2.1. C-system: Force particles

In the lowest C-system of Korean, five Sentence-Final Particles (SFPs) that encode illocutionary force are generated: declarative, interrogative, imperative, promissive and exhortative. Pak (Reference Pak2006) labels the latter three instances as ‘jussives’; the term represents the grammatical mood of commanding or exhorting. Examples in (3) are retrieved from Park (Reference Park2009). Note that Korean is an SOV language, and discourse particles are traditionally analyzed as bound morphemes attached to the verbal cluster.

In (3a–c), although the subjects of these sentences are not overtly manifested due to the pro-drop property of Korean, the specific interpretations of the subjects with respect to each jussive particle naturally arise. Pak et al. (Reference Pak, Portner, Zanuttini, Elfner and Walkow2008) argue that each jussive particle carries a relevant person feature, establishing an Agree relation with the null subject. Namely, the first-person feature, denoting the speaker (i.e. [person: 1]), lies in the promissive marker -ma , while the second-person feature, referring to the addressee (i.e. [person: 2]), is associated with the imperative marker -la . The exhortative marker -ca bears both the first and the second person features (i.e. [person: 1] ⨁ [person: 2]), rendering the subject to denote both the speaker and the addressee. The agreement between jussives and the subjects yields a proper semantic interpretation of the subjects. As for the syntactic position of jussive particles, we follow the point of view that all the illocutionary force particles are base-generated under the Force head (see Cinque Reference Cinque1999, Han & Lee Reference Han and Lee2007, Pan Reference Pan2019, Reference Pan2022, among many others), aligning with the general assumption of the Split-CP hypothesis. As all the clause-typing particles are base-generated under the same head, they are expected to show complementary distributions. For instance, the co-occurrence of jussive particles and the declarative particle -ta is generally banned, as shown in (4).

2.2. Commitment domain

Above the C-system, there is a Commitment domain. CommitP takes a C-system as its argument by adding the speaker’s epistemic stance of the truth of a proposition (see Krifka Reference Krifka, Roeper and Speas2014, Miyagawa Reference Miyagawa2022). From the perspective of pragmatics, given that the speaker publicly manifests that the proposition is true, the speaker’s liability is naturally enhanced. As an instantiation, Miyagawa (Reference Miyagawa2022) posits the Japanese particle -yo under the Commit head. Consider the Japanese examples in (5a–b) in the assertive/imperative context. The particle -yo in (5b) emphasizes the truth of the proposition, and hence, (5b) carries a stronger illocutionary force than (5a), which lacks such a particle.

(6a, b) show that in Japanese, the particle -yo must syntactically occur above the interrogative Force particle -ka , which suggests that -yo is base-generated above ForceP (C-system) (see Saito Reference Saito and Shlonsky2015). Semantically, since -yo enhances the commitment to the truth of a proposition, it is thus expected that the sentence does not exhibit a true information-seeking question. This prediction is borne out. In (6b), -yo selects ForceP headed by the interrogative particle -ka , and the morphological sequence [ -ka-yo ] gives rise to a rhetorical question reading rather than a true information-seeking question reading (see Miyagawa Reference Miyagawa2022, Miyagawa & Hill Reference Miyagawa and Hill2025).

In fact, there are different types of commitment toward a proposition, one of which is evidentiality. Given that evidentiality particles either state a mere presence of the evidence or elucidate specific sources of the evidence, the speaker’s commitment to the proposition is reinforced by evidentiality. Importantly, evidentiality renders the truth condition intact, suggesting that the evidentiality particles should not belong to the C-system (Force). Miyagawa and Hill (Reference Miyagawa and Hill2025) confirm that the sentence in (5b) can be used in the context involving evidentiality.

Following this line of reasoning, we argue that at least three classes of particles in Korean can be posited in the Commitment domain: committal particle -ci , evidentiality particles -kwun , -ney , and hearsay particles -mye , -y . All these particles exhibit the speaker’s commitment to the given proposition by anchoring the proposition to some epistemic sources.

