1. Introduction*

It is widely acknowledged that relational adjectives (RAdjs) (Bally, 1944; Beard, 1995; Fradin, 2008a, 2008b; Rainer, 2013; Fábregas, 2014, among many others), called associative adjectives (Pullum & Huddleston, 2002; Giegerich, 2015) or nominal nonpredicating adjectives (Levi, 1978), are attributive only, as suggested by Levi’s term, and that their independent occurrence in the predicate position requires some explanation. Such noncanonical expressions come in two types. In the first type, the predicative use is superficial, and the predicate consists of a Radj that modifies a formally null noun phrase (NP) or noun. In the second type, the predicative use is genuine, and the predicate is a qualitative adjective that is converted (or coerced) from a Radj. The following sentences that have nuclear in the predicate position exemplify each of these two types, (1a) the superficial use; and (1b) the genuine use (Bauer, Lieber & Plag, 2013, p. 318):

(1)	a.	Newsweek 1997: France is second—75 percent of French electricity is nuclear, which has reduced French air pollution fivefold—followed by Russia and Japan.
	b.	USA Today 2005: The outspoken Texas conservative, who displays the Ten Commandments in his office but admits he has a hard time loving his enemies, declined to run for House speaker in 1998 because he considered himself “too nuclear”.

From the interpretations of these examples, nuclear in (1b) clearly has a qualitative reading and expresses a gradable property, whereas that in (1a) does not. Rather, the latter instance retains the basic classificatory function of Radjs, so the sentence in (1a) means that 75 percent of French electricity is the nuclear type.

Levi (1978: ch. 7) argues that Radjs in the predicate position, such as (1a), and qualitative adjectives derived from relational counterparts, such as (1b), should be separate (see also Babby, 2010). Levi (1978, p. 255) advances a deletion analysis for the first type. In her view, the adjective remains relational but occurs on its own because its modified NP can be deleted under identity with a noun in the subject position of a copular sentence. For example, the copular sentence in (1a) is based on a process depicted below, in which the second occurrence of electricity of the underlying structure is deleted, stranding the Radj in a predicate position.

(2)	75% of French electricity BELONG TO nuclear	electricity
	→ 75% of French electricity is nuclear.	(= 1a)

Based on Levi’s analysis, we refer to the resulting construction as a stranded Radj copular sentence. Her own examples are provided below (Levi, 1978, p. 254).

(3)	a.	The process by which compounds are formed is transformational.
	b.	Her infection turned out to be bacterial, not viral.
	c.	His razor is electric.
	d.	Question formation in Finnish is morphemic.
	e.	The therapy David does is primarily musical.
	f.	That interpretation is presidential, not judicial.

The purpose of this paper is to shed new light on Radj stranding in English from a comparative‑structural perspective. A major question about Levi’s analysis is the licensing condition for the process. Under what conditions can NPs involving a Radj undergo the deletion process, as depicted in (2)? In other words, when can Radjs be used predicatively in the superficial sense? In our view, these questions call for an analysis based on abstract phrase structure to enable cross‑linguistic similarities and differences to be captured as simply as possible. Sections 2 and 3 concern English and Japanese data, respectively, and section 4 offers an analysis along the above lines. Below, we introduce basic facts and concepts to provide a foundation for the upcoming sections.

Levi (1978, pp. 256‑262) makes three observations about the licensing of Radj stranding. First, the combination of a Radj and the modified noun should be established as instances of what she calls complex nominals. Consider the following data taken from Levi (1978, p. 256):

(4)	a.	{Our engineers/*Those agents/*My relatives} are all chemical.
	b.	{Those agents/*Our engineers/*My relatives} are all theatrical.

The sentence in (4a) shows that Radj stranding is possible with our engineers and chemical, but it is not possible with those agents or my relatives and chemical. The same observation applies to (4b). (Note that qualitative readings of chemical and theatrical are irrelevant here.) According to Levi, the acceptability differs because chemical engineers and theatrical agents are “good” complex nominals, whereas chemical agents, chemical relatives, theatrical engineers, and theatrical relatives are not. If the combination is not established as a complex nominal, it is impossible for it to undergo the deletion process.

Second, Levi (1978, p. 258) observes that stranded adjectives are unacceptable in a relative clause with an indefinite subject:

(5)	a.	I wish I had some {musical talent/*talent which was musical}.
	b.	Rita wants to edit a {linguistic journal/*journal which is linguistic}.

Third, stranded adjectives “[…] are consistently and markedly more acceptable when used in an explicit or implicit comparison than when they are used alone” (Levi, 1978, p. 260). Thus, Levi (1978, p. 260) observes that the (a) sentences in (6) and (7) are more acceptable than the (b) sentences, which lack a contrasting adjective:

(6)	Her infection turns out to be
	a.	viral, not bacterial.
	b.	viral.
(7)	Our firm’s engineers are all
	a.	mechanical, not chemical.
	b.	mechanical.

Additionally, stranded Radjs become more acceptable when they are preceded by adverbs such as primarily, mainly, mostly, or largely (Levi, 1978, p. 260):

(8)	a.	The therapy he does is {primarily musical/?musical}.
	b.	The novelists we studied were {mostly regional/?regional}.
	c.	The equipment they sell is {mainly culinary/?culinary}.

In the acceptable cases, the adverb establishes an implied contrast between the class named by the adjective and other alternatives not overtly expressed. Below, we call these types of adverbs modifiers of extension, which indicate “whether the relation expressed by the adjective is the only one that the modified noun establishes with an entity, or whether there are other relations that are pertinent in that context” (Fábregas, 2014, p. 284).

Returning to (1a), these conditions are also met in this stranding sentence. First, nuclear electricity is established as a complex nominal. In addition, the subject is quantified by a modifier of extension (75%), so there is a clear sense of contrast between nuclear electricity and other types of electricity that should account for the remaining 25%. Levi’s observations are accurate, but how can they be explained? In a recent attempt, Nagano (2018a) argued that Levi’s conditions can be explained by viewing stranded Radj sentences as answers to implicitly or explicitly posed D(iscourse)‑linked wh‑questions for classification, such as Which type of infection was her infection? And Which type of razor does he avour? Each of these questions can be answered by uttering, “It is viral” and “His razor is electric”, for example. This analysis captures the involvement of contrast between subtypes as a natural consequence of the pragmatic context of the utterance.

