Licensing of matrix questions in Japanese and its implications

Japanese allows matrix questions either with or without a Q-particle. It is therefore often assumed that a Q-particle is optional in matrix questions and questions can be licensed by a null Q-particle or rising-intonation. This paper, however, shows that some matrix questions require an overt Q-particle, which undermines the assumption that a null Q-particle or rising intonation can license sentences as interrogative. This study thus pursues the mechanism of licensing matrix questions without this assumption. In particular, I propose that yes/noand wh-questions without an overt Q-particle are licensed via Agree with their relevant items (i.e. a polarity head and wh-phrase). I further argue that there are two positions where matrix yes/no-questions are licensed in Japanese. This two-layered approach can capture the typological distribution of Q-particles and universal properties of matrix yes/noquestions.


.[TP Subj/wh V-T]…] ka[Q]]
The main focus of this paper is how matrix questions without an overt Q-particle are typed as interrogative. It has been traditionally supposed that the Q-particle ka is optional in Japanese and questions without ka are licensed as interrogative by a null Q-particle or rising-intonation (K. Yoshida and Yoshida 1996;T. Yoshida 2012;Kuwabara 2013). This is often referred as question marker drop or Q-drop. Under this assumption, it is expected that a null Q-particle should behave the same as an overt Q-particle such as ka in matrix questions; otherwise it would be difficult for children to acquire the distribution of a null Q-particle. Especially if there is a situation where a null Q-particle cannot appear in a matrix question but an overt Q-particle can, then the acquisition of the null Q-particle cannot be done without negative evidence.
I will show that the distribution of matrix questions without a Q-particle is different from the one of matrix questions with a Q-particle in Section 2. This motivates a licensing mechanism of matrix questions without positing a null Q-particle. I then propose different licensing mechanisms for matrix questions with and without a Q-particle in Section 3. In particular, I propose that matrix questions without Q-particles are licensed via Agree with the relevant items with a [Q]-feature. I also argue that there are two positions to license matrix questions, namely, ForceP and Speech Act Phrase (SAP) (Speas and Tenny 2003;Haegeman and Hill 2013). The implication of this two-layered approach will be discussed in Section 4.

Matrix questions with/without Q-particles.
This section shows that the distribution of matrix questions with a Q-particle is different from the one of matrix questions without a Q-particle. I first show that Q-particles are not optional in matrix questions, which suggests that there are no null counterparts of Japanese overt Q-particles. (4) shows that the Q-particle ka must be absent when the verb is not a polite form, although it is optional when it follows the polite form of the verb, as we have seen in (1) and (2).
'Who came?' This shows that there is a restriction on the use of the Q-particle ka. Nevertheless, the deviance of (4a) does not necessarily mean that matrix wh-questions with a non-polite verb can never be grammatical with a Q-particle. Japanese has another Q-particle kke, which also marks a sentence as interrogative (cf. (8)), and this particle does not require a polite form to the preceding verb.
(5) Dare-ga ki-ta kke? who-NOM come-PAST Q '(Tell me again) who came?' The Q-particle kke is known as triggering a remind-me presupposition, as shown in the translation (Sauerland and Yatsushiro 2017). 2 This presupposition cannot be obtained when kke is 2 Where kke occurs is a complicated issue. Kke seems to appear in the Force head like ka, but it behaves as a discourse particle, which is often considered to appear somewhere in Speech Act projections. What is more interesting and puzzling is that kke can appear under ka, as shown in (i) (Sauerland and Yatsushiro 2017 (4b). These data thus suggest that there are no null counterparts of overt Q-particles such as ka and kke in Japanese. They also show that an overt Q-particle is not a simple realization of a [Q]-feature (Ginsburg 2009). The Q-particle ka obeys some constraint, which does not come from a [Q]-feature of ka itself given that (4b), which I assume carries a [Q]-feature, is grammatical. 3 The Q-particle kke contains not only a [Q]-feature but also other features that induce a remind-me presupposition.
