2.1 Case marking

There is fairly general agreement that noun phrases and clauses are distinguished in terms of their Case properties. While the former appears in Case positions, the latter is excluded from those positions. Stowell’s (1981) Case Resistance Principle states that clauses are not assigned Case because they are Case-assigning categories. However, unlike embedded declarative clauses, embedded interrogative clauses (called EQs in this paper) are not necessarily subject to this principle and can occur in Case positions.

The EQ occurs in the embedded subject position in (4a), after the preposition in (4b), in the ECM subject position in (4c), and in the object position of the dative construction in (4d). These are all Case positions from which a canonical clausal category is excluded, as in (3a–d).

It has been noted in the literature that Japanese EQs also have nominal aspects (Fukui 1986; Tomioka 2020 among others). They display distributional patterns parallel with those of their English counterparts.

As illustrated by (5a, c), the EQ in Japanese must carry a case particle when it appears as the subject of a finite clause and an ECM clause. The particle -ni in (5b) is a postposition. Its obligatory occurrence also indicates that the EQ that carries it is nominal and must receive a structural Case. Unlike the other examples in (5), the particle is optional in (5d). There are two possible analyses. One is that the EQ is consistently nominal regardless of the presence or absence of the particle. In this analysis, the EQ without the particle is considered to result from the optional dropping of the particle (see Kuno 1973; Saito 1985; Masunaga 1988; Hoshi 1993 for case particle dropping in the nominal context in general). The other is that while the EQ with a case particle is nominal, the EQ without it is clausal. This paper adopts the latter.

As noted by Fukuda (1993) and Kageyama (1993), unlike the accusative particle, the nominative particle is hard to drop (see (6a, b)). Likewise, the nominative particle on the subject EQ in (7a, b) cannot be dropped.

It follows that the case particle in (5d) is not optional, though it looks so. Case-marked and Caseless EQs in (5d) are not in free variation, but there are two versions of EQ, either nominal or clausal, in this pattern.

This conclusion is further supported by the test using coordination. To begin with, consider the following example.

The EQ selected by mayotta ‘wavered’ occurs with or without the postposition -ni. Despite the appearance, (8) does not involve the optional dropping of the postposition. The postposition is obligatory when the verb mayotta ‘waved’ selects a nominal complement.

If the postposition were truly optional, (9) would be grammatical without it. It follows that the optionality of the postposition in (8) is only apparent. The EQ displays categorial ambiguity: while it is nominal with the postposition, it is clausal without it.

Notice that the postposition becomes obligatory when the verb mayotta ‘wavered’ takes coordinated EQs.

Given the discussion about (8) and (9), the coordinated EQs are nominal. This conclusion is consistent with the fact that the coordination by means of the conjunction to ‘and’ is possible only with nominal categories. Consider the following examples.

The contrast between (11a) and (11b) reinforces the conclusion that the coordinated EQs in (10) are nominal, whereas embedded declaratives can never be.

2.2 Modification patterns

Another major criterion for distinguishing nominal categories from others is modification. EQs in English and those in Japanese behave differently in this respect. English resists the modification patterns typically observed in noun phrases even when they occur in a Case position. Consider the following examples.

Unlike the noun question in (12a), the EQ in (12b) is incompatible with a determiner, an adjective, and a relative clause. By contrast, Japanese nominal EQs are compatible with those elements.

(13) illustrates that the EQ co-occurs with a demonstrative and allows adjectival modification. Japanese has two classes of adjectives. One is an adjective like yayakosi-i ‘difficult-adj’ and the other is like zyuuyoo-na ‘important-adj’. They carry adjectival affixes -i and -na, respectively (see Kageyama 1982; 1993; Miyagawa 1987; Urushibara 1993; Nishiyama 1999 for this distinction from theoretical perspectives). Both can modify the EQ. The EQ is also compatible with the genitive-marked modifier onazimi-no ‘familiar-gen’. In (14), the EQ is modified by a relative clause.¹ These are all characteristics of nominal categories and are not observed with declarative clauses, as in (15a, b).

The contrast between Japanese and English in the compatibility with modification indicates that nominalization in English takes place more restrictedly than in Japanese.

A caveat may be in order regarding the modification by relative clauses. As illustrated below, although English EQs are incompatible with a restrictive relative clause, they can be followed by a non-restrictive relative clause.

It may be argued that the relative clause in (14) is also a non-restrictive clause. If that is the case, Japanese and English EQs will be no different in this respect; they are compatible with a non-restrictive relative clause.

