1.1 Strictly Head-Final Structure

Japanese phrase structure is strictly head-final, as discussed, for example, in Kuno (1978a). The verb follows its complements and the nominal head appears at the end of a noun phrase as illustrated in (1).

(1)

The PP complement in (1b) shows that the language employs postpositions, which reflects its head-final nature.

The head-finality is observed with various functional categories as well. The structure of the sentence in (2) is shown in (3).

(2)

(3)

In this structure, the light verb v* follows its VP complement and T(ense) follows v*P.

Embedded CPs have complementizers (Cs) at the end. The ka in (4a) is the question C.²

(4)

On the other hand, to in (4b) is the quotative C, which follows a direct quotation or a paraphrase of a direct quotation in the sense of Plann (1982) and Saito (2015a).

A modifier precedes the modified. Thus, a phrase has its head in the final position without exception. (5a) is an example of modification by a relative clause.

(5)

An adjective modifies a noun in (5b), and an adverb modifies a verb phrase in (5c).³

1.2 Complex Predicates

One of the characteristic properties of Japanese phrase structure is that it is often more complex than the surface form suggests. Japanese is an agglutinative language and employs complex predicates extensively. A representative example is the causative form, instantiated in (6).

(6)

This looks like a simple sentence, but it has been a consensus since Kuroda (1965a) that it involves clausal embedding. Murasugi and Hashimoto (2004), for example, proposed that (6) has the structure in (7) with a small clause complement. It is then basically identical in structure to its English counterpart.

(7)

There is abundant evidence for clausal embedding in Japanese causatives. The best-known argument is based on the interpretation of the subject-oriented reflexive zibun. In (8), the matrix subject Hanako and the embedded subject Taroo are both possible antecedents for zibun but not the embedded indirect object Ziroo.

(8)

In the causative (9), not only Hanako but also Taroo can be the antecedent of zibun, although the causee Taroo appears in dative Case.

(9)

This indicates that Taroo is a subject, which in turn implies that there is clausal embedding in (9).

Another well-known argument is due to Shibatani (1976). The VP adverb isoide ‘in a hurry’ modifies Hanako’s action of dispatching Taroo to London in (10a).

(10)

The causative sentence (10b) is ambiguous. The adverb can modify Hanako’s action as in (10a), but it can also modify Taroo’s action of heading to London. This is predicted by the structure in (7) with an embedded v*P.

1.3 Structures and Selection

Japanese phrase structure adheres to X’-theory with the head-final structure in (11).

(11)

At the same time, it provides rich data indicating that there is no fixed universal clause structure with functional categories. If the standard structure of a clause is as in (12), Japanese clause structure can diverge from it in several ways.

(12)

This can be seen with modals, complementizers, and sentence-final discourse particles.

Japanese has modal elements that do not carry tense. Among the examples are daroo, which expresses speaker’s surmise, and its negative counterpart, mai. The former takes a TP complement, as illustrated in (13).

(13)

This shows that a modal can appear between C and T, as in (14).

(14)

The complement of the other modal, mai, is either TP with present tense as in (15a) or v⁽*⁾P as in (15b).

(15)

(15b), in particular, shows that even a matrix clause need not have T as long as it has a modal with information on tense.⁴ Its structure, when embedded, is as in (16).

(16)

Another deviation from the standard structure in (12) can be seen with stacking complementizers. Japanese has three complementizers, no, ka, and to. Examples of each are given in (17).

(17)

No is employed when the embedded clause expresses an eventuality. It typically appears in the CP complement of factive verbs and perception verbs like mi ‘see’ and kanzi ‘feel’. Ka is the question complementizer for yes/no questions as well as wh-questions. To marks direct quotations and their paraphrases as noted in section 1.1. A direct quotation can be any utterance or for that matter any sound, but when to embeds a paraphrase of direct discourse, its complement must be a grammatical full clause.

Interestingly, these three complementizers can co-occur. An example is shown in (18).