In our view, however, the issue under discussion requires a structure‑based approach because Levi’s and Nagano’s points are not a prerequisite, at least directly, for the licensing of a comparative construction in Japanese. Anticipating the discussion in section 3, the acceptability of a Japanese sentence that corresponds to the English sentence in (1a), shown as (9a) below, depends on the presence of a classifier such as ‑gata ‘type’ or yurai ‘derived from’.

(9)	a.	Furansu	no	denki	wa	genshiryoku‑{gata/yurai}	da.
		France	gen	electricity	top	nuclear power‑classifier	cop.nonpst
		‘French electricity is of the nuclear type.’
	b.	??Furansu	no	denki	wa	genshiryoku	da.
		France	gen	electricity	top	nuclear power	cop.nonpst
		(Intended) ‘French electricity is of the nuclear type.’

The two copular sentences in (9a) and (9b) differ only in the morphological complexity of the predicate. Here, not modifiers of extension but classifiers are determinative. In short, English and Japanese differ in the licensing of a similar copular sentence; English imposes semantico‑syntactic conditions, while Japanese imposes a morphological condition. Optimally, an analysis of the licensing of RAdj stranding should be able to solve this cross‑linguistic puzzle.

In section 4, the puzzle is addressed by adopting Adger’s (2013) structural analysis of relational nominals, such as side of the table and photo of Lily. Contrary to the traditional view, Adger claims that it is not correct to state that side takes the table as its argument. Rather, the two are related to each other by a relation‑denoting functional head ק . ¹ק is responsible for the argument structure that encodes relationality, such as part, possession, and representation. In the present case, part, a subtype of ק, takes the table as its argument while it receives concrete content from side. The following representations are intended to provide a basic idea of this entirely new analysis of relational nominals:

 

The bubbles represent the process of predicate modification. Similar semantic embellishment processes have been widely recognized in the literature under the name of “further specification” (Levin & Rappaport‑Hovav, 1995; Baker, 2003, p. 222). The point is that traditional relational nouns per se do not take internal arguments. An independent argument‑taking relational head is involved, and side in side of the table and photo in photo of Lily simply further specify it.

Adger’s DP theory provides an approach to Radj stranding in much more general terms than Levi (1978) or Nagano (2018a) used, namely deletion within DP. A careful examination reveals that complex nominal inputs to the deletion process in (2) are the names of (sub)kinds that are structurally similar to relational noun phrases; for instance, nuclear electricity can be read as nuclear kind of electricity. In this analysis, nuclear is the specifier of the phonetically null head kind, a type of ק, while electricity is its complement. As constituents of a kind name, nuclear and electricity are in the same structural relationship as is assumed between edge and the table in Adger’s theory. Further, nuclear is stranded when the complement is deleted within this structure. The same situation applies to Japanese kind‑name counterparts. The two languages differ in how they recover the deleted complement. Crucially, Japanese classifiers, such as ‑gata and yurai in (9a), are morphological manifestations of the ק head, so their presence alone can identify the deleted material. In contrast, English does not have overt ק elements, so the recoverability depends on the pre‑establishment of a kind name itself and the presence of contrast‑making phrases, as shown in (5‑7). In short, we claim that what we call stranded RAdjs are stranded specifiers of ק‑headed DPs, and their occurrence is licensed by the recoverability of the deleted complement.

2. Data from English

In section 1, we introduced Levi’s observations about RAdj stranding in English. This section aims to refine the conditions by adding another observation concerning the types of RAdj + noun combinations.

2.1 Relational adjectives in complex nominals

RAdjs have a multifaceted nature both functionally and formally (Beard, 1991, 1995; Fradin, 2008b; Marchis Moreno, 2018; Nagano, 2018a, 2018b), and it is not that any RAdj + noun combination presents an appropriate input to the stranding process. This issue is independent of the semantico‑syntactic conditions introduced in section 1; some combinations are inherently excluded from the stranding process. To pin down the appropriate input type, we start with Levi’s (1978) classification. In Levi’s definition, the term complex nominal refers to a nominal construction in which a head noun is preceded by a modifying element that is either a noun or a RAdj. She divides English complex nominals into the following three sets (the labels are also Levi’s (1978, (1‑2), bold added)):

(10)	nominal compounds
	apple cake	deficiency disease
	time bomb	autumn rains
	doghouse	nicotine fit
	windmill	color television
	daisy chain	surface tension
(11)	nominalizations
	Markovian solution	film producer
	American attack	city planner
	presidential refusal	dream analysis
	musical criticism	quantifier lowering
	constitutional amendment	metal detection
(12)	NPs with non‑predicating adjectives
	electric clock	musical clock
	electric shock	musical criticism
	electrical engineering	musical interlude
	electrical conductor	musical comedy
	electrical outlet	musical talent

Exactly half of these complex nominals contain RAdjs (bolded). In Levi’s transformational analysis and as a widely accepted view, RAdjs are nouns in adjectival form2. Our concern here is how to correctly characterize the “NPs with non‑predicating adjectives” group. As we saw in section 1, Levi believes that instances of this group can undergo the deletion under identity and yield stranded RAdjs.

First, Nagano (2018a, 2018b) shows that RAdjs in the “nominalization” group cannot be stranded because they function as complements of their head deverbal nominalizations. For example, American attack is ambiguous between the Subject (America) + Verb (attack) interpretation and the Verb (attack) + Object (America) interpretation. Most likely, the Malkovian solution is based on the Subject + Verb relationship, while the remaining three complex nominals are based on the Verb + Object relationship3. The view that RAdjs function as complements (rather than modifiers) can be confirmed by a one‑substitution test. It is well known that in a NP, the pro‑form one can substitute for the head noun excluding its modifier, but the exclusion of its complement leads to ungrammaticality, as in (Harley, 2009, p. 134):

(13)	a.	?*That student of chemistry and this one of physics sit together.
	b.	That student with short hair and this one with long hair sit together.

