XP- and X 0 -movement in the Latin verb: Evidence from mirroring and anti-mirroring

Latin verbs appear to both obey and disobey the Mirror Generalization (Baker 1985), albeit in different inflectional forms (Cinque 1999). Given the importance of the Mirror Generalization to morphosyntactic theorizing, this situation deserves scrutiny. We develop an analysis on which all Latin finite verbs, whether mirroring or anti-mirroring , share a single, simple, virtually invariant derivation, involving one step of head movement (Asp to T) and one step of phrasal movement ( v P to [Spec,TP]). On this analysis, the Mirror Generalization is valid for Latin, despite appear-ances, but it is crucially about structures formed by operations on heads: phrasal movement can give rise to apparent violations of it (Myler 2017). The analysis extends readily to nonfinite forms, solving an anti-mirroring paradox arising among the participles. It also makes correct syntactic predictions. When the verb word is a constituent, it is a TP, and the analysis correctly predicts that it should move as a phrase, not as a head. The analysis also makes correct predictions about the positions of v P-adjuncts, and is fully compatible with what is known about leftward argument movement out of v P. Finally, unlike two competing analyses (Embick 2000; Calabrese 2019), it accounts for anti-mirroring in Latin without any stipulations placing passive Voice in an unexpectedly peripheral position within the verb word. The larger picture that emerges is one in which phonological words need not correspond to syntactic constituents, but can instead be reflexes of linearly contiguous series of morphemes suspended across potentially vast regions of syntactic space (Julien 2002).

The concept in (1) is standardly interpreted as making available the following structural diagnostic: (2) If a root and some functional heads in its extended projection (Grimshaw 1991) are assembled into a single "morphological word," then, ceteris paribus, the higher that one of those functional heads is syntactically, the farther away from the root it should be in the morphological word (and vice versa). In Latin, the heads in the extended projection of V shown in (3) do seem to be packaged into a single morphophonological word. Therefore, the MG, in conjunction with the clause structure in (3), leads us to expect that the Latin verb word should display the following morpheme order: 3 (4) Root > v > Voice > Aspect > Tense > Mood (> = 'precedes') If this were so, Latin verbs would transparently obey the MG. But in fact, they seem to both obey and disobey the MG, albeit in different parts of their paradigm, as will be shown in Sections 2.1 and 2.2. This is peculiar, and naturally gives rise to the following question: (5) Given the crosslinguistic robustness of the MG, how can it be that Latin verbs seem to both obey and disobey it (in different inflectional forms)? 4 An analysis that answers this question will be developed in Section 3. But first, we will give concrete examples of Latin verb forms that appear to obey the MG (henceforth mirroring forms) and Latin verb forms that appear to disobey it (anti-mirroring forms). This is the task of the next two subsections. 4

Anti-mirroring verb forms
In many passive forms, on the other hand, the exponents of Mood, Tense, and sometimes Aspect occur closer to the root than does the exponent of passive Voice, in seeming contravention of the MG. (The unexpectedly external position of the passive Voice morpheme is discussed in Cinque 1999: 197, citing Calabrese 1985Embick 2000: 196-199;Baker 2014: 8-9;and Calabrese 2019, §3.4.4.2.) As an example, consider the present passive subjunctive forms of laud-'praise' ((9)). In these forms, the morpheme -ē, a portmanteau that expones both subjunctive Mood and present Tense, occurs closer to the root than does the exponent of passive Voice. 7 In the 3sg, 1pl, and 3pl forms, the Mood/Tense portmanteau -ē is even separated from Voice by an agreement morpheme. (To briefly foreshadow part of the analysis, we will propose below that mood and tense features are both on T in Latin.) (9) a. laud-e-r praise-prs.sjv-pass.1sg '(that) I may be praised' b. laud-ē-ri-s praise-prs.sjv-pass-2sg '(that) you may be praised' c. laud-ē-t-ur praise-prs.sjv-3sg-pass '(that) he/she/it may be praised' d. laud-ē-m-ur praise-prs.sjv-1pl-pass '(that) we may be praised' e. laud-ē-minī praise-prs.sjv-pass.2pl '(that) you may be praised' f. laud-e-nt-ur praise-prs.sjv-3pl-pass '(that) they may be praised' Nor is anti-mirroring in passive forms restricted to the paradigm in (9). The exponent of passive Voice also occurs farther from the root than do the following morphemes, among others (see the Appendix for the full paradigm): (10) a. -bi (indicative Mood + future Tense) laud-ā-bi-t-ur praise-th-fut.ind-3sg-pass 'he/she/it will be praised' b. -bā (indicative Mood + past Tense + imperfective Aspect) laud-ā-bā-t-ur praise-th-impf.ind-3sg-pass 'he/she/it was being praised' c. -rē (subjunctive Mood + past Tense + imperfective Aspect) laud-ā-rē-t-ur praise-th-impf.sjv-3sg-pass '(that) he/she/it might be praised' All the forms examined in this subsection follow the template in (11), which appears to violate the MG. (The notation {X, Y} here means that the order of X and Y depends on the particular form at hand.) (11) Root (> Theme Vowel) > Mood+Tense (+Aspect) > {φ (Person + Number) , Voice} The question, then-as mentioned-is how it can be that some verb forms seem to violate the MG (and in a language in which other verb forms obey it, no less).