(18)

Four patterns are possible for the complementizer(s) of the embedded clause in (18), ka, no-ka, ka-to, and no-ka-to. When two or more complementizers co-occur, they must appear in this fixed order. These complementizers have been analyzed in terms of Rizzi’s hypothesis (1997) on the structure of the clause periphery. Hiraiwa and Ishihara (2002) proposed that no is the Finite head, and Saito (2015a) named to a Report head, following Lahiri’s analysis (2002) of Spanish. Then, the Japanese complementizers form the following hierarchical structure, where ‘Int’ stands for Interrogative:

(19)

Japanese clausal phrase structure is further enriched by sentence-final discourse particles. Some of them are shown in (20).

(20)

These particles appear only in matrix clauses. Wa, which is typically employed in women’s speech, and yo both convey assertion. Ne has the effect of eliciting the addressee’s response and is often translated into English as a tag question. They can co-occur but only in the order indicated in (20).

It should be clear from this brief illustration that Japanese phrase structure has various significant features aside from its head-finality. The structures introduced observe the selectional properties of the relevant heads. It was shown in (13) that the modal daroo takes a TP complement. In other words, it selects for T. The complementizer no has the same selectional property. For example, it is incompatible with a ModalP complement.

(21)

It contrasts with the other complementizers in this respect. Example (22) shows that neither ka nor to has this selectional restriction.

(22)

This is one reason no is at the bottom of the hierarchy of complementizers. As noted, to embeds a paraphrase of a direct discourse. The paraphrased direct discourse can be a question. Hence, to can take a CP complement headed by ka. On the other hand, a question is formed on a clause with a truth value. As to-headed CPs express paraphrases of direct discourse, truth values are irrelevant for their interpretation. Hence, ka cannot take a to-headed CP as its complement. Thus, the hierarchy in (19) is predicted from the lexical properties of the complementizers.

Exactly the same point can be made for the hierarchy of discourse particles. The particle wa selects for T, just like the modal daroo and the complementizer no.

(23)

This explains why wa occupies the lowest position in the hierarchy of discourse particles. The sequences wa-ne, yo-ne, and wa-yo-ne, are fine because wa and yo indicate that the speaker is making an assertion and ne solicits the addressee’s response to it. However, *ne-yo does not make sense because one cannot assert solicitation of a response. The hierarchy of discourse particles is also predicted by the lexical properties of the particles.⁵

Japanese clause structure exhibits patterns that are not observed in other languages. At the same time, the discussion in this section has shown that whether a certain head-complement structure is possible or not depends on the selectional property of the head. This conclusion is in line with Chomsky’s minimalist hypothesis (1994) that phrase structure is freely generated by the operation Merge, which takes two elements α, β‎ and forms the constituent {α, β‎}, and that language variation is due to variation in functional heads and their properties.

1.4 Some Controversial Cases

There are some controversial issues on the phrase structure of Japanese. One has to do with the position of the nominative subject in tensed clauses. It is assumed in this article that it raises to the specifier position of TP. This hypothesis was entertained, for example, in Saito (1985) and more recently in Kishimoto (2017). On the other hand, Kuroda (1988) and Miyagawa (2001), among others, argued that it can stay within v⁽*⁾P. The arguments on both sides are based on the analysis of Case, scope, and scrambling in the language, and are mostly theory-internal. It remains to be seen whether there is more direct evidence to distinguish the two positions.

Another issue concerns the structure of nominal phrases. Fukui and Speas (1986) and Abney (1987) proposed that nominal phrases are DPs (Determiner Phrases), for example, in English. According to this hypothesis, the is a D head and a genitive phrase occupies the specifier position of DP, as illustrated in (24)–(25).

(24)

(25)

Fukui (1986) noted that Japanese noun phrases differ from their English counterparts in many respects. For example, Japanese lacks a definite article, and noun phrases in the language can contain multiple genitive arguments as shown in (26).

