Kalaallisut is a member of the Inuit-Yupik-Unangan language family, which is spoken from Eastern Russia across Alaska and Canada to Greenland. It is the sole official language of Greenland, with around 50,000 speakers in Greenland and another 17,000 in Denmark (Naja Blytmann Trondhjem p.c.). The language is acquired by children in the home and taught in school. Danmark was a colonial power in Greenland and Danish language still has a big presence, especially in administration. English has a growing presence.

Kalaallisut exhibits classic properties of polysynthethesis: grammatically free argument drop, noun incorporation, high morpheme-to-word ratio, and flexible word order (though see Fortescue 1993). Verbs agree with subject and object and case alignment is ergative-absolutive. These properties are illustrated in (4) through (7).2

In (4) the object kaagi is incorporated (Sadock 1980) and all arguments are pro-dropped (Yuan 2018: 154). (4) further shows that both intransitive and transitive agreement morphology is fused with mood morphology: dependent conditional mood in the embedded clause and independent indicative mood in the matrix clause. Kalaallisut has eight moods, five of which will be relevant in what follows: causative, contemporative, indicative, interrogative and participial. Example (5) illustrates the unmarked SOV order. SOV order is accepted in any context in elicitation, it is the order volunteered in out-of-the blue contexts, and it is the dominant word order in texts (Fortescue 1993). (See also Fortescue 1984: 93ff, and Kleinschmidt 1851/1968: 98–101 for arguments that SOV is the unmarked order.) Together with (6), (5) shows that case alignment is ergative-absolutive. The transitive subject is overtly marked with the ergative suffix -p, whereas transitive objects and intransitive subjects are morphologically unmarked, as is typologically common for absolutive case. There is no morphological ergative/absolutive case distinction in pronouns and plural nouns, but we gloss them with case based on their grammatical function. Finally, the textual example in (7) illustrates high morpheme-to-word ratio for both verbs and nouns.

Complement clauses typically appear after the matrix verb and for the bulk of embedding predicates trigger 3sg object agreement on the matrix verb (Fortescue 1984: 34). Three moods are possible for complement clauses: participial mood, causative mood and contemporative mood. The participial and causative moods are used when the embedded subject is distinct from the matrix subject, as in (8) and (9). Contemporative mood (10) is used when the embedded subject is co-referential with the matrix subject. We focus on participial clauses in what follows.3

Based on the discussion above, we propose the basic clause structure in (11),

Compton (2018) collapses Mood and C, whereas Yuan (2018: 32) assumes that MoodP is dominated by subject and object agreement projections. As far as we can tell, nothing of consequence hinges on this choice. The points we make could be made under either of the alternative structures.

As noted in Fortescue (1984: 38) and Sadock (2003: 32), and discussed at length in Mikkelsen & Thrane (2024), Kalaallisut allows raising to object out of a finite clause, i.e. hyperraising to object:4

Mikkelsen & Thrane (2024) show that the configuration in (13) involves movement of illit from the lower clause into the matrix clause where it triggers object agreement on the matrix verb. They provide evidence from island and locality effects that this is movement, though we won’t repeat their data. Of interest here is that this is movement across a finite clause boundary to a low position:

The finiteness of the embedded clause is evidenced by its obligatory mood-agreement morphology—compare (15) to (16)—and its ability to express tense independently of the tense of the matrix clause (17):

These embedded clauses are therefore at least MoodPs. We will go one step further and assume that they are CPs, based on the fact that wh-movement can take place within such embedded clauses, suggesting the presence of a specifier of CP position (Keine 2020: 57).

The correlation of finiteness with the presence of a CP is in accord with most analyses of hyperraising (e.g. Martins & Nunes 2009; Deal 2017; Fong 2018; Halpert 2019), though Jenks (2025) argues that Moro hyperraising is out of a finite TP. The exact landing site of hyperraising is a little harder to pinpoint, but crucially it is lower than the landing site of focus movement. This is demonstrated in the examples in (19) and (20).5

In the well-formed (19) the hyperraised object meeqqat occurs to the right, i.e. lower than the focus fronted DP. The ill-formedness of (20) shows that hyperraising cannot target a position above a focus fronted phrase. For concreteness we take the landing site of hyperraising to be the specifier of vP, as in (21), though all that matters for assessing the Height-Locality Connection is that it is below the landing site of focus fronting (= specifier of CP).