Analysis
An optimal, maximally principled solution to the mirroring/anti-mirroring puzzle just presented should satisfy the following theoretical desiderata: (12) a. Uniformity: All the finite verb forms-whether they appear to obey the MG or not-should share a single derivation (which would keep idiosyncratic, verbform-specific stipulations to a minimum). b. Generality: Although the puzzle arose among the finite forms, the analysis should extend parsimoniously to (and ideally shed new light on the structure of) nonfinite and even deverbal forms. c. Syntactic fit: Although the puzzle is "morphological," in that it is about morpheme order, the solution should ideally make correct predictions about Latin syntax, and should be compatible with everything that is independently known about it.
This challenge can be met. The present section will develop an analysis of mirroring and anti-mirroring in Latin that satisfies all three desiderata in (12). On the analysis developed below, mirroring and anti-mirroring verb forms have the same derivation-which involves one step of XP-movement (vP-movement to [Spec,TP]) and one step of head movement (Asp to T). The following two subsections lay out this derivation for a mirroring and an anti-mirroring verb form, respectively.

Deriving the mirroring forms
Consider the following mirroring form: (13) Perfective Aspect inside past Tense laud-ā-v-era-t praise-th-pfv-pst-3sg 'he/she/it had praised' On our analysis, (13) has the following derivation: (14) Derivation of laudāverat 'he/she/it had praised'   Five aspects of this derivation deserve comment. First, the Voice head in finite clauses bears two φ-probes that trigger Agree operations: one for Person and one for Number. 8 (On nonfinite environments, see Section 4.3.) This proposal is motivated by the observation that there appears to be a tight connection in the Latin verb word between the exponence of Voice proper (active vs. passive/nonactive) and the exponence of φ-features (henceforth φ for short), whose values are obtained from the structurally most prominent nominal. 9 This is particularly clear in passive paradigms such as (9). In the 2pl form laudē-minī '(that) you may be praised', passive Voice and 2pl are realized by the portmanteau -minī. And in the forms in which passive Voice and φ are realized separately, their order is variable across forms. Thus,, and laude-nt-ur (3pl), φ precedes Voice, but in laudē-ri-s (2sg), Voice precedes φ. All this can be accommodated straightforwardly if Voice bears the φ-probes in the syntax, and Voice nodes with certain combinations of features after Agree undergo postsyntactic Fission (in the Distributed Morphology sense: Harley & Noyer 1999, §3.3), with the resulting nodes linearized in particular orders depending on their features (cf. Embick 2000: 199). 10 Secondly, Asp undergoes head movement to T (and is linearized to the left of T, just as the V-v complex is in French: Pollock 1989). The derivation in (14) shows this head movement as occurring in the syntax, but that is only for convenience: nothing will hinge on whether this step of head movement, or any other, is syntactic or postsyntactic, so we will remain agnostic on this point. (For discussion, see Matushansky 2006;Arregi & Pietraszko 2018;to appear;Dékány 2018;Newman 2018a;and Harizanov & Gribanova 2019, a.m.o.) Third, vP undergoes phrasal movement to [Spec,TP]. That is, Latin belongs to the class of languages in which the derivation of an ordinary clause involves movement of a (phrasal) verbal projection (cf. many of the papers in Carnie &Guilfoyle 2000 andCarnie, Harley &Dooley 2005). The proposal that Latin has this property is also defended by Danckaert (2012;2017a;b) and Gianollo (2016) on independent grounds (modulo certain differences concerning the precise categorial identity of the verbal projection moved, and its landing site; see also Bailey 2010). To our knowledge, however, this hypothesis has never been employed to explain the (mirroring and anti-mirroring) morpheme orders within the Latin verb word; our analysis will do just that. 11 8 Following Heck & Müller (2007), we use the following feature notation: [F:□] = unvalued feature; [ * F:□ * ] = unvalued feature that probes for a value; [ * F: val * ] = initially unvalued feature that has successfully probed for a value. 9 We assume for concreteness that Latin nominals are KPs, headed by a case morpheme K (cf. Lamontagne & Travis 1987;Bittner & Hale 1996), but nothing hinges on this. 10 The claim that the φ-probes are on Voice might seem surprising, since their counterparts in better-studied languages (e.g., English) are on T. (A reviewer raises this concern in connection with the Person probe, noting that (14) "involves agreement with person features, whose overt exponence is standardly linked to T.") But this component of the analysis is empirically and conceptually unproblematic. Empirically, the location of agreement probes within clause structure varies a great deal crosslinguistically (Cinque 1999, a.m.o.). Conceptually, nothing in the theory rules out Voice heads of the sort proposed here. Since T and Voice are both clausal functional heads, and T bears Person and Number probes in some languages, we expect Voice to in other languages: forbidding this would require a stipulation. This expectation is especially clear because Person and Number probes seem to have no more of an inherent connection to Tense than they do to Voice. That said, the view that the φ-probes are on Voice will make some additional assumptions necessary in Section 4.3; see fn. 23 for an alternative. 11 A reviewer asks what drives this vP-movement. We assume for concreteness that, in Latin, T (as well as a certain a head discussed in Section 4) bears a movement-driving feature [•v•]  Fourth, as suggested to us by Nico Baier, T is the locus of both tense and mood featuresi.e., T(ense) and Mood are not projected as separate heads in Latin. This view has two advantages over an alternative on which T and Mood are separate heads in Latin. First, it makes it easy to understand why there are no clear cases of tense and mood being exponed separately in this language (see the Appendix). Secondly, if T and Mood were separate in Latin, Mood would presumably be higher, and we would therefore have to stipulate that leftward adverbials standardly analyzed as TP-adjuncts in English were MoodP-adjuncts in Latin (since, if they were TP-adjuncts, they would interrupt the verb word, which in fact is impossible). This stipulation is avoided on the T-only view adopted here: on this view, TP-adjuncts in English correspond to TP-adjuncts in Latin, and the latter do not interrupt the verb word.
Fifth, the external argument (EA) originates in [Spec,vP], not [Spec,VoiceP] (Collins 2005;Merchant 2013). This is advantageous for three reasons. First, if the EA originated in [Spec,VoiceP], it would be incorrectly predicted to interrupt the verb word, unless it were stipulated either that [Spec,VoiceP] was linearized to the right or that the EA obligatorily moved to some position above [Spec,TP]. Both stipulations are avoided if the EA originates in [Spec,vP]. Secondly, if the EA originated in [Spec,VoiceP], then, after vP-movement, the EA would not c-command the internal argument. Although it is unclear how problematic that would be, it might make it difficult to account for the fact that, even when all the arguments scramble leftward past low adverbs, the EA precedes the internal argument by default (see fn. 32 and Section 5.3). This problem is avoided if the EA originates in [Spec,vP]. Finally, if the EA originated in [Spec,VoiceP], then Voice would have to probe upward into [Spec,VoiceP] and then, only if that was unsuccessful, downward into vP, complicating the theory of probing. But if the EA originates in [Spec,vP], Voice can uniformly probe downwards, into its c-command domain. 12 On this analysis, then, the derivation of every synthetic finite verb form in Latin involves one step of XP-movement (vP-movement to [Spec,TP]) and one step of head movement (Asp to T). The analysis could therefore be dubbed "the XP-movement + X 0 -movement Analysis," but for convenience we will shorten this to the XP/X 0 Analysis.
As illustrated in (14), the XP/X 0 Analysis accounts straightforwardly for clearly mirroring verb forms. In the form derived in (14), for example-laud-ā-v-era-t 'he/she/it had praised'-the mirroring order (perfective Aspect inside past Tense) comes about because Asp undergoes head movement to T.

