significant number of “no preference” responses (five, to be precise); we are not sure why. Overall, however, it is the intra-type pairs that received the most variable responses and the highest number of “no preference” responses.

Conversely, the pairs with the highest percentage of responses preferring the expected order (i.e., pairs with the most homogeneous responses) were all inter-type pairs: (21e), (21f) and (21j), receiving 81%, 86% and 100%, respectively.

To summarize, our study shows that there is a clear difference between adjective pairs that involve two adjectives of the same type (i.e., intra-type pairs) and adjective pairs that involve adjectives of different types (i.e., intertype pairs). The intuitions about adjective ordering are much stronger and more homogenous across speakers for inter-type pairs than for intra-type pairs. This strongly supports our division of adjectives into these three types.

4. Poslednije-Type Adjectives

In this section, we consider the adjectives of the so-called poslednie-type (cf. [Babby, 1987]), which precede quantifiers such as numerals and fulfill ™ a special semantic function. Unlike the adjectives discussed in the previous section, they do not modify the property contributed by the NP, but ratherm provide information regarding the individuals referred to or quantified over by the DP. In the presence of such adjectives, the nominal expression cannot be interpreted as property-denoting; rather, it receives a referential (or sometimes quantificational) status. It is important to note that we include in this group indefinite pronouns (e.g. kakie-to) and determiners (e.g. eti) that appear in a DP and exhibit adjective-like behavior in that they agree with the head noun in number and gender:

(22) a. posledniepjat'knig

lastfivebooks.GEN

`the last five books'

b. kakie-todesjat' podrostkov

sometenteenagers.GEN

`some (unknown) ten teenagers'

c. etisest'fil'mov

thesesixmovies.GEN

`these six movies'

In (22a), the adjectiveposlednie `last' does not modify the kind of the books in question. Instead, it specifies which particular books the nominal refers to. In other words, its function is not to modify the property but to help identify the referent. It helps us choose, out of those entities that instantiate the property of being a book, the particular individuals referred to by the DP. The item kakie-to `some' in (22b) contributes existential quantification over groups of ten teenagers and further makes sure that the speaker cannot identify the particular teenagers involved. In other words, it marks the referent as not speaker-identifiable (cf. [Kagan, 2011] for a detailed discussion of -to items). Once again, lack of identifiability is a characteristic of the referent, not part of the property denoted by the NP. Finally, the demonstrative eti `these' in (22c) is an indexical expression that provides the nominal with a referential and definite status, making sure that its referent is familiar from the context (either physical or linguistic). Thus, in all these cases, the adjectives that precede the numeral provide information about the intended referent of the nominal, about the individuals that it picks up, and not about the property denoted by the NP.

Crucially, the referential interpretation of the nominal is not a mere by-product of the lexical meaning of the adjective. Rather, the structural position in which the adjective appears plays a crucial role in determining its meaning. Evidence for this claim comes from the fact that many adjectives can appear in different positions: either in the high position above the NumP (aP-1) or lower in the structure below NumP (e.g., in aP-3). In these cases, the interpretation of the adjective and of the nominal expression as a whole depends on the position of the adjective. Adjectives merged high - in aP-1 - modify individuals rather than properties and thus indicate that the nominal as a whole is either referential or quantificational in terms of its semantic type. In contrast, when the same adjectives are merged lower in the structure, they are interpreted as modifying the property denoted by the NP. This sometimes results in interesting shifts in the meaning of the adjective itself. In the following two subsections, we will illustrate this interaction between the syntactic position of the adjective and its interpretation with a number of examples. The contrast between the higher and the lower site will be determined by two factors: the position of the adjective relative to a numeral and its case form. Adjectives appearing in the higher site precede the numeral and exhibit nominative/accusative case (we will call the AP_NOM/ACC-Num order `the A-initial pattern'), whereas adjectives appearing in the lower site follow the numeral and appear in the genitive case, which is associated with the numeral (the Num-AP<sub>GEN</sub> order will be referred to as `the Num-initial pattern'). For the purposes of this discussion, we focus on nominals as they would appear in a structural case (nominative or accusative) position. If the nominal occurs in an oblique case position, prenominal adjectives appear in the relevant oblique case, regardless of their structural position (cf. [Babby, 1987; Bailyn, 2004], and the discussion in Section 5 below). Then, in Section 4.3 we show that when occurring in the lower position, the adjective does not rule out the option of the nominal as a whole receiving a property-type interpretation, but with adjectives occurring in aP-1, the nominal as a whole cannot be interpreted as denoting a property.

5. Referentiality

Appropriately, our first set of examples contains the adjectivepervyj `first'. Consider the following pair of sentences:

(23) a. Pervye	pjat'	ucitelej	vysli	v	final.
first.NOM	five	teachers.GEN	went-out	to	final
`The first five teachers came through to the finals.'
b. Pjat'	pervyx	ucitelej	vysli	v	final.

6. (The) five first teachers came through to the finals

When the adjective pervyj `first' occurs in the higher position, above the numeral, its function is to make the referent of the nominal identifiable; as such, it makes the nominal as a whole referential. For instance, in order to figure out who came through to the finals according to (23a), one has to consider the first five teachers that participated in the contest, or the first five teachers in the row, etc., as should be specified by the context.

Interestingly, a different interpretation emerges if the adjective is merged in a lower position, to the right of the numeral, as in (23b). This sentence, unlike (23a), receives the meaning in which pervyje `first' does not affect the referentiality of the nominal, but rather determines the nature of the kind of teacher involved. This sentence can be uttered, for example, if people nominate for participation in the contest their first teachers (each person nominates the first individual who taught him or her at school). In this case, we deal with a contest in which first teachers participate. The sentence (23b) asserts that five participants came through to the finals. Here, the adjective pervyx `first' modifies the property denoted by the NP, rather than relating to particular instantiations of this property. The individuals instantiate the property `first teacher', and not merely the property `teacher'. Except for being characterized by this property, the individuals need not be first in sequence in any other sense (e.g., first to make it through to the finals, first in a row, etc.). For instance, they could be the last participants on the list. Such a meaning can even be expressed in a phrase like (24a), in which the items pervyx `first' and poslednie `last' are perfectly compatible since the former modifies the property and the latter helps to identify specific instantiations of the property.⁸ Note that (24b) with the A-initial pattern cannot receive the corresponding reading, and in fact is ungrammatical.

Our next example is provided by what we may call specificity markers, that is items that when occupying the high position, mark the nominals as either specific or non-specific. To illustrate, consider the minimal context in (25), discussed by Pereltsvaig (2006a)

(24) a. poslednie	pjat'	pervyx	ucitelej
last.NOM	five	first.GEN	teachers.GEN
`the last five first teachers'
b. *poslednie	pervye	pjat'	ucitelej
last.NOM	first.NOM	five	teachers.GEN

(25) V Mariinskom teatre tancevali...

In the Mariinsky Theatre danced

a. ...opredelennyjepjat'balerin.

certain.NOMfiveballerinas.GEN

`A certain five ballerinas danced in the Mariinsky Theatre.'

...