Based on this contrast, consider the following data taken from Giegerich (2015, p. 36)4.

(14)	a.	?Do you mean the presidential murder or the papal one?
		?Do you mean the parliamentary election or the presidential one?
		?Is this a subject review or an institutional one?
	b.	Is this the bovine strain of the disease or the feline one?
		Do you have a medical appointment or a dental one?
		Is he a rural policeman or an urban one?
		Is this a cold‑water fish or a tropical one?

The anaphoric substitution in (14a) is less acceptable than that in (14b) because the RAdjs in (14a) function as complements to the head deverbal nouns. Following Payne & Huddleston (2002, pp. 439‑451), we call the RAdjs in (11) pre‑head complements.

Moving on to the “NPs with non‑predicating adjectives” group, a difficulty with this group is that being a pre‑head modifier does not indicate that the adjective can be stranded. A discussion by McNally & Boleda (2004) strongly suggests that stranding is most compatible with pre‑head modifier RAdjs within complex nominals that function as common names. We delve into this subclass in the next section. In our view, complex nominals as common names undergo the RAdj stranding process because they have a ק‑headed structure like relational nominals.

Finally, a remark about the “nominal compounds” group is necessary because RAdjs in English can be zero‑derived. A complication arises when the member’s compound status is uncertain in terms of established word/phrase criteria, such as stress, semantics, separability and coordination (Liberman & Sproat, 1992; Bauer, 1998, 2017; Klinge, 2009; Shimamura, 2014; Giegerich, 2015). There are noun + noun combinations that should be seen as phrases, and such instances are very similar to RAdj + noun combinations except that the first element lacks an adjectival suffix. Given that most relational adjectivalizing suffixes in English are Latinate (Bauer et al., 2013, p. 314), there is a very slim chance that non‑Latinate nouns (e.g. cashmere in cashmere scarf, origami in origami birds), acronyms (e.g. PC in PC desk), and nouns of greater length (e.g. social security in [[social security] number]) will be affixed with one of these suffixes. Below, we focus on denominal adjectives with overt suffixes, but it is important to distinguish nominal compounds from noun + noun combinations which could be classified into the same group as (12) from a functional point of view. Unlike nominal compounds, the latter type of combination can undergo the same deletion process as RAdj + noun combinations and yield similar copular sentences. For example, cashmere scarf yields the following instance (Shimada, 2017, cited from Oxford Advanced Learner’s Dictionary):

(15)	The scarf is 70% cashmere.

2.2 Relational adjectives in complex common names

Advancing the view that RAdjs are intersective just like prototypical adjectives, McNally & Boleda (2004) study Catalan and English counterexamples to Bally’s (1944) generalization that RAdjs are attributive only. They observe that RAdjs occur in the predicate position when “the property denoted by the adjective is used to classify individual instances of a kind that could typically be described using the adjective” (McNally & Boleda, 2004, p. 191). Witness their English example (ibid.):

(16)

Infection with tuberculosis spreads in two ways, by the respiratory route directly from another infected person or by the gastrointestinal route by drinking milk infected with the tubercle bacillus… In infections with M. tuberculosis, the tubercle bacilli commonly affect the lungs, in which case the disease is known as pulmonary tuberculosis. By contrast, infections with M. bovis often affect the bones and joints. About 90 percent of all clinically recognized tuberculosis in humans is pulmonary. (Taken from the Britannica Guide to the Nobel Prizes)

The long citation is necessary to show that “strandable” adjectives are used to name a kind (pulmonary tuberculosis), and “stranded” sentences are produced when individual instances (about 90 percent of all clinically recognized tuberculosis) are classified based on this kind name. (We add the double quotations because McNally & Boleda do not share Levi’s deletion analysis.) The authors also compare international conference with international bakery and indicate, based on a Google search, that the former complex nominal can engender corresponding copular sentences (e.g. This conference is international), while the latter scarcely can. The difference lies in the plausibility of the two complex nominals as the names of kinds.

As a consequence of the above characterization, McNally & Boleda (2003, p. 192) correctly note that “the use of relational adjectives predicatively to subclassify ordinary individuals is most frequent in specialized discourses, where the adjectives used for subclassification of a given kind of entity are well known and the interest in such subclassification is obvious”. Thus, the linguistics literature is fertile ground for stranded RAdjs. Below, we cite a small portion of the vast instances we have collected from our academic field:

(17)	Review of Adger (2013) by Donati (2014) in Journal of Linguistics, 50. (p. 495)
	In the introduction to this book, its author, David Adger, explains that it arose from two separate papers in progress. This is still visible in the structure of the book, which is made up of two different parts. The first part (ch. 2‑3) is completely theoretical, and concerns central issues such as labels, lexicon, and cartography. It is followed by an analytical part (ch. 4‑6), which is entirely empirically oriented and devoted to an in‑depth analysis of the syntax of DP constructions in a number of languages, notably (Scottish) Gaelic.
(18)	Review of Adger (2013) by Cecchetto (2015) in Language, 91(2). (p. 484)
	I have mixed feelings about the title. It is a reference to a quotation from Aristotle and alludes to the claim that no noun is really relational, relationality being imposed on the noun from the structure it is inserted into. Therefore, all nouns would denote undifferentiated substance, not relations. The title is certainly eye‑catching. It might incorrectly suggest, however, that the book is about the count/mass noun distinction, a topic that is not covered here.
(19)	Kastovsky (2009) in Diachronic perspectives. The Oxford Handbook of Compounding (p. 332, footnote.16). Oxford: Oxford University Press.
	[…] The copulative type girlfriend, oaktree, on the other hand, sometimes wrongly subsumed under the dvandva class, is determinative, i.e. endocentric, and therefore belongs to the karmadhāraya type.
(20)	Giegerich (2012) in The morphology of ‑ly and the categorial status of ‘adverbs’ in English. English Language and Linguistics, 16(3). (p. 342).
	[…] I will therefore argue, in addition to subscribing to the single‑category claim, that adjectival and adverbial ‑ly are in terms of the morphological system of English radically different suffixes: the former is a derivational suffix while the latter is inflectional. In deadly, the suffix is derivational but non‑category‑changing (like, for example, ‑ish in greenish); but nicely is an inflected form of the adjective nice.
(21)	Traugott (1989) in On the rise of epistemic meanings in English: An example of subjectification in semantic change. Language, 65(1). (p. 34)
	The semantic theory must also allow one to tell when a polysemy has come into being. Geis and Zwicky (1971) discuss the criteria with reference to the development of causal since, which actually was siþþan in OE. Siþþan indicated the temporal relationship ‘after(ward)’, ‘from that time on’, as in Since I left home my mother has been mad at me, but invited the inference of a causal relation. So long as since had to have a temporal interpretation, it is inappropriate to speak of a temporal‑causal polysemy. However, a polysemy arose in ME when what was formerly only an inference had to be construed as the actual meaning of the form, as in Since I am leaving home, my mother is mad at me. At that stage since had become polysemous: in one of its meanings it was temporal and could have an invited inference of causality; in the other, it was causal.