(26)

Examples of this kind suggest that genitive Case is not assigned to DP Spec, but to any nominal phrase or PP within the projection of N. Given this, Fukui concluded that Japanese lacks the D projection.⁶ The hypothesis was developed by Bošković (2012), who proposed that there is a major parameter that distinguishes between DP languages and NP languages.

On the other hand, Saito and Murasugi (1990) argued that DP is present in Japanese. They first presented a generalization on ellipsis as follows:

(27)

This is instantiated by the contrast in (28) with N’-ellipsis.

(28)

Only in (28a), the specifier of DP is present, and, hence, the complement NP can be elided. Saito and Murasugi (1990) then argued that N’-ellipsis is observed in Japanese as well. Relevant examples are shown in (29).

(29)

If this ellipsis phenomenon falls under the generalization in (27), then the remnant genitive phrases in (29a–b) must be in the specifier position of DP.

Independently of this NP-DP controversy, it was proposed by Travis and Lamontagne (1992) and Fukuda (1993) that arguments in Japanese are headed by Case markers, as illustrated in (30) for the accusative o.

(30)

The main motivation for this analysis is the distribution of Caseless arguments. Kuno (1973) pointed out that the accusative Case marker o can be omitted in casual speech, but the nominative Case marker ga cannot. The examples in (31) confirm this generalization.⁸

(31)

Furthermore, the accusative marker is required when the object is moved to the sentence-initial position.

(32)

Thus, Caseless arguments are allowed only in the complement position.

The distribution of Caseless arguments, then, mimics that of CPs with null C, illustrated here.

(33)

Stowell (1981) and Kayne (1981) assumed that CP is headed by a null C when that is missing and presented an analysis for (33) in terms of the empty category principle. This analysis can be extended to (31) and (32) if Caseless arguments are headed by a null K.

Then, there are four possibilities for Japanese nominal arguments.

(34)

It remains to be seen which one of these is correct. But each has important theoretical implications. The analyses in (34b) and (34d) suggest that nominal arguments must be DPs to be properly interpreted; those in (34a) and (34c) imply that there is no semantic requirement of this kind. They suggest that D is assumed only with overt evidence, for example, the presence of a definite article. The analyses in (34c) and (34d) lead to a similar conclusion for K. K is present in Japanese as the locus of Case feature because the language has overt Case particles as lexical elements. Japanese, then, may provide evidence for possible language variation with respect to (minor) functional categories. The analysis in (34a) is assumed in the remainder of this article but only for expository purpose.

2.1 Scrambling

Japanese has a scrambling rule, which freely preposes major constituents. Examples of clause-internal scrambling and long-distance scrambling are given in (35a) and (35b), respectively.

(35)

The rule is well-motivated and has properties different from A-movement and operator movement.

One of the well-known arguments for scrambling is due to Kuroda (1980). In Japanese, floating numeral quantifiers are used productively, as in (36).

(36)

San-nin ‘3-Cl.’ in (36a) is associated with the subject and go-hon ‘5-Cl.’ in (36b) with the object.

Kuroda pointed out that a floating numeral quantifier must occur adjacent to the argument it is associated with. Example (37a) is ungrammatical because the intervening object breaks the relation between the numeral quantifier and the subject.

(37)

Kuroda then observed that (37b) is perfectly grammatical although the subject intervenes between the numeral quantifier and the object. This, he argued, is expected if the OSV order is derived from SOV by scrambling the object to the sentence-initial position; (37b) is then derived as in (38).

(38)

The object sake-o ‘sake-ACC’ originates within VP and moves to the position preceding the subject. Example (37b) is grammatical because the object is adjacent to the numeral quantifier go-hon ‘5-Cl.’ in its initial position.

Scrambling is not NP-movement. Scrambled objects, for example, are assigned Case prior to scrambling, and hence, scrambling is not Case-driven. Further, long-distance scrambling as in (35b) shows that scrambling is not subject to the locality requirements imposed on NP-movement. It was shown in Saito (1989) that scrambling is not operator movement either. The argument is based on the distribution of wh-phrases, illustrated in (39).