Deriving the anti-mirroring forms
Consider now the following anti-mirroring form: Subjunctive Mood/present Tense portmanteau inside passive Voice laud-ē-t-ur praise-prs.sjv-3sg-pass '(that) he/she/it may be praised' 2014; 2017a; b; and Gianollo 2016 for relevant discussion). We cannot attempt to evaluate the arguments for this latter view here, especially since they generally assume that all languages satisfy the EPP somehow, which, however, is controversial (see Cable 2012 for arguments to the contrary). At any rate, nothing in the analysis will hinge on the [•v•] assumption; consequently, we do not exclude the possibility that Latin vP-movement is instead driven by a feature of the vP itself (cf. Bošković 2007, a.m.o.). 12 Two final remarks are in order about vP. First, we assume that VP is linearized to the left of v, since this derives the default SOV order without further ado. (On other orders, see fn. 32 and Section 5.3.) V might undergo string-vacuous head movement to v, but, lacking evidence for this, we assume it does not.
Secondly, (14) shows the theme vowel -ā as exponing v. Alternatively, it may be postsyntactically adjoined to v (Oltra-Massuet 1999; Embick 2010). If overt verbalizers such as -fic '-fy' are v heads, the latter view would allow us to avoid positing a double-vP configuration for, e.g., ampli-fic-ā-re 'widen'. But if -fic is a V (cf. Bruening 2016), ampli-fic-ā-re does not involve a double vP regardless of the status of the theme vowel. Nothing will hinge on whether theme vowels expone v or are postsyntactically adjoined to it; the trees adopt the former treatment for convenience.
On the XP/X 0 Analysis, this anti-mirroring form has exactly the same syntactic derivation as the mirroring form examined in Section 3.1: 13

(16)
Derivation of laudētur '(that) he/she/it may be praised' In (16), as in (14), Voice probes to value its φ-features and gets the values it seeks from the structurally closest KP. (As discussed in fn. 13, Voice PASS[Pers:3, Num:sg] undergoes postsyntactic Fission, the result of which is anticipatorily shown in (16).) As before, Asp undergoes head movement to T, although this has no overt effect in (16) because Asp itself is not overt in this form. Finally, also as before, vP moves to [Spec,TP], and it thereby comes to precede the T head -ē, which expones both subjunctive Mood and present Tense.
The anti-mirroring observed in laud-ē-t-ur '(that) he/she/it may be praised' can be analyzed as involving two anomalies: Mood/Tense surfaces unexpectedly close to the root, and Voice surfaces unexpectedly far from it. On the XP/X 0 Analysis, these anomalies have a single source: both come about because vP moves to [Spec,TP], stranding Voice. This vP-movement carries the root to a position where it is linearized close to the exponent of Mood/Tense and far from the exponent of Voice.