Although we have seen that Q-particles are not optional, it does not conclusively show that Japanese does not have a null Q-particle at all. There might be a case that Japanese null Q-particle is just a simple realization of a [Q]-feature unlike overt Q-particles and it does not have any other features whose distribution is affected by a polite form or which induce a remind-me presupposition. This idea would rather be preferable given that imposing many features on a null element is against a minimalist view and is in fact problematic from the perspective of acquisition. However, even this idea cannot be maintained in Japanese. I provide two empirical arguments that null Q-particles do not exist in Japanese. First, under a reportative use of the discourse particle yo (Kuroda 1973;cf. Davis 2011), yes/no-questions cannot be formed without an overt Q-particle. (6a) has the Q-particle ka followed by yo and it contains a yes/no-question in its meaning, whereas a sentence without a Q-particle cannot form a yes/no-question under the scope of the discourse particle yo, as shown in (6b) Note that it is possible to embed a wh-question without a Q-particle under yo, as in (7). 4 (7) [ Dare-ga ki-ta no ] yo. who-NOM come-PAST Fin Report In this paper, I would like to tentatively suppose that kke occupies Force like ka, and leave the issue of where kke exactly appears for future research. 3 The question of when ka can appear is not the focus of this paper. The main concern of this paper is how matrix questions without a Q-particle such as ka are typed as interrogative. For the distribution of ka, see Miyagawa (1987Miyagawa ( , 2017 and also Sect. 4.2. 4 Although the questions in (6) and (7) are "embedded" under yo, they are still considered as matrix questions. Pure embedded questions must carry the Q-particle ka, as shown in (i).

(i)
[ Dare-ga ki-ta (no) *(ka)] osie-te. who-NOM come-PAST Fin Q tell-IMPR 'Tell me who came.' Therefore, the contrast between (i) and (7) regarding the existence of ka suggests that the wh-question in (7) is not truly embedded. Note also that the Q-particle ka is attached to the embedded wh-question in (i) without a polite form of the verb, which is not allowed in matrix wh-questions. (ii) shows that wh-questions under yo cannot carry the Qparticle ka like (4a) because the verb does not carry a politeness affix.
(ii) *[ Dare-ga ki-ta (no) ka] yo. who-NOM come-PAST Fin Q Report '(Tell me) who came.' '(Tell me) who came.' (6a) and (7) illustrate that a sentence embedded under yo can form a question with either the Qparticle ka or a wh-phrase, but it cannot be a question when both of them are missing, as in (6b). In other words, (6b) cannot carry a [Q]-feature, which would be unexpected if Japanese had a null Q-particle in the lexicon which purely represents a [Q]-feature.
Another piece of evidence that Japanese does not have null Q-particles comes from so-called in-situ focus constructions, illustrated in (8) and (9). In in-situ focus constructions, a sentence is headed by the complementizer no or n, which I assume is a Fin head (Hiraiwa and Ishihara 2002; see also f.n.1) and followed by the non-polite copula da or polite copula desu. In (8a), the Q-particle kke follows the copula da, and in (9a) the Q-particle ka follows the polite copula desu. Importantly, the overt Q-particles in (8a) and (9a) cannot be dropped, as illustrated in (8b)  This suggests that in-situ focus constructions cannot form yes/no-questions without an overt Qparticle. Nevertheless, it is not the case that in-situ focus constructions can never be a question without an overt Q-particle. The wh-question counterparts do not require an overt Q-particle, as shown in (10b) and (11b).
Thus, to form a matrix question in in-situ focus constructions, either an overt Q-particle or whphrase is necessary. The ungrammaticalities of (8b) and (9b) suggest that Japanese does not have null Q-particles. If it did, (8b) and (9b) should be able to form a question like (8a) and (9a).
In this section, we have seen that (i) there are no null counterparts of the overt Q-particle ka and kke in Japanese and (ii) even a null Q-particle that is a simple realization of a [Q]-feature does not exist. One could still impose some constraint on the distribution of null Q-particles to explain the unavailability of question interpretations in (6b), (8b) and (9b), but as discussed in Section 1, such constraint on null elements is difficult for children to learn (i.e. negative evidence would be required). I thus pursue a licensing mechanism of matrix questions in Japanese without relying on null Q-particles.

Proposal.
If null Q-particles do not exist in Japanese, the question is how matrix questions without an overt Q-particle are licensed as interrogative. I first illustrate the proposal for yes/noquestions and then for wh-questions.