2.3. Speech act domain

As the highest projections in the treetop, the Speech act phrases, saP and SAP, are posited, hosting the speaker and the hearer at their specifiers respectively (see Speas & Tenny Reference Speas, Tenny and Sciullo2003, Miyagawa Reference Miyagawa, Aelbrecht, Haegeman and Nye2012, Reference Miyagawa2022, Haegeman & Hill Reference Haegeman, Hill, Folli, Truswell and Sevdali2013, among many others). This superordinate shell-structure [saP > SAP] hosts various speech act particles in relation to discourse participants. Take the speech act particle hai in Romanian as an example. The pragmatic function of hai is to seek the attention of the addressee as an evaluative injunction. Haegeman and Hill (Reference Haegeman, Hill, Folli, Truswell and Sevdali2013) propose that the particle hai occupies the head position of SAP (i.e. the lower shell) with the vocative Dane at its specifier, as in (16a). In this configuration, Dane is the hearer, and the sentence serves to draw the attention of the addressee. In (16b), when the particle hai moves from the lower SA head to the higher sa head, deriving the ‘ hai > Dane’ order, the same particle hai will have a speaker-oriented reading, such as enhanced mitigation, as opposed to (16a).

Turn to Korean. The particle -yo expresses the speaker’s polite attitude toward the addressee, resulting from various factors such as social hierarchy. For instance, in the context in which the speaker is delivering the utterance to her teacher, omitting the particle -yo yields pragmatic infelicity, as in (17).

Based on the observation of the speech act particles in Romanian and West Flemish, Haegeman and Hill (Reference Haegeman, Hill, Folli, Truswell and Sevdali2013) propose a list of the properties of speech act particles. Choi (Reference Choi2016) argues that -yo shares some of these properties, as sketched in (18).

1. (18)

   1. a. -yo indicates the relationship between the discourse participants.

   2. b. -yo outscopes sentential negation.

   3. c. -yo interacts with the vocative in terms of the distribution.

   4. d. The interaction between -yo and the vocative is conditioned by (18a).

(18a, c, d) clearly show that the particle -yo is associated with the vocative/addressee DP. Along this line, the infelicity of the sentence in (19) is due to the pragmatic clash between the vocative particle -ya and the speech act particle -yo . Namely, when the addressee Inho is overtly realized, suffixed by the vocative particle -ya , it expresses the speaker’s social superiority to the addressee. However, the speech act particle -yo at the end of the sentence indicates the contrary, yielding an infelicitous sentence.

Choi (Reference Choi2016) argues that -yo is base-generated as the SA (Speech Act) head, the lower shell of the speech act projections, while the addressee/vocative DP is externally merged at the Spec of SAP (see also Yim Reference Yim2016 for further discussion on -yo ). Importantly, the order [e -yo ] in (19) suggests that the speech act particle -yo is located higher than the declarative force particle -e , substantiating that SAP is generated higher than ForceP. In addition, data from language acquisition suggest that -yo is posited higher than the exhortative particle -ca , one of the instances of jussive particles, as shown in (20) (see Yim Reference Yim2012). Given that jussive particles are generated under the Force head, (20) also shows that SAP is generated higher than ForceP.

1. (20) Mek-ca-yo.

   eat-exh-sa

   ‘Let’s eat.’

Furthermore, the SA particle -yo is located even higher than the particles in the CommitP domain, as shown in morphological sequences in (21). This demonstrates the hierarchical order of the three functional projections as: ForceP < CommitP/EvidP/RepP < SAP.Footnote ¹⁰

In conclusion, a full-fledged hierarchical structure of the Korean discourse projections examined in our paper is represented in (22).

(22)

2.4. Embeddability

In Korean, all the clause-typing particles under ForceP, which are positioned lower than the complementizer -ko , can be embedded, as shown in (23).Footnote ¹¹

As for the particles in CommitP and SAP, examples in (24) show that none of these particles are embeddable. This is expected as CommitP and SAP are discourse layers, which are closely related to the speech act of the utterance and the perspective of the speaker. A similar observation has been made for the Attitude particles in Mandarin Chinese, which also cannot be embedded. According to the Subjectivity Scale Constraint in Pan (Reference Pan2015, Reference Pan2019, Reference Pan2022), higher functional projections are more directly linked to the speaker’s opinion or attitude, and give rise to more subjective interpretations; therefore, it is less likely for these projections to be embedded. As a result, they are root phenomena.