Interim summary and theoretical implications
On the XP/X 0 Analysis, then, all synthetic finite verbs in Latin, whether they appear to obey the MG or not, share a single, simple derivation, involving one step of head movement (Asp to T) and one step of phrasal movement (vP to [Spec,TP]).
On this analysis, the MG is in fact valid for Latin-contrary to initial appearances-but it must crucially be understood as a generalization about structures formed by operations 13 We say "exactly the same syntactic derivation" because there is one minor postsyntactic difference: when Voice PASS bears the features [Pers:3] and [Num:sg], as in (15)/(16), it undergoes postsyntactic Fission, resulting in separate exponents for the φ-features (-t) and Voice proper (-ur). In the mirroring form laud-āv-era-t 'he/she/it had praised', by contrast ( (14)), Voice ACT is not exponed separately from the φ-features, so by hypothesis there is no Fission. As discussed in Section 3.1, Voice PASS does not always undergo Fission: Voice PASS[Pers:2, Num:pl] is realized by the portmanteau -minī. The generalizations about when Voice does and does not undergo Fission appear to be true morphological idiosyncrasies of Latin, though rather superficial ones (cf. Embick 2000: 199). Because Fission is standardly considered a postsyntactic operation (Harley & Noyer 1999, §3.3), the observations discussed in this footnote are fully compatible with our claim that mirroring forms ((13)) and anti-mirroring forms ((15)) have identical syntactic derivations. on heads (e.g., head movement). 14 As shown in Section 3.2, Latin does have some verb forms that appear to disobey the MG, but that is an illusion caused by the fact that all the verb forms are derived in part by a step of phrasal (vP) movement. Phrasal movement, being subject to less strict locality constraints than head movement, 15 can give rise to superficially anti-mirroring morpheme orders, a conclusion also arrived at on independent grounds by Buell (2005) On the XP/X 0 Analysis, the Latin verb word (e.g., laudāverat 'he/she/it had praised' or laudētur '[that] he/she/it may be praised') is not a complex head, contra Embick (2000) and Calabrese (2019, §3.4.4.2). A perhaps more surprising consequence of the analysis is that, when one or more arguments (or adjuncts) remain within vP, the verb word is not even a constituent (see (14) and (16)).
If the XP/X 0 Analysis is on the right track for Latin, then, it strongly supports an alexist or squishing view of syntax/morphology relations. (The term alexist is from Svenonius 2018, and squishing is from Myler 2017; Myler credits Glyne Piggott for the coinage.) On such a view, there is no syntactic correlate of (morpho)phonological wordhood; hence, there are no syntactic words. Instead, phonological words are autonomously assembled by the prosody from syntactic terminals, on the basis of phonological properties of the exponents of those terminals-e.g., some exponents are inherently phonologically dependent and therefore simply lean (left or right) in the prosody. A consequence of this view is that phonological words can be assembled from collections of morphemes that are linearly adjacent but suspended across large regions of syntactic space-and, in fact, phonological words need not always correspond to syntactic constituents, as is argued on independent grounds by Julien (2002) (2017); Kim (to appear); and references cited in those works; see also Halle & Marantz (1993) and much related work in Distributed Morphology.
See the Appendix for the full paradigm of all the (synthetic) finite, nonfinite, and deverbal forms of the sample verb laud-'praise', laid out in a way that makes it clear how the XP/X 0 Analysis successfully derives them. 16