The passage in (21) attests to McNally & Boleda’s point in a compact yet very clear way. First are kind names named by RAdjs (causal ‘since’, temporal ‘since’), and then follow classificational copular sentences using the same RAdjs.

Moreover, the examples cited above confirm Levi’s observation that a context of comparison or contrast facilitates RAdj stranding. In (17) and (18), we have completely and no as modifiers of extension. In (19)‑(21), different individual instances and their types are enumerated and compared.

The use of RAdjs in kind names is discussed by Gunkel & Zifonun (2009) under the term complex common name. Common names are “neither singular terms nor general descriptions but establish a distinct type of general term that is used to refer to ‘kinds’” (Gunkel & Zifonun, 2009, p. 205). Common names are names of kinds, morphologically simplex or complex. Complex common names including classifying modifiers are names of subkinds. Crucially, RAdjs are a major type of classifying modifier. Compare the instances of common names in (22a) with those of general descriptions in (22b) (Gunkel & Zifonun, 2009, p. 206):

(22)	a.	tiger, hammer, lawn mower, domestic animal, Indian elephant
	b.	students from abroad, elephants living in Africa

As diagnostics for the name/description division, common names can be used with kind‑selecting predicates such as be extinct and die out (e.g. The Indian elephant will soon die out), and they pass the so‑called test (Carlson, 1977, p. 442) (e.g. Indian elephants are so called because their largest population is found in India). Additionally, common names are natural with modification by so‑called (e.g. so‑called Indian elephant). In Japanese, common names are often used in the NP frame [N to iu mono], lit. ‘N quot say thing’, ‘what is known as N’; e.g. Indozō to iu mono ‘what is known as Indian elephant’.

From a formal point of view, the constituents of complex common names are connected together more tightly than those of a general description, as suggested by comparing (22a) with (22b) (see also Gunkel & Zifonun, 2009, pp. 212‑214). The tightness of the connection is clearly illustrated by the following list of complex common names, which are taken from Gunkel & Zifonun (2009, pp. 208‑209):

(23)	derivations and exocentric compounds
	fre			pommier ‘apple tree’ (< pomme ‘apple’),
				ouvre‑bouteille ‘bottle opener’
(24)	endocentric compounds
	eng			apple tree
	fre			mode‑homme ‘men’s fashion’
	pol			zegarmistrz ‘watchmaker’
(25)	N+N syntagmas
	a		eng	Bush administration
			lat	(scientific) canis lupus ‘wolf’
	b.		eng	women’s magazine, bird’s nest
(26)	N+NP/PP syntagmas
	a.	pol		kierowca samochodu (driver car.gen) ‘car driver’
				colow włosów (color hair.gen.pl) ‘hair color’
	b.	eng		weapons of mass destruction, bird of prey
		fre		chemise de nuit ‘night gown’, homme de la rue ‘man in the street’
		pol		krople do nasa (drop.pl for nose.gen.sg) ‘nose drops’
(27)	A+N syntagmas
	eng			urban transit, cellular division, musical critic
	fre			taches solaires ‘sunspots’, chat domestique ‘domestic cat’
				intervention militaire ‘military intervention’
	pol			hala dworcowa ‘station concourse’, sok jobłkowy ‘apple juice’
				białe wino ‘white wine’

The term “syntagmas” in (26) and (27) refers to phrases, so the formal tightness of complex common names does not indicate that the combination should be a word. Small phrases such as those discussed by Sadler & Arnold (1994) can be common names. At the same time, as Gunkel & Zifonun correctly points out, formal tightness is not a sufficient condition for name status. Although the following combinations are formally similar to those in (22a) and (27), they cannot be seen as common names:

(28)	A+N syntagmas
	Indian government, Sub‑Saharan Africa, trans‑global expedition;
	nocturnal visit, post‑lexical component;
	“Patrick Schwarzenegger, son of the world‑famous Terminator, is posting pictures of his Japanese vacation: posing with a sumo wrestler; visiting the Meiji shrine; eating lots of sushi”. (Taken from a story by CNN)

Generally speaking, relational adjectives expressing Place and Time, including those reported above, tend to be used to form general descriptions5.

In our view, being a classifying modifier of a complex common name is definitive for a RAdj to be stranded. Levi’s classification in (12) does not take into account the name/description distinction, so the term “NPs with non‑predicating adjectives” is not quite accurate in capturing the licensing condition of the phenomenon.

3. Data from Japanese

This section aims to perform a contrastive examination of the English RAdj stranding phenomenon. To the best of our knowledge, previous contrastive studies have been limited to data from European languages and have compared RAdjs per se from different European languages, as in Gunkel & Zifonun (2008) and Marchis Moreno (2018). In contrast, we compare a European language and an Asian language and demonstrate that RAdjs can be fruitfully contrasted with formally different, yet functionally similar, constructions from a distant language family.