2.5. Phasehood

This section examines the phasehood of each particle. As illocutionary force must be morpho-syntactically marked in Korean, the omission of the Force particle - ta results in ungrammaticality, as shown in (25). Importantly, sentences with Force particles can be used independently, which shows that the Force head in Korean exhibits isolability/independence at both the A-P interface and C-I interface, thus demonstrating a qualifying property for phase. This also aligns with Chomsky’s (Reference Chomsky, Martin, Michaels and Uriagereka2000) argument supporting CP as a phase in that CP includes Force.

Furthermore, empirical evidence based on long-distance Negative Polarity Item (NPI) licensing in Korean suggests that Force head invokes Spell-Out and Transfer of the structure to the interfaces.Footnote ¹³ In Korean, a polarity item such as amukesto (i.e. ‘anything’) must form a local relationship with its licensor to ensure a proper semantic interpretation, which is traditionally referred to as ‘clausemate condition’ (see Choe Reference Choe1988, Kuno Reference Kuno and King1998, Sells Reference Sells2006). In the Minimalist Program, this locality condition can be reformulated in terms of the notion of phases as the following: the polarity item and its licensor, negation, must be within the same Spell-Out domain (see Kayabaşi & Özgen Reference Kayabaşi and Özgen2018 for a similar reasoning for Turkish). Consider the following pair in (26). Both sentences instantiate a long-distance dependency of the polarity item as the NPI is located in the object position inside the embedded clause, whereas its licensor is in the matrix clause. The difference is that (26a) has a full-fledged embedded clause, headed by the Force particle -ta , while the embedded clause of (26b) is reduced, headed by the nominalizer -ki .

The contrast between (26a) and (26b) can be explained if the Force head -ta is treated as a phasal head. Adopt the weak version of the Phase Impenetrability Condition (see Chomsky Reference Chomsky and Kenstowicz2001).Footnote ¹⁴ In (26a), inside the embedded clause, by the point at which the phase head -ta is merged, the object NPI will be spelled out as part of the VP, which is the complement of the lower phasal head v. As a result, the negative licensor and the NPI end up being in different spell-out domains. By contrast, in (26b), as there is no Force phase head in the embedded clause, the transfer of the VP (including the object NPI) inside the embedded clause can be delayed until the phasal head v of the matrix clause is merged. Consequently, the negative licensor in the matrix clause and the object NPI in the embedded clause are in the same spell-out domain, and therefore, the phasemate condition is satisfied. The long-distance polarity licensing thus provides positive evidence showing that Force head invokes cyclic computation internal to syntax and that Force is a phase head.

Distinct from ForceP, CommitP and SAP are not qualified as phases. Start with hearsay particles in the Commitment domain. As shown in (27), hearsay particles under Rep head cannot directly take TP as their complement without the presence of the declarative Force particle -ta. Along this line, a Rep head itself does not show a true independence property at interface levels, and therefore, it cannot be qualified as a phase head.

However, other particles in the CommitP layer, such as committal -ci and evidential -kwun , -ney do not appear to require Force head to stand alone, as shown in (28).

There are two structural possibilities in (28). One is that ForceP is absent altogether, so the Commit head directly takes TP as its complement. The other is that ForceP is syntactically projected but with a phonologically null exponent. We argue for the latter option, substantiated by both empirical and theoretical evidence. One crucial piece of evidence is that committal particles and evidential particles can co-occur with the Force particle -ta , demonstrated by the sequences: [-ta-ci ], [-ta-ney ] and [-ta-nun-kwun ]. This substantiates two points. First, commitment particles, in principle, take ForceP as their complement, and second, -ci , -kwun and -ney cannot function as clause-typers. If these particles can incorporate the clause-typing functions, their co-occurrence with the declarative particle, -ta , will be left unexplained. In addition, recall that the Commit layer does not take part in forming a proposition of an utterance. Then, if Commit head directly takes a TP as complement without ForceP, this will result in an incomplete proposition, as the C-system is absent in this structure. For this reason, we assume that a null ForceP is still projected in (28). Along this line, akin to hearsay particles -mye and -y , the independence/isolability property of -ci , -kwun and -ney actually comes from ForceP, and therefore, the commitment particles do not stand alone on their own, failing to be qualified as phases.