A note on the locality of allomorph selection
Before we extend the XP/X 0 Analysis to nonfinite forms, it will be worth noting how it interacts with existing hypotheses about the locality domain for allomorph selection, given the allomorphy patterns observed within Latin verbs. The basic observation to be accounted for is that, within the verb word, a head can often trigger allomorphy on another head in a way that seems nonlocal in terms of hierarchical structure, given the structures proposed here.
For example, consider the verbs manēre 'stay', monēre 'warn', and vidēre 'see'. These are second-conjugation verbs: their theme vowel, when present, is -ē/-e. Yet they behave differently in the presence of perfective Aspect (Embick 2000: 189). Consider their 3sg 14 This is true of the diagnostic version of the MG ((2)). Baker's formulation ((1): "Morphological derivations must directly reflect syntactic derivations (and vice versa)") is almost completely obeyed on the XP/X 0 Analysis even if attention is not restricted to operations on heads. As a reviewer notes, together with head movement, "phrasal syntactic movement can produce a morphological derivation that directly reflects (in linearization) the syntactic computation, including both External/Internal Merge." The only caveat is that the small amount of postsyntactic Fission in Latin (Section 3.1) does not appear to directly reflect anything in the syntactic derivation. 15 For further discussion of this hypothesis, see Section 5.1, particularly fn. 24. 16 There are a couple of oddities in the imperative forms. pluperfect active indicative forms ( (17)). (We illustrate with these for maximum comparability with laudāverat 'he/she/it had praised' in (14).) In each of (17a-c), a particular V idiosyncratically triggers the realization of Asp PFV in a different way: as -s, -u, or zero. Furthermore, in (17-c), Asp PFV also triggers allomorphy on V, causing it to surface as vīd-rather than vid-. (By contrast, the long vowel in (17-a) is due to regular phonology.) On the XP/X 0 Analysis, however, V and Asp are not structurally close to one another: V is within the vP in [Spec,TP], and Asp is a subconstituent of the complex T head. The XP/X 0 Analysis is therefore incompatible with the hypothesis that, for two elements α and β, α cannot trigger allomorphy on β if the two are separated by a maximal projection boundary (Bobaljik 2012: 13). A fortiori, the analysis is incompatible with the hypothesis that α cannot trigger allomorphy on β if they are separated by the boundary of any non-minimal (non-X 0 ) projection, even an intermediate projection (Bobaljik & Harley 2017: 150). As a reviewer notes, however, the XP/X 0 Analysis dovetails naturally with Ostrove's (2018) stretching-based approach to Vocabulary Insertion (see Embick 2010 for related ideas). Ostrove (2018: 1281) defines stretch as follows: Let T be an ordered n-tuple of linearly ordered terminal nodes ⟨t 1 , …, t n ⟩ such that for all t ∈ T, t = t 1 or t is an element of the extended projection of t 1 . a. For all k = 1 … n, t k is a stretch. (Every node is a trivial stretch.) b. For any n > 0, if t k is a stretch, then ⟨t k , …, t k + n ⟩ is a stretch.
Ostrove argues that, in Irish, a single exponent can sometimes realize the sequence of terminals Mood-√root-v, to the exclusion of T, even though the former three nodes form neither a span ("a complement sequence of heads […] in a single extended projection"-Svenonius 2012) nor a morphosyntactic constituent. On his analysis, this is because that sequence is nonetheless a stretch, and Vocabulary Insertion can target stretches.
Ostrove, then, invokes stretches primarily as targets for Vocabulary Insertion. However, his definition of stretch is modeled on Merchant's (2015: 288)

definition of span, and
Merchant argues for what he calls the Span Adjacency Hypothesis: "Allomorphy is conditioned only by an adjacent span." Pursuing the stretching-based approach, it is natural (in fact, a minimal extension) to develop it further by proposing (19): Stretch Adjacency Hypothesis Allomorphy {is conditioned only / can be conditioned} by an adjacent stretch.
(Which version is more adequate is an open question.) If (19) is on the right track, the allomorphy patterns in (17) become unsurprising: in all three forms, the form of Asp PFV is conditioned by the linearly adjacent stretch V-v; and in (17-c), the form of V is conditioned by the linearly adjacent stretch v-Asp PFV , even though the latter is not a constituent. Naturally, the proposals in (18)-(19) deserve much more detailed testing, and they are discussed further in Section 4.3. What is important here is that, as a reviewer notes, the XP/X 0 Analysis is compatible with (a minimal extension of) one independently motivated existing approach to Vocabulary Insertion: Ostrove's (2018) stretching approach.

Extension: Deriving nonfinite verb forms
Section 3 developed an analysis of the derivation of Latin finite verb forms that relies crucially on vP-movement. On that analysis, all the finite verb forms in the language, whether mirroring or anti-mirroring, have the same derivation-one that involves one step of head movement and one step of vP-movement.
This result in itself furnishes a strong argument that Latin makes use of vP-movement (see Bailey 2010;Danckaert 2012;2017a;b;and Gianollo 2016 for similar proposals). Nonetheless, it is worth asking whether there is any (further) independent evidence for this conclusion-e.g., other empirical phenomena that initially appear puzzling, but receive a satisfying explanation if vP moves in Latin.
This section argues that the answer is yes-and specifically, that a phenomenon of exactly this type emerges when we turn our attention from the finite to the nonfinite verb forms. The discussion that follows will focus on two nonfinite forms in particular: the Perfect Passive Participle and the Future Active Participle (cf. Matthews 1972: 83-86;Embick 2000: 215-216;Calabrese 2019, §3.4.4.1.5; and references cited there). We will argue that the Future Active Participle has what initially seem to be contradictory properties, but the apparent paradox vanishes if both participles are derived partially by vP-movement.