However, Japanese non-restrictive relative clauses behave differently from their English counterparts in one crucial respect. Those in English can be associated not only with nominal but with clausal categories.

By contrast, as discussed above, Japanese relative clauses cannot modify clausal categories (see (15b)). The Japanese equivalent of (17) becomes ungrammatical with the relative clause.

The contrast indicates that Japanese relative clauses are limited to nominal modification, whether they are used restrictively or non-restrictively. It follows then that (14) is still valid as evidence for the nominal nature of Japanese EQs.

3.1 The n as a nominalizer

Which head nominalizes a clause is a matter of controversy. Analyzing clausal nominalization in Ndebele, Pietraszko (2019) observes that a clause in that language is nominalized by directly merging D with CP. A similar D-CP structure is proposed for nominalized clauses in languages like Spanish (Plann 1981), Greek (Roussou 1991), Polish, and Georgian (Borsley & Kornfilt 2000). Consider the following examples.

Each sentence here contains a D-related morpheme, such as a determiner and a demonstrative merged with a clausal category.

In this connection, Alexiadou (2020) argues that higher parts of the clausal spine, such as TP and CP, are nominalized only by D. In contrast, lower parts like VoiceP and vP are nominalized by another nominal functional head n (see also Iordăchioaia 2020 for a similar view). Thus, there is a division of labor between D and n. Alexiadou discusses two types of nominalized infinitives in Spanish to support this distinction.

(20a) is called the verbal infinitive, and (20b) is called the nominal infinitive. Although both involve nominalization of the infinitival construction, they differ in the manner of nominalization. The difference is reflected in modification patterns. (20a) allows adverbial, rather than adjectival, modification. (20b) exhibits the opposite pattern. The occurrence of de ‘of’ in (20b) indicates that the infinitive temer ‘fear’ is nominalized and that the adjective is merged with the already nominalized constituent. On the other hand, the impossibility of adjectival modification in (20a) indicates that the part escribir ella novelas is not a nominal but a clausal projection. It is directly merged with D, whereby it is converted into a nominal category.²

While the analysis based on the D-CP/TP structure seems viable for languages listed in (19), it does not apply to nominal EQs in Japanese. Instead, clausal nominalization is attained by another nominal functional head n, rather than the determiner type head D. Japanese nominal EQs behave differently from Spanish verbal infinitives in that they tolerate adjectival modification.

As discussed in section 2.2, Japanese EQs occur with demonstratives, adjectives, and genitive-marked constituents. These elements are used for modifying nominal categories, which means that they target a constituent that has already been nominalized, as stated in Kornfilt & Whitman’s (2011) Functional Nominalization Thesis.

The adjectival modification in (21) indicates that the EQ in these examples must have a nominal character before the adjective is introduced into the structure.

One may postulate a structure like (23) for (21) and argue that adjectival modification is still possible in this structure because the adjective targets a nominal projection D′.

One problematic aspect of this analysis is that it fails to exclude adjectival modification in the so-called poss-ing gerunds in English.

The poss-ing construction displays verbal internal structure except for the genitive marking on the subject. Its external distribution is the same as that of noun phrases. Since Abney (1987), these properties have been attributed to the structure in which D and VP are directly merged.

The absence of adjectival modification in the poss-ing construction indicates that an adjective cannot be merged with D′.⁴ It follows that (23) is not an appropriate structure of (21) and that a nominal category other than D must be merged with EQ so that the adjectival modification may be possible. Based on this consideration, we propose a structure where the interrogative CP (i.e., EQ) is nominalized by n rather than by D, as schematically illustrated below.

3.2 Subclasses of empty nominals

We assume that n is a nominal categorizer postulated in the framework of Distributed Morphology (Halle & Marantz 1993). In this framework, a lexical noun traditionally labeled N is not atomic but syntactically derived via the merger of an acategorial root and a nominal categorizer n. For example, a noun dog has a syntactically complex structure, where a root √dog is merged with the categorizer n. In our analysis, nominal EQ gets its nominal status by merging CP with the phonologically empty nominal categorizer n.

A corollary of this analysis is that the n in question is semantically vacuous, lacking a semantic denotation of its own. It simply serves to convert a clause into a nominal category. An advantage of this analysis is that it can account for the limited distribution of EQs. As illustrated below, only inquiry-type predicates (Tomioka 2020), namely, question-embedding predicates meaning ‘to ask/question’ can select them.