Finally, (29) shows that the SA particle -yo cannot be used independently without the co-occurrence with the declarative force particle -e, and therefore, SAP does not show an independence property at interface levels and is not treated as a phase under our analysis.

2.6. Summary

Discourse particles in Korean, ranging from force particles through commitment particles to speech act particles, head different functional projections. Based on the observation on the co-occurrence of these particles, we establish their hierarchical order as: ForceP < CommitmentP/EvidentialityP/ReportativeP < SpeechActP. Furthermore, we conclude that only ForceP is a phase, whereas CommitP and SAP are not. One important point is that RepP, which is in the commitment domain, is not a phase in our analysis but nevertheless has ϕ-features. This shows that non-phasal heads could possess uninterpretable features at narrow syntax, see Table 1.

Table 1.

Syntactic positions of the discourse particles and their phasehood

Table 1. Long description

The table consists of six columns: Domain, Sub-domain, Particles, phi-features, Embeddability, and Phasehood.

* The Speech Act domain includes the particle -yo. It is marked with an X for phi-features, Embeddability, and Phasehood.

* The Commitment domain is divided into three sub-domains:

* Commit P includes the particle -ci, marked with an X for all three properties.

* Evid P includes particles -kwun and -ney, marked with an X for all three properties.

* Rep P includes particles -mye and -y. It is marked with a checkmark for phi-features but an X for Embeddability and Phasehood.

* The Force domain is divided into two groups:

* Particles -ta and -nya are marked with an X for phi-features but checkmarks for Embeddability and Phasehood.

* Particles -la, -ma, and -ca are marked with checkmarks for all three properties: phi-features, Embeddability, and Phasehood.

3.1. Head-movement as a syntactic operation in Korean

Voluminous work substantiates that head-movement is difficult to implement in the syntax proper, as it violates many well-motivated syntactic constraints. To list a few, first, it violates the Extension Condition (see Chomsky Reference Chomsky, Hale and Keyser1993), which states that syntactic movement always yields a bigger structure, blocking the potential anti-cyclic movement in syntax. Expectedly, this poses a serious problem for head-to-head adjunction, as the adjunction from the lower head to the upper one does not target the treetop, and hence, does not contribute to the growth of the tree. This can be evidenced by V-to-v, v-to-T and T-to-C movements in Korean. In addition, head-adjunction does not systematically have any semantic effects, unlike A or A'-movement (see Chomsky Reference Chomsky and Kenstowicz2001). Therefore, head-movement operation is assumed to fall under a different part of grammar: either as PF-movement or as a post-syntactic operation (i.e. morphological merger; see Matushansky Reference Matushansky2006, Vicente Reference Vicente2007, Harizanov & Gribanova Reference Harizanov and Gribanova2019, among many others).Footnote ¹⁵ Nonetheless, in Korean, there is empirical evidence to support that head-movement takes place in the narrow syntax (see Lee Reference Lee2008 for four pieces of evidence based on the dislocation of a VP/vP-constituent in Korean).Footnote ¹⁶ Examine the sentence in (30) (examples retrieved from Lee Reference Lee2008).

In (30), the adverbial phrase in Jongro can only modify the event of the main clause (i.e. the event of John’s saying something), rather than the event of the embedded clause (i.e. the event of Mary’s buying a book), which suggests that in Jongro is base-generated in the main clause. This naturally bars the possibility that in Jongro is first base-generated in the embedded clause and then undergoes movement to the sentence-initial position. However, in (31), both the adverbial phrase in Jongro and the object of the embedded clause book occur in the sentence-initial position preceding the subject of the matrix clause John, and in Jongro is immediately followed by book, and in this case, in Jongro, unambiguously modifies the event of the embedded clause ‘Mary bought that book’.