3.1 Nominal predicates

Japanese copular sentences connect two noun phrases by the copula da, typically taking the form: NP₁ {wa/ga} NP₂ da (lit. NP₁ {top/nom} NP₂ cop). As in English (cf. Huddleston, 2002, pp. 266‑272), they have ascriptive and specifying uses. Compare the following sentences (the labels and data are both from Kishimoto, 2012, p. 41)6:

(29)	a.	Predicational sentence
		Kare	no	onīsan	wa	(kanari no)	keppeki‑shō	da
		He	gen	brother	top	(a good bit gen)	cleanliness‑classifier	cop.nonpst
		‘His brother is obsessively concerned with cleanliness.’

	b.	Specificational sentence
		Konkai	no	kaji	no	genin	wa	tabako	no	fushimatsu	da
		the latest	gen	fire	gen	cause	top	cigarette	gen	mishandling	cop.nonpst
		‘The cause of the latest fire is a cigarette that wasn’t put out properly.’

In the sentence in (29a), the copular verb da, which inflects as datta in the past form, is used ascriptively; the sentence ascribes to the subject the property of being obsessed with cleanliness. In contrast, (29b) illustrates the specifying use of da, which defines a variable and specifies its value. Our discussion is concerned with word‑level predicates used in the ascriptive type, such as keppeki‑shō ‘obsession with cleanliness’. Traditionally, they are called nominal predicates because they are categorially nouns.

Shimada (2004: ch. 5) observes that nominal predicates have a distinct morpho‑syntactic character of endocentricity; such nouns are headed by various classifiers. Wei (2007) submits a similar view for Mandarin Chinese. Thus, keppeki‑shō ‘obsession with cleanliness’ is headed by ‑shō. This bound classifier is used to name types of diseases or personal tendencies, attaching to nouns that are strongly associated with the named disease (30) or tendency (31):

(30)	Diseases
	a.	arukōru‑izon‑shō
		alcohol‑dependency‑classifier
		‘alcoholism’
	b.	kafun‑shō
		pollen‑classifier
		‘pollinosis, pollen allergy’
	c.	kenbō‑shō
		forgetfulness‑classifier
		‘forgetfulness; amnesia’
(31)	Tendencies
	a.	aki‑shō
		boredness‑classifier
		‘a tendency to become bored easily’
	b.	are‑shō
		dryness‑classifier
		‘a predisposition to having dry skin’
	c.	kori‑shō
		enthusiasm‑classifier
		‘a tendency to become totally immersed in something’

Each of these ‑shō words can replace keppeki‑shō in (29a) and does not affect the naturalness of the construction at all.

Endocentricity can be observed with other instances of nominal predicates, as discussed in Shimada (2004, pp. 153‑157). Anticipating the upcoming discussion, both (a) modificational and (b) predicational forms are illustrated for each nominal predicate. The former further divides into two types based on the element that connects two NPs: no (the genitive marker) or dearu (an attributive form of the copula da ‘be’). The parentheses indicate the meaning of the classifier; if it is preceded by a hyphen, it is morphologically bound.

(32)	a.	haregi‑sugata	{no/dearu}	wakamono	(‑sugata ‘dress, guise’)
		gala dress‑classifier	{gen/cop}	young man
		‘a young man dressed in his best clothes’
	b.	Wakamono	wa	haregi‑sugata	da
		Young man	top	gala dress‑classifier	cop.nonpst
		‘The young man is dressed in his best clothes.’

(33)	a.	nihon‑sei	{no/dearu}	Tokei	(‑sei ‘made’)
		Japan‑classifier	{gen/cop}	watch
		‘a Japanese watch’
	b.	Sono	tokei	wa	nihon‑sei	da
		That	watch	top	Japan‑classifier	cop.nonpst
		‘That watch was made in Japan.’

(34)	a.	dekaruto‑ha	{no/dearu}	gakusha	(‑ha ‘sect, faction, school’)
		Descartes‑classifier	{gen/cop}	scholar
		‘a Cartesian scholar’
	b.	Sono	gakusha	wa	dekaruto‑ha	datta
		That	scholar	top	Descartes classifier	cop.pst
		‘That scholar was Cartesian.’

(35)	a.	hakuai‑shugi	{no/dearu}	shōbōshi‑tachi	(shugi ‘‑ism, belief’)
		altruism‑classifier	{gen/cop}	fireman‑pl
		‘altruistic firemen’
	b.	Shōbōshi‑tachi	wa	hakuai‑shugi	da
		fireman‑pl	top	altruism‑classifier	cop.nonpst
		‘Firemen are altruistic.’

(36)	a.	gakusei	{no/dearu}	shachō
		Student	{gen/cop}	president
		‘A president who is a student’
	b.	Wagasha	no	shachō	wa	gakusei	desu
		Our company	gen	president	top	student	cop.nonpst.polite
		‘The president of our company is a student.’

The bold‑faced expressions are all headed by a classifier, except the one in (36). Semantically, all of them, including the one in (36), refer to kinds. Given the discussion in section 2, it is safe to conclude that they are common names. Their category (N), tight‑knit form, and class‑denoting semantics, as well as the be extinct and so‑called tests, attest to this conclusion. We have no account of the exceptional morphological character of the example in (36), but it might be interesting to note that kyōshi ‘teacher’, another common name denoting a profession or social status, is also non‑headed yet predicative (Baker, 2003, p. 163, footnote 40 for profession names).

Let us consider the examples in (32‑36) (a). When a nominal predicate constitutes a nominal modification structure, it is linked to the head noun either by no, the genitive marker, or by dearu, a form of the copula. This relationship is also true of the example in (29a): keppeki‑shō {no/dearu} onīsan (lit. cleanliness‑classifier {gen/cop} brother) ‘my brother who is germophobic’. While a common view regards the genitive/copula distinction as a mere morphological alternation of the same construction, the discussion in section 2.2 casts doubt on such a view. For example, take the so‑called test. The no‑linked forms are very natural with iwayuru ‘so‑called’, but the dearu‑linked forms are not; e.g. (33a) > iwayuru nihonsei no tokei (lit. so‑called Japan‑classifier gen watch) vs. *iwayuru nihonsei dearu tokei (lit. so‑called Japan‑classifier cop watch). The compatibility with a be extinct type predicate also reveals the same contrast, in which only the no‑linked form is acceptable without any difficulty; e.g. (29a) > keppeki‑shō no dansei wa moo mirarenai (lit. cleanliness‑classifier gen men top now cannot_be_seen) ‘Germophobic men are extinct’. The two forms are also morpho‑syntactically distinct. Obviously, the dearu‑form is larger than the no‑form because dearu is a tensed verb form, while no is not. Thus, only the former can inflect, as in N dearu N (non‑past form) ～ N deatta N (past form). Given these clear differences, it is safe to conclude that the no‑linked N‑N combination is a common name, while the dearu‑linked N‑N combination is a general description. Since nominal predicates are common names themselves, as discussed above, the larger combinations at issue constitute names of subkinds when linked by no7.