The verbs tazuneta ‘asked’ and touta ‘questioned’, question-embedding predicates (Uegaki 2019; Tomioka 2020; see also Uegaki & Sudo 2019), can select an interrogative clause but not a non-interrogative clause. Conversely, the non-question-embedding verb semeta ‘blamed’ is incompatible with an interrogative clause. This contrast is predictable if n is semantically vacuous. It selects a question denotation (for example, a set of propositions, following Hamblin 1973) and returns the same value. The resulting nominal phrase remains interrogative, allowing only a question-embedding predicate to select it. The nominal projection above the EQ must be, as it were, transparent for s-selection.

This restriction suggests that the nominalized EQ in (27a) lacks the DP layer. Its absence over the semantically vacuous nP is reflected in the unavailability of a demonstrative with the EQ selected by an inquiry-type predicate.

On the assumption that the demonstrative is located in DP, its exclusion from (28) indicates that the EQ here does not contain DP.⁵ In this connection, it is worth recalling that we noted in section 2.2 that some EQs are compatible with demonstratives.

The predicate hanasiau ‘discuss’ is not an inquiry-type, though it embeds a question. The grammaticality of the demonstrative in (29) suggests that the EQ here has been turned into a different semantic type via the merger of n, and hence, it is now compatible with D. It has referential content that tolerates the modification by the demonstrative. The contrast between (28) and (29), therefore, suggests that while the EQ in (28) is nominalized by the semantically vacuous n, the one in (29) utilizes a semantically active content noun (see e.g., Moulton 2015, 2020 on content nouns). Thus, there are at least two major classes of empty EQ-nominalizers.

The EQ involving the semantically active n does not denote a question per se, even though it is selected by a question-embedding predicate. There are several ramifications in this pattern. One is a case where the EQ can be analyzed as denoting something equivalent to ‘question’ or ‘issue’. The EQ with this interpretation may have a structure akin to a noun complement clause.

(30a) means that John answered a particular question. It can be paraphrased into the noun complement construction by adding the parenthesized elements toyuu situmon ‘comp question’ without significant change in meaning. Likewise, (30b) can be optionally accompanied by the element toyuu mondai ‘comp issue’. The EQs describe the specifics of the head nouns situmon ‘question’ and mondai ‘issue’. They play the same role in the version without the overt head noun. Based on this parallelism, we propose that the EQs here involve a semantically active empty noun whose meaning is something like ‘question’ or ‘issue’. It is merged with the interrogative CP, as illustrated below.

The silent content noun converts a set of propositions denoted by the interrogative CP into a contentful individual (Moulton 2020). Consequently, the EQ is compatible with the demonstrative ano ‘that’ (see (29)).

It is worth noting that an EQ nominalized by the semantically vacuous n (i.e., an EQ selected by an inquiry-type predicate) cannot be converted into the noun complement construction. For instance, the EQ in (32), which is selected by an inquiry-type predicate like tazuneta ‘asked’ and touta ‘questioned’, cannot be followed by the phrase toyuu situmon ‘comp question’.

The denotation of the EQ nominalized by the semantically vacuous n remains unchanged. It denotes a question, namely, a set of propositions.

The hidden noun complement structure with a silent content noun is observed cross-linguistically. Spanish has a nominalized clause that involves the determiner lo followed by the preposition de and the declarative clause headed by the complementizer que (Picallo 2002; Moulton 2020 among others).

(33) is analyzed as involving the noun complement structure with a null noun equivalent to overt nouns like ‘idea’ or ‘proposal’, as indicated by the translation. The presence of such a noun is verified by the preposition de, which is obligatory when the CP occurs as the complement of a noun.

Moulton (2020) observes that the relevant empty noun is equivalent to all-purpose content nouns such as kes in Korean. Japanese also has such an item, namely, koto. As illustrated below, koto is replaceable with a more concrete expression toyuu kangae ‘comp idea’.

The same mechanism is available in EQs. Each sentence in (30a, b) involves a semantically active all-purpose nominalizer, as schematically illustrated in (31). The only difference is that the relevant nominalizer, the counterpart of koto in the declarative clause, is phonologically empty in the EQ.

Another instance of EQ involving a silent content noun expresses a possible answer to the question. Consider the following example.

This sentence does not mean that John tracked down the question itself. It means that he found the answer to the question; that is, he identified the person who had stolen the money. Let us call this type the answer-denoting EQ. Although the EQ shows up in the shape of an interrogative clause, it virtually refers to (an) individual(s) that serve(s) as the answer to the wh-phrase. In this sense, (36) is in parallel with a relative clause.