Recall that the adverbial phrase in Jongro itself cannot be fronted alone. As a result, the possibility of scrambling the embedded object that book and the adverbial in Jongro separately is barred. Given the interpretation of the sentence, in Jongro can only modify the embedded VP buying that book, and therefore, in Jongro and that book must stay inside the same embedded VP. Naturally, a plausible analysis would allow the entire embedded VP as a chunk to undergo fronting to the sentence-initial position. Note that VP-fronting is generally assumed to take place at narrow syntax, targeting an A'-position. If head-movement is truly a post-syntactic operation, the embedded verb sa ‘buy’ is expected to be fronted together with the adverbial and the object during the VP-fronting procedure in syntax. Then, to achieve the word order, the embedded verb sa ‘buy’, now in the fronted VP located at the Spec of the matrix CP, will be somehow ‘lowered’ to its original position, and then be raised to the embedded T and C. It is hard to imagine that such a procedure is possible even under the scenario of post-syntactic operations. Therefore, a more reasonable derivation would allow head-movement to take place before the operations in the syntax proper, such as VP-fronting, and along this line, head-movement can be regarded as a syntactic operation. (32) shows the derivation of (31). The verb of the embedded clause V-buy will be first moved to join T and then to Force-ta and finally to join the complementizer -ko to form a complex head sa.ss-ta-ko . At a later stage, the VP, now only consisting of the adverbial in Jongro and the object book will be fronted together to the Spec of CP.

Given that VP-fronting in Korean is a syntactic operation, head-movement, which takes place prior to the fronting, may well be a syntactic operation (see Lee Reference Lee2008, for a further discussion). Therefore, considering the logical ordering with respect to other syntactic operations, we propose that head-movement in Korean is a narrow syntactic operation. Importantly, this piece of evidence seems to suggest that head-movement can take place before all the syntactic operations in a general fashion, since in the Minimalist Program, all the syntactic operations such as Agree (feature valuation and feature deletion), Move and Transfer, happen at the phase level simultaneously. We will come back to this point in later sections.

3.2. Edge-feature in Korean

One of the arguments supporting the phasehood of discourse particles in Mandarin is that they appear to bear an EF, a feature that allows an A'-type of specifier to be projected (see Pan Reference Pan2022). In Chomsky’s system, EPP and EF are strictly distinguished. The phase head v* bears both an EPP and an EF: EPP is satisfied by the merging of an external argument (EA), and EF on v* allows for the creation of an additional landing site, which serves as an escape hatch for cyclic wh-movement. The highest phase head C also bears an EPP and an EF: EPP can be inherited by T to project an A-specifier hosting EA, and EF serves as the final landing site for wh-movement. In a simplified way, it seems that EPP allows a phase head to project an A-type of specifier (i.e. to create an argument structure), whereas EF projects an A'-type of specifier (i.e. to establish a scopal or discourse-related dependency). Korean is a head-final language (SOV), and discourse particles are suffixed to the verbal cluster; therefore, it is not clear whether Korean discourse particles project a specifier. Nevertheless, following the general idea that phase heads possess EPP/EF (see Chomsky Reference Chomsky, Martin, Michaels and Uriagereka2000, Reference Chomsky and Kenstowicz2001, Reference Chomsky, Freidin, Otero and Zubizarreta2008), and that the v* phase projects an EA by means of EPP, let us assume that the diagnosed phase head (i.e. Force) in Korean also bears an EF.

Korean is a consistent head-final language. It has been argued in Korean syntax that the subject is not surfaced at Spec TP but stays at Spec v*P (see J.-M. Yoon Reference Yoon1990, J. Yoon Reference Yoon1994 for empirical evidence and the relevant syntactic tests).Footnote ¹⁷ If we take the surface head-finality as pure externalization effects, then nothing special is needed to justify the consistent final-order of Korean, as it could be regulated by the Headedness Parameter. However, alternatively, if we assume that the externalized final order could result from syntactic operations in syntax for purely syntactic reasons, the derivation could proceed as follows in (33) (see Kayne Reference Kayne1994, Lee Reference Lee2007).Footnote ¹⁸ Below is the sample derivation.