(39)

In these examples, the embedded clauses are questions as indicated by the question complementizer ka. In (39a), the wh-phrase dare ‘who’ is within the embedded clause and the embedded CP is interpreted as a wh-question. In (39b), on the other hand, dare is the matrix object and is located outside the embedded question. In this case, the sentence is totally ungrammatical. This illustrates the generalization that a wh-phrase must be part of a question sentence to be properly interpreted.

Given this generalization, it is somewhat surprising that (40a–b) are both grammatical.

(40)

Example (40a) is straightforward, as the wh-phrase nani ‘what’ is contained within the embedded question. In (40b), this wh-phrase is scrambled out of the embedded question to the initial position of the main clause. The wh-phrase is not contained within a question, just as in (39b), but nevertheless, the example is grammatical. Saito (1989) proposed on the basis of examples of this kind that a scrambled phrase can be interpreted at its initial position instead of its landing site. This, which is called the radical reconstruction property of scrambling, indicates that scrambling is not operator movement because if it were, the scrambled phrase would be interpreted at the landing site as an operator.⁹

The radical reconstruction property may suggest that scrambling is a PF stylistic rule, as assumed in Chomsky and Lasnik (1977). However, scrambling has semantic effects. For example, Webelhuth (1989) and Mahajan (1990) examined German and Hindi scrambling, respectively, and showed that a phrase preposed by clause-internal scrambling can serve as the antecedent of an anaphor. Tada (1993) and Nemoto (1993) demonstrated that this phenomenon is observed in Japanese as well. Example (41a) is ungrammatical as the anaphor otagai ‘each other’ is not bound.

(41)

However, the same sentence becomes grammatical when the intended antecedent karera ‘they’ is scrambled from the object position to the sentence-initial position as shown in (41b).

The contrast in (41) does not obtain with long-distance scrambling. But long-distance scrambling too can affect interpretation. This is observed when scrambling is applied to anaphors, as discussed by Dejima (1999). Zibunzisin ‘self.self’ is a local, subject-oriented reflexive. So, only Ziroo, the subject of the most deeply embedded clause, can be its antecedent in (42a).

(42)

But (42b) indicates that the middle subject Hanako also becomes a possible antecedent when zibunzisin is scrambled to the initial position of the most deeply embedded CP. In (42c), zibunzisin is scrambled long-distance out of a CP to the initial position of the middle clause. This makes the main clause subject Taroo qualify as the antecedent of zibunzisin. Thus, the landing site of long-distance scrambling counts when the local relation of an anaphor and its antecedent is calculated. Scrambling in Japanese can affect the interpretation of anaphors whether it is clause-internal or long-distance, and, hence, it should be considered a syntactic movement operation with unique properties.

How Japanese scrambling should be analyzed and why the language allows this unique syntactic operation are still active research topics. The reader is referred to Miyagawa (2010, 2017), Saito (2015b, 2016), and the references cited there for various proposals on these questions.

2.2 NP-Movement

NP-movement in Japanese is observed in passive and unaccusative sentences. The best-known evidence for this is presented in Miyagawa (1989). His argument is similar in form to Kuroda’s argument (1980) for scrambling. Recall that floating numeral quantifiers must be adjacent to the noun phrases they modify. Thus, (43a–b) are ungrammatical.

(43)

Miyagawa (1989) pointed out against this background that (44a–b) are grammatical, in sharp contrast with (43a–b).

(44)

In (44a), for example, kono kagi-de ‘with this key’ intervenes between the numeral quantifier and the associated noun phrase, just as in the ungrammatical (43a). Yet, this example is perfectly grammatical. Miyagawa (1989) pointed out further that the unaccusative hypothesis correctly predicts the grammaticality of (44a–b). According to this hypothesis, the theme argument of an intransitive verb originates in the object position and moves to the subject position. Then, the structure of (44a) is as in (45).

(45)

Doa ‘door’ is adjacent to huta-tu ‘2-Cl.’ in its initial position.