3.1. YES/NO-QUESTIONS WITHOUT A Q-PARTICLE. The contrasts between the a-and b-examples in (6), (8), and (9) suggest that the licensing conditions of yes/no-questions with and without a Qparticle are different. I propose that matrix yes/no-questions without an overt Q-particle are typed as interrogative in the Speech Act (SA) projection, that is, the speaker-oriented discourse domain located above the matrix CP (Speas and Tenny 2003;Haegeman and Hill 2013). According to Speas and Tenny (2003) and others, there are at least two projections in the SA domain. I argue that matrix yes/no-questions without a Q-particle are typed as interrogative when the topmost SA head obtains a [Q]-feature via Agree with the Pol(arity) head in the IP domain (Bailey 2013;Holmberg 2015). I assume that the Polarity head carries a [Q]-feature. The structure of the yes/no-question in (1b), repeated below as (12a), is illustrated in (12b).
(12) a. Hanako-wa ki-masi-ta? Hanako The licensing mechanism of (12b) can then be viewed as turning the whole utterance into a question at the highest position in the SA projection.
Recall that a sentence with ka can form a yes/no-question under the scope of the discourse particle yo, but one without ka cannot. To explain the contrast above, I first show that the discourse particle yo is a SA element. I assume that a particle which cannot be embedded is a SA element given that the SA projection is located above the matrix CP domain. As in (14a), yo cannot appear in the embedded clause when the matrix subject Hanako binds the embedded subject zibun 'self.' The use of the anaphor zibun in (14a) confirms that the clause is truly embedded. Note that if the subject is watasi 'I' and the "embedded" clause can be considered as a direct quotation, then the particle yo can appear in that clause, as shown in (14b).
(14) a. Hanako-wa [ zibun-ga hannin da (* yo) to] kokuhaku-sita. Hanako-TOP self-NOM culprit COP Report C confession-did 'Hanako confessed that she is the culprit.' Hanako-TOP I-NOM culprit COP Report C confession-did 'Hanako confessed "I am the culprit." ' Therefore, the fact that a sentence with yo cannot be truly embedded suggests that yo is a SA element. The current proposal then straightforwardly accounts for the contrast between (13a) and (13b). Their structures are illustrated in (15).
When the Q-particle ka, which appears in the Force head, is present as in (13a), it marks a sentence as interrogative under the scope of the SA element yo. On the other hand, I have proposed that when it is absent, then a yes/no-question is licensed at the highest position in the SA projection. Since (13b) does not contain an overt Q-particle, it cannot be interpreted as a yes/noquestion under the scope of the discourse particle yo.
The current approach can also account for the impossibility of yes-no questions in in-situ focus constructions without an overt Q-particle. The relevant examples are repeated below. I argue that the deviance of (16b) and (17b) is due to an effect of Relativized Minimality (Rizzi 2004). Following Hiraiwa andIshihara (2002, 2012) and Kuwabara (2013), I assume that the copulas da and desu are a Foc(us) head, not T like English copulas such as is. Kuwabara (2013) reports that these copulas can be added to a constituent on which the speaker puts some emphasis. 67 6 According to Hiraiwa and Ishihara (2012, f.n. 10), a copula tends to be grammaticalized into a focus particle crosslinguistically (see Cable 2010for Tlingit, Hiraiwa 2005 for Bùlì and Li and Thompson 1981 for Mandarin). 7 The following examples also show that the copula desu is different from English copulas like is in terms that it does not mark tense information of adjectives. Note that the copula desu is optional whether the adjective in (i) is a present form or past form.
This suggests that the copula desu functions as something different from tense unlike English is. I assume that desu in (i) is also a Focus head. Interestingly, (i) with desu cannot be a yes-no question without a Q-particle (Yoshida and Yoshida 1996).
(ii) a. * Omosiro-i desu? b. * Omosirok-atta desu? interesting-PRES Foc interesting-PAST Foc The ungrammaticality of (ii) can also be attributed to a focus intervention effect under the proposed approach. This account is however not conclusive since it does not sound like the copula desu in (i) is semantically a focus element. I would like to leave this issue for future research. In (20), the overt Q-particle itself types the sentence as interrogative at ForceP. So far, I have argued that a yes/no question with an overt Q-particle such as ka and kke is licensed as interrogative at ForceP, whereas one without it is at the highest position in the SA projection. In other words, there are two positions where a yes/no-question is licensed. I will now show that Japanese has another Q-particle which appears in the topmost SA head, that is, the tagquestion particle ne. This is consistent with the two-layered approach proposed in this paper. A sentence ending with the particle ne induces a tag-question interpretation, as shown in (21a), 8 Interestingly, in-situ focus constructions without a Q-particle can form a yes/no-question with the past form of the copulas datta or desita, as shown in (i).