There is another construction that should be distinguished from the no‑linked N‑N combination. In a noun modificational construction, when a predicate is not classifying, it is linked to the head noun not by no but by na, as in the following:

(37)	a.	totemo	{yūfuku/ritchi} na	shachō
		very	rich	president
		‘a very rich president’

	b.	Wagasha	no	shachō	wa	totemo	{yūfuku/ritchi}	desu
		Our company	gen	president	top	very	rich	cop.nonpst.polite
		‘Our president is very rich.’

Shimada’s analysis neatly distinguishes these examples from the types in (29a) and (32‑36) because yūfuku ‘rich’ (a borrowing from Chinese) and ritchi ‘rich’ (a borrowing from English) are not endocentric words. Semantically, what they denote are not kinds but gradable qualities. Categorially, the predicates in (37) are adjectives, so they are nominalized by the productive A‑to‑N nominalizing suffix ‑sa (Shimamura, 1990), as in (38a) below. In contrast because nominal predicates are nouns, they cannot be nominalized, as pointed out by Nagano (2016, p. 58) with the example in (38b).

(38)	a.	shachō	no	{yūfuku‑sa/ritchi‑sa}
		president	gen	rich‑ness
		‘the richness of the president’

	b.	*kono	kabin	no	chūgoku‑sei‑sa
		this	vase	gen	China‑made‑ness
		‘(Intended) the fact that the vase is made in China’

In summary, this section has revealed that nominal predicates are common names that denote kinds, and noun modificational constructions in which they are linked to the modified noun by no are the names of subkinds8. Nominal predicate + no + noun combinations and nominal predicate + dearu + noun combinations should be distinguished with respect to the name/description status.

3.2 Nominal predicates as equivalents of stranded adjectives

Most readers have probably noticed the striking convergence between Japanese nominal predicates and English RAdjs; for example, compare the Japanese sentence and its English translation in (29a), (33), (34), and (35). Certainly, Japanese nominal predicates do not always correspond to English RAdjs, as suggested by (32) and (36). However, crucially, Nagano (2016) shows that English RAdjs, particularly those occurring in stranded RAdj copular sentences, almost always correspond to Japanese nominal predicates9.

Nagano (2016) observes that RAdjs that are not pre‑head complements are translated into Japanese in two ways, using either a simplex form or a complex form10. Let us consider the following italicized common names that express relations of Material, Shape, Style, Type, Genealogy, and Possession, respectively. N1 stands for the base noun of the adjective and N2 the modified noun. For each English expression, there is a shorter translation (a) and a longer one (b) (Nagano, 2016, pp. 52‑55):

(39)	N2 made of N1
		wheaten bread
	a.	komugi	no	pan
		wheat	gen	bread
	b.	komugi‑sei	no	pan	(‑sei ‘made’)
		wheat‑classifier	gen	bread

(40)	N2 in the shape of N1
		triangular room
	a.	sankaku	no	heya
		triangle	gen	room
	b.	sankaku‑kei	no	heya	(‑kei ‘form’)
		triangle‑classifier	gen	room

(41)	N2 in the style or pattern of N1
		striped jacket
	a.	shima	no	uwagi
		stripe	gen	jacket
	b.	shima‑gara	no	uwagi	(gara ‘pattern’)
		stripe‑classifier	gen	jacket

(42)	N2 of the type of N1
		European economy
	a.	yōroppa	no	keizai
		Europe	gen	economy
	b.	yōroppa‑{gata/ryū}	no	keizai	(‑gata, ‑ryū ‘type, kind’)
		Europe‑classifier	gen	economy

(43)	N2 in the line or family of N1
		Slavic language
	a.	surabu	no	gengo
		Slav	gen	language
	b.	surabu‑kei	no	gengo	(‑kei ‘line, belonging to’)
		Slav‑classifier	gen	language

(44)	N2 with N111
		four‑legged creature
	a.	yotsuashi	no	ikimono
		four‑leg	gen	creature
		tripodded camera
	b.	sankyaku‑tsuki	no	kamera	(‑tsuki ‘with’)
		three‑leg classifier	gen	camera

Nagano (2016) considered the semantic differences between the (a) and (b) forms, but we are most interested in the longer modifiers in (b) that can be used in ascriptive copular sentences of the type identified in (29a). This usage occurs naturally because the forms in (b) but not those in (a), are internally headed and hence are nominal predicates. Thus, compare the following sentences corresponding to the two forms in (39):

(45)	a.	??Kono	pan	wa	komugi	da.
		This	bread	top	wheat	cop.nonpast
		(Intended) ‘This bread is made of wheat.’
	b.	Kono	pan	wa	komugi‑sei	da.
		This	bread	top	wheat‑classifier	cop.nonpast
		‘This bread is made of wheat.’

The sentence in (45a) is acceptable as a specifying sentence but not as an ascriptive sentence. In contrast, (45b) shows that the presence of a classifier drastically changes the acceptability as an ascriptive. The same observation applies to the cases in (40‑44).

Moreover, Nagano (2016, pp. 58‑60) observes that the genitive marker of the longer modifiers in (39‑44) (b) can be preserved in the predicate position. Compare the sentence in (45b) with the following one (cited from Nagano, 2016, p. 59 with a slight modification):

(46)	Kono	pan	wa	komugi-sei-no	da.
	This	bread	top	wheat-classifer-gen	cop.nonpast
	‘This bread is made of wheat.’