We have now established the locality profile of Kalaallisut hyperraising, namely that it can cross a finite clause boundary. This is surprising from the perspective of the Height Locality Connection, according to which movement to a low position is more restricted than movement to a high position. This contrast is schematized in Table 3.

Kalaallisut has a focus construction in which the focused element appears in initial position followed by a clitic (=una for singular foci, =uku for plural foci) and the verb is realized in the participial mood:6^,7

This construction has been noted and described in previous literature (Fortescue 1984: 74ff, 184ff; Molich 2022: 100–107), but, as far as we know, no theoretically oriented analysis has been developed. To begin to fill this gap we start by establishing five analytic properties of the construction. First, though homophonous with the absolutive forms of the singular and plural cataphoric demonstrative pronouns, =una and =uku are dedicated focus clitics in (22) and (23). Second, focus fronting is derived by movement. Third, focus fronting is clause bounded. Fourth, =una/=uku are focus Q-particles in the sense of Cable (2010), notated Q_foc. Q_foc takes the focused phrase as their complement and projects a Q_focP. Finally, there is a dedicated focus C head that attracts Q_focP to its specifier. The resulting derivation for (23) is sketched in (24).

The following sections establish each of these properties in turn.

As noted above intervention arises when a closer goal satisfies the probe, causing it to halt without interacting with a more distant goal bearing the probed-for feature(s). In Halpert’s (2019) of hyperraising in Zulu, there is no intervention in hyperraising to subject configurations because the hyperraising probe is only satisfied by a goal that has ϕ-features and the ability to discharge the EPP on T. While Zulu finite CPs have ϕ-features, they cannot serve as subject. Therefore the embedded finite CP in the hyperraising configuration does not meet the combined requirement of the hyperraising probe, lacking the ability to satisfy the EPP. The hyperraising probe is therefore able to probe past the finite CP and interact with and attract the highest DP goal inside the embedded CP. This derives hyperraising. Because English CPs can satisfy the EPP, the embedded CP intervenes between the raising probe and the highest DP inside the embedded CP. This explains the lack of hyperraising in English. Halpert’s account of Zulu hyperraising to subject carries over to Kalaallisut hyperraising to object, though we replace the EPP feature on the probe with a D feature. (This avoids positing an EPP feature and instead takes advantage of preexisting category features.) Thus the Kalaallisut hyperraising probe has the specification in (57), using the plus sign for composite, specifically conjunctive, specification of the satisfaction condition.14).

To assess intervention we need to determine the featural content of the potential goals for the probe, i.e. the embedded clause and the highest nominal inside of it. Specifically, we need to determine whether these phrases bear ϕ and/or D features. We will assume, uncontroversially, that the presence of ϕ-features is reflected in the ability to control ϕ-agreement, while the presence of a D feature is reflected in the ability to bear case. It is immediately clear that Kalaallisut nominal phrases bear both ϕ and D. As seen in (58), both subjects and objects control agreement on the verb and they also bear case. This is morphologically visible on the ergative subject, but not on the absolutive object because absolutive case is null. There are however further, oblique, cases that we can look to, including the allative goal in (59).

Turning to clausal complements, they control 3sg object agreement on the matrix verb, as seen in (60), indicating that finite clauses bear ϕ-features.15

Importantly, clausal complements are unable to bear case, indicating that they are not nominalized by a D-shell.16 This is not immediately discernable because clausal complements typically occur in an absolutive argument position (complement of transitive V) and absolutive case is null. Therefore baseline examples like (60) are compatible with complement clauses bearing absolutive case. There are however two pieces of evidence that clausal complements do not bear case in Kalaallisut. First, they do not participate in the kinds of case alternations that DP complements do. This is illustrated for anti-passivization in (61)–(64).

(61) and (63) are baseline transitive examples, with an ergative-absolutive case frame and transitive agreement. (62) is the antipassive form of (61): the case frame is absolutive-instrumental and the verb shows intransitive agreement. Now consider (64). Again the verb is antipassive, carries intransitive agreement and its complement is marked with instrumental case. The only difference between this ungrammatical construction and the grammatical example in (62) is that (62) involves instrumental case marking of a DP, whereas (64) involves instrumental case marking of a clausal complement. This contrast follows straightforwardly on the assumption that clausal complements are not DPs.