A similar argument can be constructed for NP-movement in passive. The following examples are grammatical, just like (44):¹⁰

(46)

These examples are predicted to be grammatical if the subject originates in the object position within VP and moves to the sentence-initial position.

The movement in (44) and (46) is not scrambling. As noted in section 1.2, the Japanese reflexive pronoun zibun is subject-oriented. Thus, the subject Hanako is the only possible antecedent for zibun in (47a).

(47)

The interpretation is not affected even when the object Taroo is scrambled to the sentence-initial position as in (47b). Hanako is still the only subject of the sentence. On the other hand, the nominative phrase in passive and unaccusative sentences can be the antecedent of zibun. Example (48) shows this for passive.

(48)

The movement in passive and unaccusative sentences thus must be NP-movement to the subject position.

The movement exhibits the locality expected of NP-movement. The contrast between (49b) and (50b) is discussed in Inoue (1978).

(49)

(50)

In (49b), the object moves across the dative argument to the subject position. The apparently similar movement is illicit in the causative (50b). This is predicted because the dative argument in (50) is the embedded subject as discussed in section 1.2, and NP-movement across a subject is prohibited.

Raising in Japanese is discussed most extensively with complex predicates. First, start in English can be a raising predicate as well as a control predicate; (53a) can be an example of control, whereas (51b) with the weather it instantiates the raising case.

(51)

The difference can be seen in the contrast in (52).

(52)

Kuno (1978b) pointed out that the same ambiguity is observed with the verb hazime ‘start’ in Japanese when it is employed as the head of a complex predicate. The examples in (53)–(54) are the Japanese counterparts of (51)–(52).

(53)

(54)

In (53b), ame ‘rain’ originates in the object position of the unaccusative verb hur ‘fall’ and raises to the subject position of the sentence. Control and raising with complex predicates are discussed extensively in Koizumi (1995).

2.3 Operator-Movement

Overt operator movement was proposed for comparatives by Kikuchi (1987) and for clefts by Hoji (1987). Kikuchi (1987) demonstrated that island effects are observed with Japanese clausal comparatives. The contrast in (55) instantiates this generalization.

(55)

Example (55a) is grammatical with the gap in a to-headed CP, but (55b) is ungrammatical because the gap is contained within a relative clause. Kikuchi, then, proposed that clausal comparatives are derived by null operator movement as in (56).

(56)

Hoji (1987) pointed out that clefts, too, exhibit island effects, as illustrated in (57).

(57)

The gap is contained within a relative clause in (57b). The ungrammaticality of the example, then, suggests that clefts are derived by movement.

It should be mentioned that Hoji (1987) pointed out a further contrast between (57b) and (58).

(58)

The focus is a PP in (57b) whereas it is an NP in the grammatical (58). Hoji proposed that the two examples have the structures in (59a, b), respectively.

(59)

That is, a cleft sentence is derived by empty operator movement when the focus is a PP, but when the focus is an NP, the gap can be pro and need not be produced by movement.

This analysis is confirmed by another contrast shown in (60).

(60)

This example shows that an overt resumptive pronoun is allowed only with an NP focus. The pro in (59b) is then nothing but a covert counterpart of the pronoun in (60). Also, as a pronoun is disallowed in PP clefts, the gap in (59a) cannot be pro but must be produced by movement.¹¹

Although it is well-established that comparatives and clefts in Japanese involve operator movement, many attempts have been made to develop the original analyses by Kikuchi (1987) and Hoji (1987). Among the important works for comparatives are Ishii (1991), Beck et al. (2004), Bhatt and Takahashi (2011), and Sudo (2015). An alternative analysis for clefts is proposed in Hiraiwa and Ishihara (2002), and multiple-foci clefts are discussed in Koizumi (1995) and Takano (2002, 2020) together with their theoretical implications.

Analysis of wh-questions in Japanese is also a lively research topic that relates to operator movement. There are two major approaches for the analysis, one with operator movement and the other without. The latter originated with Kuroda (1965b) and is based on the observation that the interpretation of a wh-phrase depends on the associated quantificational particle. The examples in (61) illustrate the fact.