Deriving the Perfect Passive Participle
The Future Active Participle (e.g., laudātūrus '[that is] going to praise') is derived from the same "stem" (descriptively speaking) as the Perfect Passive Participle (e.g., laudātus 'having been praised'-the "stem" in this case being laudāt-), with the addition of an element -ūr. Because the Perfect Passive Participle lacks -ūr and is thus structurally simpler, we will discuss first its derivation and then that of the Future Active Participle Livy, Ab Urbe Condita 32.29.1 …mūrus et porta dē caelō tāc-t-a era-nt. wall.nom and gate.nom from sky.abl touch-ptcp-n.nom.pl be.impf-3pl '…the wall and the gate had been struck by lightning.' ('…the wall and the gate were having-been-touched from the sky.') Anna laud-ā-t-a es-t. [C] Anna praise-th-ptcp-f.nom.sg be-3sg 'Anna {was / has been} praised.' ('Anna is having-been-praised.') On our analysis, the derivation of (21) is, in part, the following. (On the syntax of the auxiliary, see fn. 19. On the apparent φ-lessness of Voice in nonfinite forms, see Section 4.3.)

(22) Partial derivation of (21)
On this analysis, the participial morpheme -t here is a categorizing little a head-i.e., an adjectivalizing head (Marantz 1997, a.m.o.)-that takes a full-fledged (phrasal) verbal projection as its complement. 17 What is most important here about (22) is that the vP moves to [Spec,aP], and it is in this way that the V-v sequence laud-ā-comes to immediately precede -t. Finally, the aP is the complement of an Agr(eement) head that bears unvalued Gender The derivation in (22) yields the correct morpheme order, but no argument has yet been given that [ vP … laud-ā-] surfaces to the left of -t as a result of vP-movement. (After all, if the complement of -t were simply linearized to its left, no vP-movement would be needed to derive the morpheme order.) The argument for vP-movement will emerge only once we have examined the more complex of the two participles under discussion: the Future Active Participle.

Deriving the Future Active Participle
The Future Active Participle is shown in context in (23) (adapted from Pinkster 1987: 194) and (24) Mārcus Annam laud-ā-t-ūr-us es-t. [C] Mark.nom Anna.acc praise-th-ptcp-prosp-m.nom.sg be-3sg 'Mark is going to praise Anna.' As mentioned, the Future Active Participle is built from the same structural core as the Perfect Passive Participle. Whichever allomorphs of the root and the participial morpheme appear in the Perfect Passive Participle, the same ones appear in the Future Active Participle (with a few exceptions -Aronoff 1994: 177), along with the extra element -ūr that is unique to the latter: vehere 'carry', vec-t-us 'having been carried', vec-t-ūr-us '[that is] going to carry'; iubēre 'order ', ius-s-us, ius-s-ūr-us;ferre 'bear', lā-t-us, lā-t-ūr-us, etc. (Embick 2000: 216). This systematic correspondence shows that the two participles must have similar syntactic derivations. Further scrutiny, however, reveals that the Future Active Participle has properties that seem rather contradictory. These have to do with the suffix -ūr, which we take to be a prospective Aspect head, accounting for the participle's "future" orientation.
First, on the XP/X 0 Analysis, when T is merged in, vP obligatorily moves to [Spec,TP] (and it never moves away). Thus, the fact that, alongside [C] Anna laudāta est 'Anna was praised', Latin also allows [C] Anna est laudāta might seem problematic, since this might seem like optional vP-movement. But the problem is illusory: [(Anna) laudāta] (where Anna may or may not move out) is not a vP but a (participial/adjectival) AgrP, and can therefore move around rather freely, as adjectival phrases can generally in Latin (see also fn. 32). This accounts for the fact that participle and auxiliary need not be adjacent. For example, [C] Anna laudāta nōn est 'Anna was not praised' is derived from (i) by movement of [ AgrP Anna laudāta] leftward past the higher particle nōn 'not' (possibly with further, string-vacuous leftward movement of Anna). This proposal is independently (if indirectly) motivated by the ability of XPs quite generally to move leftward out of vP and past nōn (Section 5.3).
Secondly, (i), contra Embick (2000), does not treat [C] Anna laudāta est 'Anna was praised' as having essentially the same syntax as synthetic passives (e.g., [C] Anna laudātur 'Anna is praised') but different features and consequently a different morphological realization. Rather, according to (i), [C] Anna laudāta est is built from esse 'be' and a (participial/adjectival) AgrP. One might worry that this might predict that such structures should always be interpreted as stative (they are in fact often interpreted as eventive  (i) have is an empirical matter. The eventive interpretation of (i) is presumably due to material within [ vP Anna laud-ā-].
Naturally, (i) requires further elaboration and testing, and comparison with alternatives-which, for space reasons, we leave for future work.
On the one hand, -ūr occurs farther from the root than does the participial suffix -t, which we analyze as a categorizing a head (Section 4.1). This would seem to suggest that -ūr is syntactically higher than -t, given the MG-particularly since Section 3.3 showed that there is good evidence that the MG holds for Latin.
On the other hand, that conclusion seems backwards. Laud-ā-t-ūr-us '[that is] going to praise' is fundamentally a participle, not a prospective AspP (the latter of which would be a verbal category in the extended projection of V). Crucially, both Future Active and Perfect Passive participial phrases can serve as adnominal adjuncts, as in (25) Cicero, Tusculanae Disputationes 4.14 opīniō ven-t-ūr-ī bonī belief.nom come-ptcp-prosp-n.gen.sg good.gen 'a belief in good to come' ('…good that is going to come') The paradox, then, is that the Future Active Participle's categorial identity seems to be determined by the participial morpheme -t, but the prospective Aspect morpheme -ūr occurs outside of -t-another case of anti-mirroring.
Just as the Mood/Voice anti-mirroring in certain finite forms (which is really two puzzles: the unexpectedly internal position of Mood and the unexpectedly external position of Voice) was argued in Section 3.2 to come about through vP-movement, so too, we propose, does this instance of anti-mirroring: Partial derivation of Mārcus Annam laudātūrus est [C] 'Mark is going to praise Anna' This analysis solves the paradox laid out above. On the analysis in (27), participial -t is higher than prospective Aspect -ūr, explaining why it is the former that determines the categorial identity of the participle. But although -t is higher than -ūr, it ends up linearly closer to the root than -ūr does, because vP moves to the specifier of -t (i.e., to [Spec,aP]), carrying the root along. Let us recapitulate. Section 4.1 proposed that, in Perfect Passive Participles such as laudā-t-us 'having been praised', [ vP … laud-ā-] ends up to the left of the participial morpheme -t (= a) because it undergoes phrasal movement to [Spec,aP]. This is not obvious if one considers only the Perfect Passive Participle. However, the hypothesis that vP moves to the specifier of -t pays a significant empirical and theoretical dividend in connection with the Future Active Participle: it explains straightforwardly why, in this participle, -t occurs closer to the root than -ūr does, even though it determines the categorial identity of the participle and therefore must be higher than -ūr. 20 The hypothesis that Latin makes use of vP-movement, then-which was adopted in Section 3 to explain anti-mirroring in certain finite verb forms-is indeed supported by evidence independent of those forms: it solves an otherwise thorny paradox involving the Future Active Participle. 21