A second observation that supports clausal complements lacking case and therefore lacking a D feature comes from a general ban on having two absolutive DPs within a single finite clause (Compton 2018: 164). This is manifested in the ungrammaticality of (65), where the two internal arguments of a ditransitive are both absolutive. One must be oblique, as seen in (66) and (67).

If clausal complements bear (null) absolutive case we expect it to be ungrammatical to have a second absolutive DP in the clause. We can test this with ditransitive embedding verbs like aperi- ‘ask’. This verb takes an external DP argument (the person asking), a finite complement clause (what is asked) and a DP indirect object (the person being asked). As (68) shows the indirect object surfaces in the null absolutive case controlling object agreement on the matrix verb.

If the complement clause also bears absolutive, this would be an exception to the ban on multiple absolutive DPs in a single finite clause. However, if clausal complements do not bear case, (68) is consistent with the ban on multiple absolutives. We conclude that clausal complements do not bear case and that they lack a D feature.

With this much in place we turn to the derivation of hyperraising. The relevant structure is given in (69).

When v Merges with VP, the hyperraising probe on v probes for ϕ. The first ϕ-goal is the embedded CP. The CP, however, does not satisfy the probe on account of missing a D feature. The hyperraising probe thus probes past the CP node and interacts with the highest DP inside the embedded clause. By virtue of bearing both ϕ and D, the embedded DP satisfies the probe and moves to the specifier of the matrix vP. Being satisfied, the probe halts. We have now derived the existence of hyperraising to object in Kalaallisut. Next we need to explain why cross-clausal focus fronting is impossible in Kalaallisut.

Before we analyze the lack of cross-clausal focus fronting, we revisit our informal movement analysis of intraclausal focus fronting from section 3.2. There are two parts to the analysis: a focused element embedded in a Q_foc shell and a dedicated C head, C_foc. Together these force movement of the Q_focP to the specifier of C_foc, resulting in its observed clause-initial position. Within the interaction-satisfaction framework, this comes down to the probe in (70).

To see this in action, consider the derivation in (72) for the example in (71).

The Q_focP meeqqa-n=uku is Merged as a sister of V. When C_foc is Merged with MoodP, it probes for Q_foc, finds meeqqa-n=uku and moves it to its specifier. Note that, in contrast to the hyperraising probe, the focus probe is “flat”: its satisfaction condition is identical to its interaction condition. In other words, all that is required to focus front is being focused; there is no additional category specification of the satisfaction condition. In section 3.2.4 we motivated this by observing that a wide range of syntactic elements can focus front. This sets focus fronting apart from hyperraising which only applies to DPs, which we modeled with a composite probe. The flatness of the focus probe is key to explaining the absence of cross-clausal focus fronting, a task we take up to next.

In principle, cross-clausal focus movement could take two forms: one-fell-swoop movement from inside the embedded clause to the specifier of a matrix C_foc or successive cyclic movement through the specifier of embedded C heads. As detailed below, neither is available in Kalaallisut.

For one-fell-swoop movement there are two scenarios to consider. The first is a configuration in which the embedded CP is itself focused. In our Q-particle analysis that corresponds to the embedded CP being encased in a Q_foc shell. That Q_focP satisfies the matrix focus probe and therefore intervenes between the Q_foc-probe on matrix C and any Q_focP inside the embedded clause.17 This results in focus fronting of the embedded CP, as in (73) and (74).

The fact that the interaction and satisfaction conditions are identical guarantees that a focused embedded CP will intervene for focus probing: by virtue of being encased in a Q_foc shell it can satisfy the probe, which therefore halts without interaction with any potential goals inside the embedded clause. Here we see the importance of the focus being flat. If there was a hypothetical specialized focus probe [int:Q_foc, sat:Q_foc + D^M], it would be able to bypass the CP and Agree with a DP encased in a Q_foc shell inside the embedded clause. This would trigger cross-clausal movement, on analogy with hyperraising. In other words, on the present analysis, it is the flatness of the focus probe that sets focus fronting apart from hyperraising.18

Next we need to rule out one-fell-swoop movement when the embedded CP is not focused, as in (75).