The resemblance between the answer-denoting EQ and the relative clause construction has been noted in the literature. One notable characteristic of the answer-denoting EQ is that it can co-occur with a constituent that refers to the individual(s) that serve(s) as the answer to the wh-phrase. Tomioka (2020) notes that an internally-headed relative clause displays the same behavior. Compare the following pair.

The bracketed part of (38a) is an internally-headed relative clause. The nominalizer no refers to the noun okane ‘money’ inside the relative clause. It can be replaced with the DP sono okane ‘the money’. The DP sono hannin ‘the culprit’ in (38b) plays a similar role. It corresponds to the individual that serves as the answer to the wh-phrase inside the EQ. Based on this similarity, Tomioka postulates a covert nominalizer in the answer-denoting EQ that plays a similar, though not identical, role to the nominalizer no in the internally-headed relative clause.

While the exact semantic nature of the covert nominalizer has yet to be figured out, it is worth noting that it behaves differently from the semantically vacuous nominal head. The EQ involving the semantically vacuous nominal head cannot be followed by the DP denoting the answer.

The EQ selected by an inquiry-type predicate like situmonsita ‘questioned’ involves a semantically vacuous nominalizer. The sentence becomes unacceptable with the DP referring to the individual that constitutes the answer to the wh-phrase.

To sum up, there are at least two types of empty EQ-nominalizers: the semantically vacuous and the semantically active types.⁶ The former simply creates a nominal projection over an interrogative CP. Since the nominal projection inherits the interrogative nature of the EQ, it can be selected only by inquiry-type predicates like tazuneru ‘ask’, tou ‘question’, and situmonsuru ‘question’. On the other hand, the nominal heads of the latter type have meanings of their own such as ‘question’ or ‘issue’, and an EQ they nominalize has a hidden noun complement structure. Another instance of EQ nominalized by a silent content noun refers to a possible answer.

In the proposed analysis, the choice of n for nominalizing an EQ depends on the semantic selectional property of the matrix predicate. Consider the following example.

The apparently identical EQs here are nominalized by different n’s. The EQ occurring with the predicate situmonsita ‘questioned’ is nominalized by the semantically vacuous n. The one selected by hanasiatta ‘discussed’ involves the silent noun whose meaning is akin to ‘issue’. The one selected by sitteiru ‘know’ denotes the complete answer to the question.

4.1 NP- and DP-languages

Bošković (2005; 2008; 2009) proposes categorizing languages on the basis of the presence or absence of DP inside noun phrases. According to him, noun phrases in languages with overt articles like English have a DP layer above AP, which in turn dominates NP as in (42a).⁷ On the other hand, those in article-less languages like Serbo-Croatian take a different shape, as in (42b). They lack the DP layer, and AP is contained in NP instead of forming an independent projection in the nominal spine.

While retaining Bošković’s insights, we argue that although Japanese is an article-less language, noun phrases still involve DP. We further show that it is not totally assimilated with languages with articles but partially exhibits characteristics of article-less languages.

Bošković (2005) observes that demonstratives and possessives in article-less languages are not determiners but adjectives. Being adjectives, they are not located either in D or in Spec-DP but adjoined to NP. He further notes that demonstratives as adjectives display the following characteristics. First, they exhibit adjectival morphology. Second, the order between demonstratives and adjectives is relatively free. Third, a possessive cannot be modified by another possessive.

Demonstratives and possessives in Serbo-Croatian exhibit all of these characteristics, as illustrated below.

The demonstrative nekim ‘some’ in (43a) has the same inflectional ending -im that appears on the adjective maladim ‘young’. The order between a possessive and an adjective can alternate freely (see (43b)). Since the possessive and the adjective are members of the same category (i.e., A(djective)), they occur in the AP layer in (42b), where they can be aligned in either order. (43c) shows that the modification of a possessive by another results in ungrammaticality. Notice that English does not exhibit any of these characteristics, as indicated by the translation given to each example.

Since Japanese is also an article-less language, it is predicted that the same pattern as in (43) is obtained. However, the prediction is borne out only partially. Consider the following examples.