This fact is important for our deletion analysis. If (45b) and (46) involve a deletion within the DP similar to that depicted in (2), it is expected that the genitive marker can be preserved.

In brief, Japanese nominal predicates are the equivalents of English stranded RAdjs.

Returning to the research question of this paper, we have revealed that, both in English and in Japanese, complex common names produce the construction in which we are interested. However, the two languages differ strikingly in certain ways. Specifically, a general analysis of RAdj stranding should address the following cross‑linguistic puzzles:

Cross‑linguistic puzzle 1:

Classifying modifiers in complex common names are adjectives in English but nouns in Japanese.

Cross‑linguistic puzzle 2:

Copular sentences are greatly facilitated by the semantico‑syntactic context of contrast in English, but there is no such condition for Japanese counterparts.

First, the two languages are clearly different in the category of the classifying modifier. Second, Japanese is free from Levi’s third licensing condition, as is evident if we compare the data in (6), (7), and (16‑21) with those in (29a), (32‑36) (b), (45b), and (46). The following section will address these puzzles using Adger’s (2013) theory.

4. Analysis

4.1 A new syntax of relational nominals

One of the essential themes of Adger (2013) is the traditional recognition in generative linguistics regarding the structural parallelism between verbs and nouns (Chomsky, 1970). It is widely believed that relational nominals, such as edge and friend, take complements as transitive verbs do. Complex event nominals (Grimshaw, 1990) are also discussed in the same context. However, Adger challenges this long‑accepted tenet and denies that nouns can take complements as verbs do. In his view, a noun and its apparent complement are not in a sisterhood configuration, but rather, their relationship is negotiated by a third element, which is a real head of the DP structure building. Thus, he derives edge of the table as follows (Adger, 2013, p. 78) using the relation‑denoting functional head ק (see section 1):

(47)

Compare this structure with the usual one in which edge and of the table are in a strictly local configuration [N PP]. In the new structure, the PP phrase is some distance from the relational nominal and is rather selected by ק, a category that is dedicated to the function‑argument structure that encodes relationality. This head is rooted by a light root, such as part, poss(ession), or rep(resentation), which does not enter into pronunciation; that is, part, poss, rep and others are purely abstract functions. The noncapitalized roots are called heavy roots, and they correspond to lexical items in the traditional term. In this analysis, edge is a specifier and a predicate modifier of the relational head, as in: λx.edge (x)∧part (x, the‑table). This specifier semantically embellishes the content of the light root.

Adger’s analysis might strike one as unnecessarily complicating the structure of relational nominals. However, there is solid empirical motivation for the change. Transitive verbs and relational nominals are crucially different in the optionality of their complement. Consider the following examples (Adger, 2016, pp. 57‑63):

(48)	a.	an edge (of the table)
	b.	Lilly killed *(the mouse)

The author investigates relational nominals cross‑linguistically and arrives at the following generalization:

(49)	Across languages, relational nominals systematically take their apparent arguments optionally, in contrast to verbs, which vary idiosyncratically in whether any particular argument is optional. (Adger, 2013, p. 61)

Adger argues that the systematic optionality of the alleged complement observed only in the nominal domain remains unclear under the traditional NP/VP parallelism. His new analysis, however, can capture the relevant optionality as reflecting two different structures hosting √edge. That is, edge in an edge of the table is the specifier of the ק head, as shown in (47), while an edge is based on a separate syntactic derivation that does not involve ק.

4.2 Application to our data

Adger (2013) does not mention the phenomenon we are concerned with in this paper. However, given the widely accepted view that RAdjs are nouns in disguise and the empirical fact that Japanese nominal predicates are nouns, it is not far‑fetched to suggest that his DP theory could be applied to our data. Moreover, let us recall that there are many noun + noun combinations that are equivalent to RAdj + noun combinations (section 2.1 and (25, 26)). Concretely, we argue that some complex common names possess the same structures as relational nominals and that the phenomenon that we have been calling RAdj stranding can be seen as the deletion of the element in the complement position of the ק head. Below, we introduce our idea using the same schema that we used to introduce Adger’s analysis in section 1.

 

As before, the bubbles represent the process of predicate modification; thus, nuclear in nuclear electricity and genshiryoku in genshiryoku‑gata no denki are specifiers of the ק head, which is rooted as kind or type. The classifier ‑gata is a morphological realization of the head, while such an element is missing in the case of English, just as the ק head in (47) remains silent in this language. On the other hand, electricity and denki ‘electricity’ are each selected by the ק head as its complement, producing a constituent that denotes ‘kind of electricity’. A drastic change from the common analyses of these expressions is that in our analysis, nuclear is not a modifier with respect to electricity. Rather, it is a modifier of kind, which then connects it to electricity. This view is entirely plausible considering that a RAdj + noun combination can be rephrased as RAdj + type/kind + of + noun when it functions as a common name (e.g. nuclear electricity = nuclear type of electricity).

With respect to the Japanese case, we depart from the traditional view that genshiryoku‑gata modifies denki, and we consider instead that genshiryoku ‘nuclear power’ modifies ‑gata ‘type’, which takes denki ‘electricity’ as its complement. The view that the classifier is in the ק head is supported not only by its very general semantics but also by its heavy counterpart kata ‘type’ behaving differently. Compare the following two DPs involving ‑gata (50a) and kata (50b), respectively:

(50)	a.	genshiryoku‑gata	no	denki
		nuclear power‑classifier	gen	electricity
		‘nuclear electricity’

	b.	denki	no	Kata
		electricity	gen	type
		‘type of electricity’

The phrase in (50b) is a typical relational nominal phrase, with kata ‘type’ taking its (apparent) complement denki ‘electricity’. Just like √edge in an edge of the table, √kata occupies the specifier position of the kind head and surfaces as such in the final realization. In contrast, the classifier ‑gata is always augmented by other specifying material, such as genshiryoku ‘nuclear power’, when it takes a complement, as in (50a). Therefore, *gata no denki (lit. classifier gen electricity) or *denki no gata (lit. electricity gen classifier) ‘(intended) type of electricity’ is illicit. To explain the difference between (50a) and (50b), the classifier should be seen as a light root and the free‑standing form as its heavy root counterpart12. In summary, we extend Adger’s analysis of relational nominal to our data in the following manner:

(51)
(52)

We are now in a position to address our research question and the two related puzzles. In our view, RAdj stranding is licensed by the involvement of the ק‑headed structure. When nuclear electricity has the structure of ‘nuclear kind of electricity’, nuclear, the specifier, can be stranded as a consequence of the deletion of the complement. The same applies to the Japanese counterpart genshiryoku‑gata no denki. Under this analysis, stranded adjectives are still attributive only, pace McNally & Boleda (2003), because they modify the head √kind.