(61)

In (61a), dare indeed seems to be construed as an interrogative wh-phrase. However, it is part of universal quantification in (61b), an example cited from Takahashi (2002). Then, the quantificational force originates in the particles, ka and mo, instead of dare. Kuroda (1965b) made this observation and called the wh-words ‘indeterminate pronouns’. He explicitly stated that they function as “yet unbound variables.” Nishigauchi (1990) developed this idea and proposed an analysis in terms of Heim’s unselective binding (1982). According to this analysis, mo serves as a universal quantifier and dare is interpreted as a variable in (64). The analysis is developed further in Shimoyama (2001).¹²

The other approach assumes that a wh-phrase in a question is an interrogative operator and moves covertly to its scope position. The analysis was inspired by Huang’s analysis (1982) of Chinese and was developed by Lasnik and Saito (1984), Richards (2001), and Watanabe (1992), among others.¹³ Among the motivations for this approach is the fact that the wh-island effect is observed between a wh-phrase and its scope position, as discussed in detail in Nishigauchi (1990) and Watanabe (1992). This can be seen in the following examples:

(62)

Example (62a) shows that the question C ka yields a yes/no question in the absence of a wh-phrase. The most deeply embedded subject Taroo is changed to dare ‘who’ in (62b). This example demonstrates that a wh-phrase and the associated ka can be separated by a clause boundary. Finally, ka is substituted for the most deeply embedded C to in (62c). This example should be ambiguous. If the wh-phrase is associated with the lower ka, it should have the interpretation in A, which is indeed possible. On the other hand, if it is associated with the higher ka, the lower ka should be interpreted as ‘whether’ and the interpretation in B should obtain. But this interpretation is more difficult. This wh-island effect suggests that Japanese wh-questions are derived by movement.

Saito (2017) developed the movement analysis in a way that is consistent with the variation in wh-interpretation. The proposed analysis works out Nishigauchi’s suggestion (1990) that a Japanese wh-phrase obtains quantificational force from the associated quantificational particle. It is illustrated for a wh-question in (63).

(63)

A wh-phrase enters the structure as an operator with unspecified quantificational force. It covertly moves to the specifier position of a quantificational particle and obtains specific force from the particle. In the case of (63), the wh-phrase becomes an interrogative operator because of the question particle ka. Similarly, the wh-phrase dare in (61b) moves covertly to the specifier position of the particle mo ‘also’ and receives the force of a universal quantifier.¹⁴

The precise analysis of wh-questions in Japanese is still controversial. But if the classical covert movement analysis or its extension just outlined is on the right track, wh-phrases in Japanese are subject to covert operator movement.

2.4 Head Movement

It is controversial whether head movement is observed in Japanese. Japanese phrase structure is strictly head-final, and a modifier always precedes the modified. Further, negation is an adjectival suffix that carries tense, as shown in (64).

(64)

Therefore, it is impossible to see, for example, whether a verb moves to Tense across an adverb or negation. Nevertheless, it has been proposed by Otani and Whitman (1991) and Koizumi (1995) that a verb raises to Tense in Japanese.

Otani and Whitman’s argument (1991) is based on the fact that null objects can receive sloppy interpretation as well as strict interpretation, as shown in (65).

(65)

Japanese is a radical pro-drop language and allows a null pronoun in any argument position. However, if a pronoun occupies the object position in (65b), only the strict reading is predicted. Example (66), as a sequel to (65a), is unambiguous.

(66)

Hence, the sloppy reading of (65b) requires an independent account. Otani and Whitman (1991), following Huang’s analysis (1987) of similar examples in Chinese, proposed that the object gap in (65b) can be produced by V-stranding VP ellipsis. The analysis is illustrated in (67).

(67)

The verb moves out of VP to T, and the remnant VP, which contains only the object, is elided. This analysis correctly predicts the sloppy reading, and if correct, implies that V raises to T in Japanese.