Recall that in the previous section, we argued that head-movement takes place prior to other syntactic operations. Thus, at the lower phase level (i.e. v*P), in Step 1, V is moved to join v* via head-movement, v* agrees with IA, deleting the u[ϕ] on v* and uCase on IA, and EA is merged to satisfy the EPP on v*. Here, we adopt the weaker version of the Phase Impenetrability Condition (PIC₂), according to which, there is no transfer at the v*P-level at this stage. Continuing with the higher CP phase level, in Steps 2–3, T and C are merged. Then, V-v* will be moved to join T, and V-v*-T will be moved to join C, to finally form V-v*-T-C. T agrees with EA, deleting the u[ϕ] on T and uCase on EA. In Korean, EA does not raise to the Spec of T. Given PIC₂, the complement of the lower phase head (v*), i.e. VP, will be transferred (see Step 4). Suppose CP here is ForceP. As discussed in Section 2, ForceP is a phase in Korean, and hence, bears an EF. The empirical observation is that in Korean, there is no morphological reflex of valuing uninterpretable features on the Force head, and therefore, Force does not probe any particular goal.Footnote ¹⁹ As a result, adopting the algorithm suggested by Pan (Reference Pan2022), we contend that the EF is satisfied by the comp-to-spec movement. Along this line, in Step 5, the entire complement of C, TP, is raised to the Spec of C to satisfy the EF. At last, the entire structure is transferred. As a result, the externalized final order will be ‘EA > IA > V-v*-T-C’. Note that the head-movement of V-v* to T to C happens before the comp-to-spec movement to satisfy the EF on C. Recall that in Section 2, we argue that CommitP and SAP are not phases. However, as will be detailed in Section 4, both the Commit head and SA head can eventually get an EF from the lower phasal head Force as a consequence of the head-movement of Force to join Commit and then SA.

5.1. In comparison to Chinese

A common property of discourse particles in Korean and in Chinese is that in both languages, the co-occurrence of the discourse particles from different layers is possible, as shown in (39): [ le éryǐ a ] for Mandarin and [ la-y-yo ] for Korean. The order among the three particles in each language is strict.

As for the phasehood of the particles in Chinese, Pan (Reference Pan2022) argues that functional projections split from CP hosting SFPs are all phases in that SFPs exhibit an independence/isolability property at interface levels. Along this line, all of the three particles, le , éryǐ , and a, are all phasal heads as they can be used independently to form fully grammatical sentences. A sentence with each specific particle is not only phonologically isolable but also yields a new adequate semantic unit with a specific semantic interpretation or discourse function. Pan (Reference Pan2022) further assumes that all the SFPs in Chinese possess an EF, as they are all phasal heads. In particular, EFs for the projections such as S.AspP, OnlyP, ForceP, and AttP are satisfied by a comp-to-spec movement as a last resort. The complement of each particle will be raised systematically to its specifier to satisfy the EF, resulting in a systematic cyclic roll-up style of movement. Importantly, discourse particles in Chinese are not bound morphemes, which do not have to be attached to verbal elements to form a verbal cluster. Turning back to Korean, we argue that only Force head projects a phase, and hence, EF is restrictively endowed on the Force head only. In addition, we show that the syntactic head-movement takes place from the Force head to the Commit head and then to the SA head, which renders the EF to be relocated to the topmost projection. Importantly, this head-movement takes place prior to any other syntactic operations, so EF is pied-piped up to the topmost projection in the peripheral domain before being satisfied, which leads to a crucial difference between Korean and Chinese. While there are cyclic roll-up style movements (i.e. multiple comp-to-spec movements) in Chinese, the comp-to-spec movement only takes place once at the highest functional projection in Korean. As a result, although the surface final order for the discourse particles in the two languages in (39) appears to be similar, the derivational mechanisms are different for Korean and Chinese.

5.2. Phasehood and the highest phrase of an extended projection

Two diverging views exist on phases. In Chomsky’s (Reference Chomsky, Martin, Michaels and Uriagereka2000, Reference Chomsky and Kenstowicz2001) rigid model of phase, specific categories (C, v*) are phase heads, endowed with uninterpretable features as well as EPP/EF. Merging a phase head into a structure signals the timing for syntactic operations, such as Agree (feature valuation, feature deletion, etc.), Move and Transfer. Syntax can tell whether a lexical item is a phase head or not only by inspecting the value of its features: uninterpretable features are unvalued, thus lacking the value slot: u[F:__]. Uninterpretable features can be inherited by a non-phasal head. This system ensures computational efficiency. Alternatively, Bošković (Reference Bošković2014) argues for a contextual model of phases, in which the highest phrase of an extended projection is a phase. Thus, phasehood is dynamically determined by context. It is worthwhile noting that both views lead to the same result, which is that the highest phrase in an extended projection is always a phase bearing an EPP/EF. For instance, in Chomsky’s model, a phase head is always the last functional head in a given sub-array to be merged; therefore, the projection headed by this phasal head must be the highest phrase in a given extended projection. For Bošković, phase heads are not fixed categories, and phasehood is not associated with fixed categories but is contextually determined. Whatever phrase located in the highest position in an extended projection is a phase. For instance, in the nominal domain, an NP or QP without D can be a phase, as long as it is the highest projection in this given domain. Crucially, Bošković’s diagnosis of phasehood is based on the extraction test: whether a given phrase possesses an EPP/EF to project a specifier serving as an escape hatch for A'-movement.