Building on McCloskey (2002), Abels (2003), and Abels (2012) among many others, we assume that embedded C heads obligatorily carry a ‘spurious’ Q_foc feature (to use McCloskey’s term). As the name suggests, spurious features are divorced from semantics. A genuine Q_foc-feature is hosted by a C head with focus semantics (see below (82) on what that amounts to), whereas a spurious Q_foc-feature is hosted by a C head which lacks focus semantics. Within the interaction-satisfaction framework of Agree we understand spurious features as probe features as opposed to goal features. While a Q_foc probe feature on the embedded C head is not able to satisfy the matrix Q_foc probe, it will defectively intervene between the matrix C_foc probe and any embedded Q_focP (Karimi 2013), as seen in (76).

However, if embedded C heads are endowed with a Q_foc probe features, we need to explain why there is no successive cyclic movement derivation for (75), as schematized in (77).

We propose that this follows from the intermediate Q_foc probe not triggering movement in Kalaallisut. While a probe Q_foc feature on the embedded C head intervenes for one-fell swoop movement into the matrix clause, it does not trigger movement to its specifier and therefore does not enable successive cyclic movement. On this view the difference between languages with successive cyclic A-bar movement and languages without reduces to whether these intermediate Q_foc probes trigger movement to their specifier or not. In the notation of Deal (2025), this difference is encapsulated in the probe specifications in (78a) and (78b).

In contrast a genuine focus C head bears the same probe in languages with and without successive cyclic movement:

A second configuration to consider is one where both the matrix and embedded C heads are genuine C_foc heads and both trigger movement to their specifier. If this is a possible configuration, we might expect cross-clausal focus movement to be possible: the Q_foc probe on the embedded C head Agrees with embedded C_focP and moves to its specifier. From that position Q_focP should be able to undergo a second step of focus movement into the matrix clause, since the embedded C head no longer intervenes. For the second author, this configuration does not arise, because embedded focus fronting is not possible to begin with (80).

However, embedded focus fronting is possible for the third author (81).

And yet, the third author does not allow cross-clausal focus movement (75). We propose that this is an instance of Criterial Freezing (Rizzi 2006). The first movement step is to a criterial position; the focus phrase is interpreted in the embedded clause. It therefore cannot move again to a second criterial position, namely specifier of matrix C_foc.19

There is one final question to consider, namely what is the output of the derivation in (77)? There is no movement of the embedded Q_focP to the specifier of the embedded C head, because the Q_foc probe feature on the C head does not trigger movement. Nor is there sequential focus movement from the specifier of one focus C head to another, because of Criterial Freezing. If Agree is a fallible operation (Preminger 2011), this account incorrectly predicts that Q_focP can be left in-situ in embedded clauses as in (82).

In this structure, the embedded C would Agree with meeqqa-n=uku satisfying C’s Q_foc feature but not move it to its specifier. The challenge is thus to determine what goes wrong in that derivation. To account for this we borrow from Cable’s (2010: 64–73) semantics for wh-questions, specifically three key assumptions. First, the sister of the Q-Particle has a non-trivial focus semantic value (i.e. is focused). Second, the Q-Particle denotes a variable over choice functions, each of which apply to the set of alternatives denoted by its sister. Finally, the interrogative C head is an operator that binds the variable over choice functions introduced by QP. These carry over to focus fronting as follows. The sister of the Q_foc-Particle is focused, i.e. induces alternatives. The Q_foc particle denotes a variable over choice functions that pick from these alternatives. Finally the C_foc head binds the variable denoted by Q_focP. We propose that, at least in Kalaallisut, this binding requires a local relationship between C_foc and QP. In the simple cases where QP originates in the same clause as its binder, this local relationship is achieved by movement. In the embedded case we are concerned with here there is no way to obtain this local relationship. First, the embedded C head is not a binder; it is an ordinary C head, not an operator, and thereby divorced from focus semantics. The actual C_foc is in the matrix clause and therefore cannot Agree with Q_focP in the embedded clause. Consequently, the required local configuration for binding of the choice function does not obtain and the structure is semantically uninterpretable.20 The basic pattern is now accounted for. Hyperraising is possible because the embedded CP does not intervene between the hyperraising probe on v and the embedded goal. Cross-clausal focus fronting is impossible because the embedded CP always intervenes for probing into the embedded clause. Before we move on to consider predictions of the proposed analysis, we need to explain the morphology of the focus clitics, specifically their number agreement with the focused phrase.