Hanako-ga keeki-o tukuru n *da / ok da-tta? Hanako-TOP cake-ACC make Fin COP/ COP-PAST 'Is it Hanako that will bake a cake?' b. Hanako-wa keeki-o tukuru n *desu / ok desi-ta? Hanako-TOP cake-ACC make Fin COP.POLITE/ COP.POLITE-PAST 'Is it Hanako that will bake a cake?' It seems that the copulas that contain a past-morpheme do not behave as a Focus head and thus they do not induce a focus intervention effect in (i). In fact, unlike da and desu (cf. 18), these copulas cannot be used to emphasize a constituent.
(ii) *[Gosai-no otokonoko-ga]-datta/desita yo, Syopan-no Gensousokkyoukyoku-o five-year-GEN boy-NOM-COP.PAST/COP.POLITE.PAST Report Chopin-GEN Fantasie.Impromptu-ACC migotoni hiki-masi-ta. beautifully play-POLITE-PAST Moreover, unlike desu (cf. f.n.7), they cannot appear with an adjective that contains a tense-morpheme, as shown in (iii). This suggests that datta and desita behave as a tense-marked copula like English is and thus they cause the ungrammaticality of (iii) due to the doubling of tenses.
whereas it cannot be attached to a wh-question, as in (21b). (21b) thus suggests that ne is a tagquestion particle by nature.
(21) a. Hanako-wa ki-ta ne? Hanako-TOP come-PAST TAG 'Hanako came, didn't she?' b. * Dare-ga ki-ta ne? who-NOM come-PAST TAG This particle can be analyzed as a SA element given that it cannot be embedded, as shown in (22).
(22) * [ Hanako-wa ki-ta ne] kii-ta. Hanako-TOP come-PAST TAG ask-PAST I argue that this particle also carries a [Q]-feature, like ka and kke, as well as a feature that induces a tag-question interpretation. Recall that an in-situ focus construction cannot be licensed as interrogative without a Q-particle or wh-phrase (cf. 8-11). The acceptability of (23) as a question thus shows that it is the particle ne that types these sentences as a question.
(23) a. Hanako-wa ki-ta n da ne? Hanako-TOP come-PAST Fin Foc TAG 'Hanako came, didn't she?' b. Hanako-wa ki-ta n desu ne? Hanako-TOP come-PAST Fin Foc TAG 'Hanako came, didn't she?' In addition, importantly, the Q-particle ne occupies the highest position in the SA projection among other discourse-oriented particles. (24) shows that it must follow other discourse particles such as wa and yo (Saito and Haraguchi 2012).
(24) a. Hanako-wa ki-ta {wa ne / *ne wa} b. Hanako-wa ki-ta {yo ne / *ne yo} c. Hanako-wa ki-ta {wa yo ne / *ne wa yo / *wa ne yo} The above data confirms that ne is a Q-particle that appears in the topmost SA and induces a tagquestion interpretation. This is consistent with the idea that a yes/no question can be typed as interrogative at the topmost SA. This point will be further discussed in Section 4.1 in terms of the typology of Q-particles.

WH-QUESTIONS WITHOUT A Q-PARTICLE.
In Section 2, we have seen that unlike yes/no-questions, matrix wh-questions do not require an overt Q-particle. Matrix wh-questions are licensed as interrogative without a Q-particle, as shown in the a-examples above, whereas matrix yes/no-questions are not, as in the b-and c-examples. Based on this data, one could pursue the idea that there is a null Q-particle only in matrix wh-questions assuming that a Q-particle for wh-questions is different from one for yes/no-questions. This idea is difficult to deny with empirical data, but I will not pursue this line of research because it is not desirable from the perspective of acquisition. Children have to figure out from the input that (i) a Q-particle for wh-questions is different from one for yes/no-questions although the two morphemes share the same phonological shape such as ka and kke, and (ii) the former can be null but not the latter, which would require negative evidence. Rather, I assume that Q-particles in whquestions and yes/no-questions are the same as they look so (Uegaki 2018), and pursue a licensing mechanism of matrix wh-questions without positing a null Q-particle.