The first cross‑linguistic puzzle concerns why nuclear is an adjective while its Japanese counterpart is a noun. In our view, this case is a morphological difference between the two languages concerning the pronunciation of the underlying ק structure. Starting with the Japanese structure in (52), ‑gata is in the head position, so its pronunciation involves the morphological amalgamation of this material and the material in the specifier position. The resulting composite is categorially nominal because it is internally headed by ‑gata, which is categorially nominal. The morphological amalgamation co‑occurs with word‑order inversion. Let us compare (50a) with (50b) again. While no in (50b) comes in front of kata ‘type’, no in (50a) comes after the nominal composite headed by ‑gata. There is some technical discussion regarding the status of the inverted no, but here, it suffices to consider that Japanese uses the genitive marker to link two NPs (Saito, 2012).

Next, let us consider the case of English. It should be noted that the structure in (51) does not lead to the following expression:

(53)	nucleus of electricity

This phrase means ‘nucleus part of electricity’ rather than ‘nuclear kind of electricity’, so it should be seen as headed by the light root √part, rather than √kind. In our view, the heavy root √nucleus takes different forms depending on the subtype of ק. When ק is rooted with √part, its specifier is pronounced in the nominal form (nucleus), but when rooted with √kind, its specifier is pronounced in the relational adjectival form (nuclear). Most likely, English distinguishes a √kind‑rooted structure from a √part‑rooted structure by the final form of the specifier, realizing the former structure by relational adjectival forms or their zero‑marked counterparts (as in, e.g. cashmere scarf, PC desk; see §2.1).

The second cross‑linguistic puzzle is also related to a morphological difference between English and Japanese. One important contribution of this paper is that it has detected the existence of overt realizations of Adger’s ק category: Japanese classifiers. The inventory of morphemes differs from one language to another; English happens to lack relevant morphemes, while Japanese happens to possess them. (Note that the reverse is true in the case of relational adjectival forms: English possesses them, but Japanese does not.) Assuming this fundamental difference, we propose that English RAdj stranding – or the deletion of the complement of the ק head – is greatly facilitated by certain semantico‑syntactic contexts of contrast because ק remains null in this language. In contrast, Japanese nominal predicates exhibit no preference for such contexts because ק is morphologically realized by the classifiers. As we saw in (45), all that is required in Japanese is classifiers. If we suppose that the stranded use of a Japanese nominal predicate is systematic because morphological realization of ק enables the identification of the deleted complement, then it is possible that the stranded use of a RAdj is not systematic due to the absence of such substance. To identify the deleted element, English employs modifiers of extension and other contrast‑evoking strategies to compensate for the absence of a classifier‑like substance. Hence, the second cross‑linguistic puzzle indicates that the notion of subkind is activated differently between the two languages: either morphologically or semantico‑syntactically.

As correctly predicted based on the above analysis, Japanese ascriptive copular sentences do not require a classifier when there is a modifier of extension in the sentence. Shimada (2017) observes that the material‑denoting simplex noun kashimiya ‘cashmere’ can be ascriptive when modified by a modifier of extension. Compare the following three sentences ((54a) from Takahashi, 2015; (54b‑c), from Shimada, 2017):

(54)	a.	??Kono	sētā	wa	kashimiya		da.
		This	sweater	top	cashmere		cop.nonpast
		(Intended) ‘This sweater is made of cashmere.’
	b.	Kono	sētā	wa	kashimiya‑sei			da.
		This	sweater	top	cashmere‑classifier			cop.nonpast
		‘This sweater is made of cashmere.’

c.

Kono

sētā

wa

100%

kashimiya

da.

This

sweater

top

100%

cashmere

cop.nonpast

‘This sweater is 100% cashmere.’

The first two sentences in (54a, b) are parallel to those in (45a, b), showing the function of the classifier ‑sei ‘made’. Interestingly, (54c) further shows that a modifier of extension is functionally similar to the classifier, which is exactly what our analysis predicts13.

Before closing this section, we emphasize that our proposal in (51) concerns complex nominals as complex common names only. As discussed in section 2.1, there are different types of complex nominals involving RAdjs. In particular, RAdjs as pre‑head complements (e.g. presidential murder) should be analyzed differently. Additionally, we believe that a relational adjective can modify a lexical noun directly, yielding structures of direct adjectival modification [ap + np] (Baker, 2003: ch. 4). If our analysis is on the right track, such a structure should underlie complex nominals as general descriptions.

5. Conclusion

In this paper, we have provided a new analysis of the non‑qualitative predicative use of English RAdjs. Through an in‑depth comparison between English and Japanese, we have uncovered a previously unrecognized generalization of the phenomenon: the independent use of a denominal relational expression in an ascriptive copular sentence is licensed by the status of the common name. At the same time, we have shown that the two languages differ in two important respects: the category of the relational expression and the involvement of modifiers of extension. We have argued that Adger’s theory of DP syntax can be naturally extended to our case and that our empirical findings about RAdjs and their Japanese counterparts can be neatly captured by his structural analysis of relational nominals. The core of our proposal can be summarized as follows: (i) the phenomenon should be seen as an instance of the DP complement deletion; and (ii) the differences between English and Japanese basically stem from their morphologies. Our analysis is entirely novel, so some points require further theoretical and empirical investigation. Nonetheless, we believe that we have successfully shown the merits of a comparative‑structural approach to address this issue.