(44a) indicates that the demonstrative is morphologically distinguished from the adjective. While the adjective has an adjectival present tense inflectional ending, the demonstrative cannot carry it.⁸ As illustrated by (44c), modifying a possessive by another is possible. This is also a characteristic that is not found in Serbo-Croatian. The only property that Japanese shares with Serbo-Croatian is the alternation between the adjective and the demonstrative/possessive, as in (44b). Apart from this property, possessives and demonstratives in Japanese behave in parallel with those in English, a language with an overt article. In this sense, Japanese is a hybrid language with characteristics of both types of language (see e.g., Oda 2022; 2023 for recent arguments against the clear-cut distinction between DP- and NP-languages).

The morphological difference between demonstratives and adjectives (see (44a)) indicates that they belong to different grammatical categories in Japanese. (42b) is inappropriate for capturing their categorial difference since it presupposes that they belong to the same category (i.e., adjective). Alternatively, we propose the following structure for Japanese noun phrases.⁹

This structure is similar to the Serbo-Croatian-type noun phrase in that the demonstrative in the adjective-demonstrative (Adj-Dem) order is located in the NP layer. Adopting the widely made assumption that prenominal modifiers occur in the projection below DP (Abney 1987; Cinque 1994; Bošković 2005 among many others), we propose that an adjective is contained in NP. The Adj-Dem order indicates that the demonstrative in this order belongs to the NP layer. (45) is similar to the English-type noun phrase as well in that it has the DP layer hosting the demonstrative that appears in the demonstrative-adjective (Dem-Adj) order.

4.2 Two types of demonstratives and their positions

In the proposed structure, the word order alternation between demonstratives and prenominal modifiers in Japanese arises from a different structural factor than in Serbo-Croatian. As mentioned above, demonstratives and adjectives in Serbo-Croatian belong to the same category, which allows for word order alternation. On the other hand, demonstratives in Japanese can occur in distinct projections, namely, in DP and NP. This suggests that the Dem-Adj and Adj-Dem orders are not necessarily equivalent, though they look so.

One of the significant features of demonstratives resides in their anaphoric function.

The demonstrative refers back to the transfer to Kyoto, which was mentioned in the previous context. Under the proposed structure (45), the Dem-Adj order results from the placement of the demonstrative inside DP. In contrast, the Adj-Dem order is rooted in the placement of the demonstrative inside NP.

It is worth noting that the demonstrative does not always display free alternation with the adjective. In addition to the anaphoric use, the demonstrative may have the meaning equivalent to ‘that kind of’. Consider the following example.

Speaker A’s first utterance implies that there are several kinds of transfer: transfer with a promotion, demotion, pay raise, pay cut, and so on. Speaker B singles out one type, and the demonstrative in Speaker A’s second utterance refers to this particular type. Let us call this use the kind-specifying use for the sake of expository convenience.

Although ano is ambiguous between anaphoric and kind-specifying types, the ambiguity can be removed by using the demonstrative anna. As mentioned in note 8, it cannot be used deictically and anaphorically but can only act as a kind-specifying modifier. Thus, it is incompatible with contexts that force deictic and anaphoric readings.

It is worth noting that unlike the anaphoric demonstrative, anna is infelicitous in the position preceding a genitive-marked modifier, as indicated by (49B).

Given the oddity of (49B) with anna preceding the genitive-marked modifier, it is more reasonable to assume that the kind-specifying demonstrative can only occur in the NP-internal position. Additionally, the asymmetry between anna and the genitive-marked modifier in their relative ordering indicates that the two orders are not entirely equivalent.¹⁰ The discussion so far leads to modifying (45) to reflect the different distribution of anaphoric and kind-specifying demonstratives.¹¹

4.3 The extraction from nominal constituents

This subsection provides arguments to support the idea that the demonstrative in the Dem-Adj order is placed inside DP. Campbell (1996) argues that the demonstrative occupying Spec-DP is an overt manifestation of a specificity operator that induces specificity effects.¹² Noun phrases exhibit variations in the extraction of constituents. While a non-specific noun phrase allows extraction, a specific noun phrase does not.

Along the line of the oft-made observation that extraction out of DP proceeds through Spec-DP (Stowell 1989; Giorgi & Longobardi 1991; Szabolcsi 1994; Gavruseva 2000 among others), Campbell argues that the specificity operator in Spec-DP prevents the wh-phrase from using that position as an escape hatch. We also show in what follows that the demonstrative in Spec-DP functions as an operator that blocks extraction.