In our analysis, Korean appears to share the views on both rigid and contextual models of phases. Complying with Chomsky’s rigid view, EF is restrictively endowed on the phasal head, i.e. Force. Importantly, although the highest functional projections in the C domain in Korean, SAP or CommitP, are not phases, they can still get an EPP/EF from the lower phase head, Force, via head-movement, which is, ultimately, in line with Bošković’s contextual model of phase. Importantly, in our analysis, the highest phrase in a syntactic domain always has EPP/EF. Note, however, that there is a crucial distinction between Bošković’s diagnosis of phasehood and our analysis. Bošković’s diagnosis of phasehood relies on the presence of an escape hatch for A'-extraction: if a given projection has EF to project an escape hatch, then this projection is considered a phase. However, in our analysis, the diagnosis of phasehood relies on whether a given functional projection in the CP domain can represent a fully grammatical independent sentence demonstrating interface properties. Therefore, the highest functional projection in the CP domain in Korean is not lexically a phase, but it can still obtain an EF from a lower phasal head via head-movement.

5.3. Language variation

Suppose that a given lexical array contains at least two functional heads, a higher one and a lower one. Each of them could be a phasal head or a non-phasal head. A third condition lies in whether a given language exhibits head-movement. When these three conditions work together, different combinations could show up with very interesting typological patterns, as shown in Table 2. The list only shows some investigated cases and is not exhaustive.

Table 2.

CP domain in different languages

Table 2. Long description

The table consists of six columns and six rows of data categorized by Case.

* Case i: Highest head is Phase C. Lower head is Non-Phase T. Head-movement is present (check mark). Language is English.

* Case i (continued): Highest head is Phase C-Force. Lower head is Non-Phase T. Head-movement is present (check mark). Language is Korean. Example 3.

* Case ii: Highest head is Phase C. Lower head is Non-Phase T. Head-movement is absent (X mark). Language is Chinese. Example 1.

* Case iii: Highest head is Phase C-Attitude. Lower head is Phase C-Only. Head-movement is absent (X mark). Language is Chinese. Example 39 a.

* Case iv: Highest head is Non-phase S A. Lower head is Phase Force. Head-movement is present (check mark). Language is Korean. Examples 17 and 20.

* Case v: Highest head is Non-phase. Lower head is Phase. Head-movement is absent (X mark). Language is unknown (question marks).

In Case (i), the highest functional head in a given syntactic domain is a phasal head, such as C in English, and the lower head is not a phasal head, such as T in English. English exhibits head-movement. The Korean ‘TP < ForceP’ pattern also belongs to this category. The only difference between English and Korean is that in English, C agrees with a wh-phrase in terms of Q-feature, and the wh-phrase is moved to the Spec of C to satisfy its EF, whereas in Korean, the Force head does not agree with any specific goal, and therefore the entire complement of Force will raise to its specifier to satisfy the EF. In a similar scenario, if the language does not have T-to-C movement, this illustrates Case (ii), such as Chinese, among many other languages. Case (iii) illustrates languages in which both the higher projection and the lower projection in a given syntactic domain are phases, and there is no head-movement, such as the ‘OnlyP< AttitudeP’ pattern in Chinese. In languages described in Case (iv), the highest functional head is not a phasal head, but the lower one is a phasal head, and there is head-movement in this language. This illustrates the ‘ForceP < SAP’ pattern in Korean. Case (v) minimally differs from Case (iv) in that the given language does not exhibit head-movement, which requires further investigation.