The realization of Q_foc as =una or =uku is governed by the number feature of the focused DP. If that DP is plural, Q_foc is realized as =uku (83); if it is singular, Q_foc is realized as =una (84).

So far we have been silent on how this difference is derived, but it is important for understanding the interaction between hyperraising and focus fronting discussed in section 5.1, so we will tackle it here. Our proposal is that the differing number morphology on the Q_foc head arises from Q_foc probing its sister and copying back its ϕ-features, which include number. More precisely, this probe is specified to interact with any feature (f) and be satisfied by foc: [int:f, sat:foc]. When it probes the sister of Q_foc, i.e. the focused phrase, it copies all the features of that goal (Joshi 2022; Grishin 2023; Deal & Royer 2025), including foc, category features and ϕ, if present. It is satisfied by foc and halts. At Vocabularly Insertion, the copied features govern the insertion of =una vs. =uku. The plural form =uku is inserted if the probe copied plural, whereas =una is inserted if it didn’t copy plural.

There are thus two contexts where =una will be inserted: when the probe on Q_foc copied singular (as in (84)) and when there was no number feature to copy.21 This latter situation is found where the focused phrase is a ϕ-less adverbial as in (87) and (88). In these ϕ-less contexts, the focus particle must surface as =una, indicating that =una is the elsewhere form:

In (87) the probe on Q_foc copies adv and foc from ippassaq. The most specified Vocabulary Item matching this feature combination is (86) and =una is inserted. In (88) the Q_foc probe copies adv, wh, and foc and, again, the most specified Vocabulary Item that matches this feature combination is (86) and the elsewhere form =una is inserted.

First, recall that on its own focus-fronting cannot cross a finite clause boundary:

However, if hyperraising brings a focused DP into the matrix clause, we predict that the DP can undergo subsequent focus fronting, because the embedded C head no longer intervenes between the matrix C head and the focused phrase. This is indeed the case, as seen in (90). The schematic tree in (91) illustrates the two-step movement.22

How does the first movement step come about, given that the focused DP is encased in a Q_focP? We propose that this is possible because of the prior ϕ-Agree operation between Q_foc and its sister, which derives the number inflection on the Q-Particle, plural =uku vs. elsewhere =una. As above, if a probe interacts with a goal, the probe copies all features from that goal. Therefore, in addition to ϕ-features, Q_foc copies the category feature of the focused element. If the focused element is a DP, that allows it to satisfy the hyperraising probe ([int:ϕ sat:ϕ+d^M]). Once hyperraised, its inherent Q_foc feature allows it to be focus fronted within the matrix clause.

The grammaticality of (90) also shows that the ungrammaticality of (89) is not due to a semantic or pragmatic problem with the intended content. The problem is syntactic: focus fronting cannot cross a clause boundary.

As seen in (92) and (93), wh-movement is also not restricted to DPs, indicating that the wh-probe is also flat: [int:wh, sat:wh^M].23

Given spurious wh-features on embedded C heads (ala McCloskey 2002; Abels 2003; 2012), the intervention analysis predicts that wh-movement is similarly clause bounded. The examples in (94) and (95) bear this out.

The explanation for the lack of successive cyclic focus fronting carries over to wh-movement. Unlike the matrix wh-probe, the wh-probe on intermediate C heads does not trigger movement and therefore does not result in successive cyclic movement through the specifier of intermediate C heads. In cases of embedded wh-movement, Criterial Freezing prevents the wh-phrase from moving on to the specifier of matrix C_wh. As with focus fronting, hyperraising can feed wh-movement in the matrix clause giving the appearance of cross-clausal wh-movement in (96).

Summing up this section, we add wh-movement as a second clause-bounded movement to a high position, as in Table 8.