Recall that an overt Q-particle or a wh-phrase is necessary to form a question under the discourse particle yo or in in-situ focus constructions (see [25][26][27]. This means that a wh-phrase behaves similarly to an overt Q-particle in licensing a sentence as interrogative. I thus propose that matrix wh-questions are typed as interrogative via Agree with a wh-phrase. The licensing mechanism I propose for matrix wh-questions is the following: (28) wh-questions without a Q-particle: a. There are three assumptions I adopt in (28). First, I assume that a wh-phrase has a [Q]-feature (Chomsky 2000, 128;cf. Cable 2010;Chomsky 2013;Narita 2011). In addition, following Watanabe (1992Watanabe ( , 2003 and Reinhart (1998), I suppose that a wh-phrase is decomposed into a wh-operator and a core noun, represented as NP in (28), and the wh-operator undergoes movement to at least SpecFocP, as in (28a). Furthermore, I assume that it is Force that obtains a [Q]feature via Agree in wh-questions, as in (28b). 9 This proposal accounts for (25a), where a wh-question is formed under the scope of the discourse particle yo. The structure of (25a) is given below:

(29) [SAP [ForceP [FocP wh-Op[Q]i [FinP [DP ti NP] come-T no] ] Force[Q]] yo].
In (29), the empty Force head undergoes Agree with the wh-operator to obtain a [Q]-feature, and thus the wh-question is licensed and interpreted under the scope of the SA element yo.
In addition, the proposed approach accounts for the grammaticality of (26a) and (27a) as well. In wh-questions, a wh-operator first undergoes movement to at least SpecFocP. Then, the Agree relation between the wh-operator and Force is not blocked by the Focus head da or desu in wh-in-situ focus constructions because it does not structurally intervene between them, as illustrated in (30).
Therefore, wh-in-situ focus constructions circumvents a focus intervention effect observed in yes/no-question counterparts (cf. 19).

Implications.
I have claimed that some matrix questions are licensed at ForceP and others at SAP. Table 1 summarizes my proposal regarding where matrix questions in Japanese are licensed.

ForceP
SAP yes/no-questions with ka/kke yes/no with ne yes/no without a Q-particle wh-questions with ka/kke without a Q-particle Table 1. the position where matrix questions are licensed As shown in Table 1, I am particularly proposing two licensing positions for matrix yes/no-questions. In this section, I discuss two implications of this proposal.
4.1. TYPOLOGY OF YES/NO-QUESTION PARTICLES. The proposed two-layered approach to matrix yes/no-questions are consistent with the typology of Q-particles. Bailey (2013) observes that Qparticles for yes/no-questions can be divided into two types cross-linguistically. Q-particles in some languages can be embedded like ka, whereas those in other languages cannot, like ne. For instance, the Q-particle ma in Mandarin cannot appear in embedded questions, as shown in (31) and Yoshida 1996). In this alternative analysis, that the Force head obtains a [Q]-feature is just a byproduct of whoperator movement, which may motivate the reason why wh-questions are typed as interrogative at ForceP, not SAP.
This division of Q-particles can be captured under the proposed approach. Q-particles that can be embedded like ka are a Force head, whereas those that cannot such as Japanese ne and Mandarin ma are a SA head. Interestingly, (32) is not grammatical even with rising intonation, which is commonly believed to license matrix yes/no-questions. In other words, rising intonation does not assign a [Q]-feature to a sentence. Rather, it is just a phonological realization of [Q] in matrix yes/no-questions. If so, the question that immediately arises is why rising intonation occurs at the sentence final position not only in head-final languages but also in head-initial languages like English. Greenberg (1963) names this property Universal 8.