Oku (1998) and Kim (1999), however, pointed out that there are examples that are similar to (65) but cannot be explained in terms of V-stranding VP ellipsis. Oku, for example, showed that sloppy interpretation is possible not only for null objects but also for null subjects. One of his examples is given in (68).

(68)

If the subject is located outside an elided VP, as Otani and Whitman (1991) assumed, then (68b) cannot be accounted for with VP ellipsis. Both Oku (1998) and Kim (1999) concluded that Japanese/Korean allow arguments such as subjects and objects to be directly elided. This came to be known as ‘argument ellipsis’.

Supporting evidence for argument ellipsis has been presented in many works including Takahashi (2008, 2014) and Sakamoto (2016), and it is well established. The remaining issue is whether V-stranding VP ellipsis exists along with argument ellipsis. If it does, it should be maintained that V moves to T in Japanese. Oku (1998) argued against this possibility with examples such as (69).

(69)

The issue is whether (69b) can mean ‘John washed the car but not thoroughly’. V-stranding VP ellipsis predicts that this interpretation is possible. Oku’s judgement is that the example can only mean ‘John didn’t wash the car’. On the other hand, Funakoshi (2016) pointed out that there is variation in judgement with examples of this kind and argued in detail that the interpretation in question is possible when appropriate context is provided. The issue here is complex in part because the interpretation is predicted not only by V-stranding VP ellipsis but also by extension of argument ellipsis to adverbs.

Koizumi (1995, 2000) argued that V moves to T and further to C in Japanese. His argument is based on multiple-foci clefts, exemplified in (70b).

(70)

In the cleft sentence (70b), formed from the basic sentence in (70a), both the subject and the direct object occupy the focus position.

However, Koizumi (2000) argued that what is focused in (70b) is a single constituent. He proposed that V raises to T and then to C as in (71), and the shaded remnant TP is placed into the focus position in (70b).

(71)

According to this analysis, a “multiple-foci cleft” always has a remnant VP or TP in the focus position.

A piece of evidence for this analysis is that the “double foci” must be clause-mates in examples of this kind. The contrast between (72b) and (72c) instantiates this generalization.

(72)

If there are two foci in (72b) and each of them is associated with its gap independently, (72c) is also expected to be grammatical. Koizumi’s analysis, however, correctly predicts the contrast. Example (72b) is analyzed straightforwardly: What is focused in (72a) is the embedded TP after V+T moves out of the TP to C, but (72c) cannot be derived. The only possibility is to apply massive movements as in (73) and to focus the shaded VP.

(73)

But the movement of the embedded V+T+C to the matrix V is illicit.

Although Koizumi’s analysis is quite elegant, Takano (2002) raised questions on the analysis and proposed an alternative. His proposal on “double-foci clefts” is that one focus adjoins to the other and then the formed complex constituent is placed in the focus position. Sohn (1994) showed on independent grounds that the operation to form a constituent of two clause-mate phrases as in (74) is available in Japanese and Korean.

(74)

Among the evidence is the contrast between (75b) and (76b).

(75)

(76)

Huang (1982) observed that the adjunct wh, naze ‘why’, is disallowed in a relative clause, as indicated in (75a). Example (75b) shows that scrambling of this wh out of a relative clause is strictly prohibited, but (76a) is only marginal because scrambling targets an argument wh, nani ‘what’, in this example. Sohn (1994) pointed out that (76b) has the same grammatical status as (76a) and surprisingly can be properly interpreted as the two wh-phrases originating within the relative clause. His analysis is that naze adjoins to nani and gets a “free ride” to the sentence-initial position. Takano (2002) proposed that “double-foci clefts” are possible in Japanese because two foci can be combined and form a constituent just like the two wh-phrases in (76b). He showed that the properties of “double-foci clefts,” including the clause-mate condition, follow from this analysis.

Thus, Koizumi’s argument (2000) for head movement is not conclusive at this point. Advancement in the analysis of ellipsis and clefts should make it clear whether head movement is observed in Japanese.