Deal (2017) investigates hyperraising to object in Nez Perce (Sahaptian, Idaho) and develops an analysis in terms of delayed opacity. The intuition behind the Delayed Opacity analysis is that a low probe can “pull” a phrase out of a phase before that phase seals off, but a high probe cannot. In other words, contra Keine’s Height-Locality Connection, lower is freer. Crucially “low” means below then next higher up phase head and “high” means above the next higher up phase head. Specifically, if a probe is below the lowest matrix phase head, it can interact with a goal inside an embedded clause, because the embedded CP phase is not sealed off until that lowest matrix phase head is Merged. In contrast, if a probe is above the lowest matrix phase head it cannot probe into the embedded CP, because that lowest matrix phase head seals it off. To understand the predictions of the two analyses, we need to establish which categories are phase heads in Kalaallisut. Here we rely on Compton & Pittman’s (2010) argument that vP is not a phase in Inuit, and that the only phase heads and C and D. Under this assumption, the Delayed Opacity analysis accounts equally well for the data considered so far: because the hyperraising probe is both low (below the matrix C phase head) and fails to intervene both analyses predict hyperraising to be possible. Because the focus fronting probe is both high (not below the matrix C phase head) and acts as an intervener, both analyses predict cross-clausal focus fronting to be impossible, as noted in Table 9.24

In principle, however, it is possible to tell the two analyses apart. If a non-intervening probe is high, the intervention analysis predicts cross-clausal movement to be possible, whereas the Delayed Opacity analysis predicts it to the impossible. Conversely, if an intervening probe is low, the intervention analysis predicts cross-clausal movement to be impossible, whereas the Delayed Opacity analysis predicts it to be possible. These predictions are schematized in Table 10.

We are not aware of any phenomena in Kalaallisut that instantiates the first column. It is true that regular clausal agreement has been argued to involve a high probe—in C by Compton (2018) and in dedicated Agr projections above MoodP by Yuan (2018)—but if v is not a phase, Agree with Voice-internal subject and object DP goals will involve regular within-phase Agree under either analysis. We therefore set the first column aside and turn to the second column: are there instances of an intervening low probe? Here we suggest that this is instantiated by low wh-movement. The examples of wh-questions considered so far all have the wh-phrase in initial position, which we assume to be specifier of CP. However, Kalaallisut also has wh-medial questions:

Following Sherkina-Lieber’s (2005) analysis of Inuktitut, we assume that wh-medial orders involve short wh-movement to a position between TP and VoiceP. As above, wh-movement is implemented as Q_qP movement.25

With these two assumptions in place, we can consider the predictions made by either analysis. The intervention analysis predicts low wh-movement into the matrix clause to be ungrammatical, because the wh-probe is an intervenes. The Delayed Opacity analysis, in contrast, predicts low wh-movement into the matrix clause to be possible because it is to a position (spec-FP) below the next up phase head (matrix C):

In fact cross-clausal low wh-movement is ungrammatical:

If the analysis of low wh-movement is corroborated, the ungrammaticality of (101) and (102) would provide an argument in favor of intervention analysis over the delayed opacity analysis.

The second alternative analysis is inspired by Lee & Yip’s (2024) analysis of hyperraising in Cantonese and Vietnamese, which combines composite probing with phase unlocking (e.g. Rackowski & Richards 2005; van Urk & Richards 2015; Branan 2018; Davis & Branan 2019; Hedding & Yuan 2025; Little 2025).26 By virtue of ϕ-Agree with the embedded CP phase, the hyperraising probe unlocks the phase and probes the embedded subject DP, which results in hyperraising. While a composite hyperraising probe is central to both the intervention and phase unlocking analyses, the logic of the two analyses is the inverse of each other. In the intervention analysis a phrase is free to move across a clause boundary (no Phase Impenetrability Condition) unless another goal intervenes. On a phase unlocking analysis a phrase is unable to move across a phase boundary unless the phase is unlocked by Agree.

Within the interaction-satisfaction theory of Agree, unlocking of a phase is the result of a probe interacting with a phasal XP but not being satisfied by it. We assume the same probes as on the intervention analysis: a composite hyperraising probe with differing interaction and satisfaction conditions—[int:ϕ, sat:ϕ+D^M]—and a flat focus probe with identical interaction and satisfaction conditions; [int:Q_foc, sat:Q_foc]. The assumptions about the featural make-up of CP (ϕ-features but no D feature) are also the same. The hyperraising probe can interact with the embedded CP by virtue of the embedded CP bearing ϕ-features, but it not satisfied by it, since the embedded clause lacks a D feature. This explains the existence of hyperraising in Kalaaallisut.