Akuzawa (2017) notes that an argument of an event nominal (aka a verbal noun) can occur either inside or outside the phrase headed by the event nominal.¹³

The PP Sapporo-kara Tokyo-e(-no) ‘from Sapporo to Tokyo’ is an argument of the event nominal tenkin ‘transfer’. The genitive marking on the PP in (52a) indicates that the PP is located inside the noun phrase (i.e., the bracketed part), whereas the lack of genitive marking in (52b) indicates that the PP is outside the noun phrase. According to Akuzawa, (52b) results from the syntactic movement of the argument. The Proper Binding Condition effect in (53a) and the ban on the resumptive pronoun in (53b) suggest the involvement of movement.

In the movement analysis, the ungrammaticality of (53a) is attributable to the unbound trace left inside the fronted noun phrase. (53b) is also accounted for naturally since a resumptive pronoun cannot replace a trace.

Akuzawa (2017) points out that when the event nominal co-occurs with a demonstrative, its argument cannot be placed outside the noun phrase.

The crucial difference between (52b) and (54b) lies in the presence or absence of the demonstrative. If one assumes, as we do in this paper, that the demonstrative occupies Spec-DP and prevents extraction out of a noun phrase, one may be inclined to ascribe the ungrammaticality of (54b) to the unavailability of this position as an escape hatch for the PP.

Notice, however, that the noun phrase ano tenkin ‘that transfer’ in (54b) only consists of the demonstrative and the noun. It is not immediately clear whether the demonstrative occupies Spec-DP or NP-internal position. Therefore, it is desirable to use a noun phrase containing a prenominal modifier to test the validity of the assumption that the demonstrative in Spec-DP blocks extraction. Bearing this in mind, consider the following dialogues.

The demonstrative in (55B) is used anaphorically, referring back to the transfer to Kyoto that is at issue. Its occurrence before the prenominal modifier suggests that it occupies Spec-DP. In this case, the PP-extraction degrades the sentence. By contrast, the demonstrative in the second sentence of (56B) is kind-specifying. It occurs in the post-adjectival position, as discussed earlier. The PP-extraction is legitimate in this case. The contrast between (55B) and (56B) confirms the view that the demonstrative in Spec-DP blocks the extraction.^14,15

4.4 Back to EQs

The discussion so far has revealed that the demonstrative in Spec-DP gives rise to the blocking effect, whereas elements inside NP do not. In other words, the demonstrative inducing the blocking effect endorses the involvement of DP in a noun phrase. Bearing this in mind, let us consider whether the same blocking effect arises in the nominalized EQ with a demonstrative. Consider the following example.

This sentence involves scrambling out of the EQ. Although it is admittedly awkward, presumably because the prenominal modification of an EQ causes certain awkwardness, it remains within the realm of grammatical sentences when the demonstrative is absent (see note 1 for discussion). Recall that a prenominal modifier such as a relative clause is contained in the NP layer (or the nP layer in the terminology of the present paper) rather than in the DP layer. The well-formedness of (57) in the absence of the demonstrative indicates that a constituent inside NP does not interfere with extraction. The extracted phrase makes use of Spec-DP as an escape hatch. The blocking effect induced by the demonstrative is attributable to the unavailability of Spec-DP for the scrambled phrase.¹⁶

The involvement of DP in the nominalization of EQ receives support from a different angle. As shown below, the nominal EQ can be referred back to by the pronoun sore ‘it’. Consider the following dialogue.

In (58B), the pronoun sore is juxtaposed with the EQ. The mild deviancy is due to the Double Accusative Constraint, but it does not go so far as to be ungrammatical (cf. Hiraiwa 2010). In particular, the sentence improves with a pause after the EQ.

Although sore can refer to a referential nominal, it cannot refer to a predicative nominal.

The noun kinzoku ‘metal’ in (59a) is an argument. Combined with the demonstrative sono ‘that’, it functions as a referential DP. Being referential, it can be referred to by sore (Noguchi 1995). The same noun in (59b) is a predicative nominal, which is a non-referential NP rather than a referential DP. It resists modification by the demonstrative and cannot be referred to by sore. It follows from the contrast in (59) that sore can only refer to a referential element. The presence of D plays a crucial role in inducing referentiality. Viewed in this light, the possibility of referring to the EQ with sore in (58B) suggests the presence of D in the EQ.

By contrast, the embedded clause introduced by to ‘that’, a canonical clausal category, does not exhibit nominal behavior. It resists demonstratives, adjectives, relative clauses, and case-marking.

It does not allow the apposition of sore, either, which indicates that clauses lack the D component.