A note on the ϕ-lessness of Voice in nonfinite forms
Before we consider the syntactic predictions of the analysis, a few words are in order about the apparent φ-lessness of Voice in nonfinite forms such as the Perfect Passive Participle laudāta 'having been praised' ( (22)) and the Future Active Participle laudātūrus 20 Another positive consequence of this analysis is that the relation between Agr and a is identical in the two participles: in both, Agr takes aP as its complement. By contrast, if the Gender and Number probes were on a, and the agreement morpheme resulted from postsyntactic Fission, then the morpheme order in the Perfect Passive Participle could still be accounted for, but not that in the Future Active Participle. The former would still be correctly predicted to be (e.g.) laud-ā-t-us 'having been praised', but the latter would be incorrectly predicted to be (e.g.) *laud-ā-t-us-ūr (intended: '[that is] going to praise'). This problem could be fixed by positing a Local Dislocation operation (Embick & Noyer 2001), but such a stipulation is unnecessary on the analysis in (27). 21 Embick (2000: 215-219) analyzes -t (which he refers to as -t-/-s-) as a default realization of Asp surfacing in deverbal structures-those built from a verbal core but lacking T. (He focuses on the same two participles as we do.) He writes: "The motivation for positing the head Asp in such cases lies in the syntax and the semantics […] in that aspectual properties are implicated in nominalizations and other deverbal formations. The affixes -s-and -t-are then the default realizations of this aspectual head…" (Embick 2000: 216). On his analysis of the Future Active Participle, Asp is lower than -ūr (which he analyzes as a modal element of category Mod), so this participle straightforwardly obeys the MG: However, Embick does not specify what the relevant "aspectual properties" are, and it is not clear that all structures containing -t-/-s-really do have "aspectual properties"-or, at least, aspectual properties attributable to -t-/-s-. For example, occī-s-ūr-us '[that is] going to kill' and am-ā-t-ūr-us '[that is] going to love' seem to have nothing in common aspectually besides prospectiveness, which is contributed by -ūr, not -t-/-s-. Turning to nominalizations, occī-s-or 'slayer' and am-ā-t-or 'lover' also seem to have little in common aspectually; likewise for occī-s-iō 'murder' and amā-t-iō 'love, caressing'. If the claim is that, in deverbal environments, -t-/-s-can realize any Asp head regardless of its features or denotation, then the link between the exponent -t-/-s-and the head Asp is considerably weakened. Furthermore, even if deverbal structures with -t-/-s-do turn out to invariably have aspectual properties, and even if these are contributed by a head Asp, it will not follow that -t-/-s-is Asp, and Embick does not argue further that it is. In view of all this, the claim that -t-/-s-is Asp strikes us as undermotivated. The same remarks apply to Calabrese's (2019, §3.4.4.1.5) analysis, which also incorporates the claim that -t-/-s-is an Asp head below -ūr (and that there is thus no anti-mirroring here); the morphological structure he posits is shown in (ii) (adapted from his (389)  The theoretical issues raised by -t-/-s-(including whether all its uses should be unified) deserve deeper investigation than space permits here; our point here, though, is that the -t-/-s-in the Perfect Passive and Future Active Participles is plausibly a categorizing a head, and if so, the hypothesis that Latin makes use of vP-movement makes available an elegant explanation for the apparent anti-mirroring in the Future Active Participle.
'[that is] going to praise' ( (27)). If, as argued in Section 3.1, Voice bears Person and Number probes in finite clauses, an account is needed of why it does not expone φ-features in nonfinite forms like those just mentioned. There are at least two possible accounts of this that are compatible with the XP/X 0 Analysis. One possibility, mentioned in fn. 17, is that Voice is simply absent from these participles. (If so, it must be possible for a head in the middle of a functional sequence to sometimes not be merged in, as argued by Rizzi 1997: 314-315;Starke 2004, § §2.1-2.2;Bošković 2016: 42;and Erlewine 2016: 475.) Alternatively, this may be a matter of Vocabulary Insertion. In particular, it may be that Voice does successfully probe for φ-features in nonfinite forms, but it is simply a fact of Latin morphology that the Voice node, φ-features and all, is realized as null in these environments. Vocabulary Items that would yield this result for the Perfect Passive and Future Active Participles are given in (28). The conditioning elements to the right of the slashes are stretches (Ostrove 2018; see Section 3.4). 22 As long as insertion of (28-a) precedes the Fission that Voice PASS sometimes undergoes in other contexts (Section 3), Fission, and hence φ-exponence, will be bled, as desired.
The choice between the two accounts just laid out is left open here. One final remark is in order. Consider the following infinitives: (29) a. laud-ā-r-e praise-th-prs.inf-act 'to praise' b. laud-ā-r-ī praise-th-prs.inf-pass 'to be praised' Here, as in the participles, Voice does not overtly expone φ-features. But here it cannot be structurally absent, because it is overt. The following Vocabulary Items are therefore needed: Asp PFV and a do not belong to the same extended projection, since a starts a new one. Therefore, if Asp PFV is projected in Perfect Passive Participles, and consequently mentioned in (28-a), the definition of stretch ((18)) will have to be revised, dropping the requirement that the elements of a stretch belong to the same extended projection, a possibility that Ostrove (2018Ostrove ( : 1281 himself considers. This may be necessary anyway (if stretches are the targets for Vocabulary Insertion), since in some English idiolects, including the second author's, me + 's can surface as my even when me is a final conjunct: %the principal and my boxes (= '[the principal and me]'s boxes'). Here, the terminals elsewhere realized as me and 's are adjacent, but belong to different extended projections; thus, they will form a stretch (= Vocabulary Insertion target) only if stretches can straddle extended projection boundaries. (Such a revision-like stretching more generallymay run into overgeneration issues, an important topic for future work.) However, if Asp PFV is not projected in Perfect Passive Participles (and hence not mentioned in (28-a)), (28-a) will not require a revised definition of stretch.
Here too, as long as insertion of (30-b) precedes the Fission that Voice PASS sometimes undergoes, Fission, and hence φ-exponence, will be bled. 23 Naturally, this discussion raises questions that merit more detailed investigation. What is important here, however, is that although Voice in Latin was argued in Section 3 to bear φ-probes, it is perfectly possible on the XP/X 0 Analysis to capture the fact that Voice does not overtly expone φ-features in nonfinite forms-particularly on Ostrove's (2018) stretching-based approach to Vocabulary Insertion or some similar approach.