Turning to focus fronting, there are two cases to consider. If the embedded CP is itself focused, it satisfies the focus probe and probing halts without the probe interacting with a possible goal inside the embedded CP. This is movement of a phasal CP, not movement across it, and no phase unlocking is involved. If the embedded CP is not focused, the Phase Impenetrability Condition prevents the focus probe from probing into the embedded CP. The focus probe cannot unlock the embedded phase and render the Phase Impenetrability Condition inactive because the embedded CP does not bear a focus feature that would allow the matrix focus probe to interact with it. Without interaction there is no phase unlocking. The same reasoning applies to wh-movement, which too can target wh-phrases of various categories (minimally DPs and adverbs). The relevant probe therefore has identical interaction and satisfaction conditions [int:wh, sat:wh] and is unable to unlock the embedded CP phase. The lack of unlocking depends only on probe specification and not on probe location. Therefore the lower wh-probe on Foc responsible for low wh-movement (Sherkina-Lieber 2004) also cannot unlock the embedded CP, which explains the ungrammaticality of cross-clausal low-wh movement as well.

Since the empirical coverage of the phase analysis is as good as that of the intervention analysis, we must consider their analytic and theoretical assumptions. Since the two analyses make the same analytic assumptions—i) CPs bear ϕ-features, but not D features; ii) the hyperraising probe is composite; and iii) the focus and wh-probes are flat—we cannot differentiate them on analytic grounds. That leaves us with theoretical considerations. Here we believe the intervention analysis comes out on top. If phases define cyclic spell-out domains and spell-out renders the complement of the phase-head impenetrable for syntactic operations, the spell-out domain of a lower phase is spelled out before a higher phase head is Merged. Then how is phase unlocking by a phase external probe possible? Despite sporadic mentions of this, e.g. Lee & Yip (2024: fn. 23), it seems to us a fundamental architectural mystery. In contrast, the intervention analysis relies solely on well-established conditions on Agree.27 In addition, the intervention analysis does not assume phases (Halpert & Zeijlstra 2024), and the phase unlocking analysis does. Other things being equal, this makes the intervention analysis more attractive on grounds of theoretical simplicity.

In addition to arguing for the Height-Locality Connection, Keine (2020) develops a theory of locality, according to which individual probes may be associated with a horizon.28 A horizon is a particular category feature that determines which domains are inaccessible to the probe. For instance, if a probe has horizon T it cannot probe into TP. If a probe lacks a horizon, it can keep probing. In this system, the Kalaallisut hyperraising probe lacks a horizon, whereas the Q_foc probe on C has horizon C. The latter means that the focus probe will abort its search when it encounters the embedded CP and therefore be unable to probe into it. This correctly captures the clause-boundedness of focus fronting. Setting phases aside for the moment, leaving the hyperraising probe without a horizon also seems to capture its behavior: nothing forces it to abort its search upon encountering CP and it can therefore probe into the CP and hyperraise an embedded DP. However, Keine (2020) argues that there are two types of locality constraints: horizons and phases. If so, the question of how the hyperraising probe probes into the embedded CP—given that it is a phase—is left unanswered. We could add phase unlocking to the horizon system and create a double-tiered analysis that employs both phase unlocking and horizons. If the phase unlocking analysis can capture the same facts on its own, that seems preferable to a dual analysis. A second point is that the intervention analysis is grounded in language specific facts that the horizon analysis does not take into account. The ability of the hyperraising probe to probe past an embedded CP is predicated on the fact that clausal complements do not bear a D feature in the language, as evidenced by their inability to bear case. That connection is lost under the horizon analysis, where the larger search domain of the hyperraising probe is attributed to its lack of a horizon. Second, the fact that an embedded CP does intervene for focus fronting and wh-movement is a consequence of these probes being flat: whatever they interact with satisfies them. That in turn is forced by distributional facts: as shown in section 3.2, any phrase capable of being focused can focus front. Similarly, there is no category restriction on which phrases can front in wh-questions, indicating a non-picky probe. On the horizon analysis, the difference between cross-clausal hyperraising on the one hand and clause-bounded focus fronting and wh-movement on the other is basically an accident.