NATURE OF INTONATION IN
(33) Universal 8. When a yes-no question is differentiated from the corresponding assertion by an intonational pattern, the distinctive intonational features of each of these patterns are reckoned from the end of the sentence rather than from the beginning. (Greenberg 1963, 80) In this subsection, I argue that the idea that matrix yes/no-questions can be licensed in the SA projection is needed in order to account for Universal 8. In particular, we need to assume that (i) SAPs are always head-final and (ii) matrix yes/no-questions always mark [Q] in the head-final SA projection whether or not they have a Q-particle. These assumptions ensure that a [Q]-feature in matrix yes/no-questions always appears in the head-final SA head (i.e. the end of the sentence), and question intonation is a phonological realization of that [Q]-feature. The first assumption is not so strange given that a Q-particle appears in the sentence-final position even in head-initial languages, as in Mandarin. Recall that the Q-particle ma in Mandarin is a SA element because it cannot be embedded (cf. 31b). Interestingly, this SA element appears in the sentence-final position although Mandarin is a head-initial language (cf. 31a). This is consistent with the idea that SA heads are universally head-final even in head-initial languages like Mandarin. 10 In fact, Bailey (2013) observes that a sentence-final Q-particle in general cannot be embedded in head-initial languages such as Mandarin, Vietnamese, Ewe, Kayah Li,Langi,Miya,and Nung. 11 This word-order mismatch across languages supports the idea that SAPs are universally head-final. 10 Here, I provide a simple assumption that SAPs are universally head-final. Alternatively, we can also take Haegeman and Hill's (2013) view of SAPs in head-initial languages. They argue that the whole clause embedded under the SA head undergoes movement to SpecSAP under the assumption that SAPs are head-initial in head-initial languages. This movement approach also derives head-"final" SAPs. 11 It should be noted that a sentence-final Q-particle in head-initial languages such as ma in Mandarin should be reanalyzed as a disjunction according to Bailey (2013). The motivation of this reanalysis is mainly to avoid the Final-Over-Final Constraint (FOFC; Biberauer, Holmberg, and Roberts 2014), as formulated in (i).
The second assumption is different from what I have assumed in the previous sections, that is, matrix yes/no-questions with ka/kke are licensed as interrogative at ForceP. However, they are not incompatible. It is possible to consider that a [Q]-feature is marked at both ForceP and SAP, as in (34) In (34a), ka/kke appears in the Force head, and thereby ForceP is typed as interrogative. In (34b), a SAP is introduced above the interrogative ForceP, and the SA head receives a [Q]-feature from the interrogative Force head. This feature sharing between Force and SA can be considered as a result of either selection or Agree. This revised analysis does not harm the accounts for the data in the previous sections. The existence of the SA projection in matrix yes/no-questions might be supported by the distribution of the Q-particle ka. Miyagawa (1987Miyagawa ( , 2017 argues that the Q-particle ka must be selected by a head. According to him, the contrast in (35) is attributed to this selection requirement on ka. Based on Stowell (1981) and Fukui (1986), he assumes that bridge verbs like say selects an interrogative clause, whereas non-bridge verbs such as whisper, murmur, and shout does not.
(35) a. Bill-wa [CP dare-ga kuru ka] itta. Bill-TOP who-NOM come Q said 'Bill said who will come.' b. ?*Bill-wa [ CP dare-ga kuru ka] donatta. Bill-TOP who-NOM come Q shouted 'Bill shouted who will come.' (Miyagawa 2017: 27) Recall that in matrix wh-questions, ka can appear when it follows a politeness-marked verb, whereas it cannot without a politeness-marked verb, as shown in (36). Miyagawa argues that ka can appear in (36a) because the politeness affix -masi requires the SA projection, and ka is selected by that SA head. On the other hand, as in (36b), ka cannot appear without a politeness affix.
(36) a. Dare-ga ki-masi-ta ka? who-NOM come-POLITE-PAST Q 'Who came?' b. * Dare-ga ki-ta (no) ka? who-NOM come-PAST Fin Q 'Who came?' (i) A head-final phrase αP cannot dominate a head-initial phrase βP, where α and β are heads in the same extended projection. (Biberauer et al. 2014, 171) Note that if a sentence-final particle in head-initial languages constitutes the C head of CP and selects a head-initial domain (e.g. TP), then it violates this constraint. Bailey then argues that questions with that particle should be analyzed as disjunctive sentences where the second disjunction is elided. She then assumes that such questions are licensed as interrogative by intonation, which I argue against here. In this paper, I assume that in contrast to Bailey, the particle in question is a Q-particle, which occupies the SA head. In fact, this analysis is not inconsistent with the FOFC given that the SA projection is a different extended projection from the C domain (Sheehan et al. 2017).