Syntactic predictions and consequences of the analysis
The previous sections developed an analysis of Latin verbal morphology that is highly syntactic. On this analysis, Latin verb forms are built largely by narrow-syntactic operations-including, notably, XP-movement.
It is therefore crucial to determine what predictions the XP/X 0 Analysis makes about other aspects of Latin syntax, and whether these are correct. That is the objective of this section. Section 5.1 lays out and confirms the prediction that the verb word should move as an XP, skipping intervening heads, and not as a head. Section 5.2 lays out and confirms the prediction that the putative vP-movement to [Spec,TP] should be able to carry along vP-adjuncts. Finally, Section 5.3 considers how the XP/X 0 Analysis interacts with what is independently known about leftward argument movement out of vP in Latin.
It is possible that (some versions of) an alternative analysis-particularly an analysis on which the Latin verb word is a complex head, henceforth the Complex-Head Analysiscould also account for some or all of the syntactic observations presented in this section. The investigation of this possibility and a detailed comparison between the Complex-Head Analysis and the XP/X 0 Analysis are tasks that, for space reasons, must be left for future work. The discussion in this section will restrict itself to showing that the XP/X 0 Analysis leads to the expectations about syntactic phenomena mentioned above, and that those expectations are indeed met. (Section 6, however, will compare the XP/X 0 Analysis and two existing versions of the Complex-Head Analysis that aim to account for the antimirroring in passive forms, arguing in favor of the former.)

Prediction A: The verb word should move as an XP, not as a head
On the XP/X 0 Analysis, the Latin verb word is a sequence of morphemes-most of them phonologically dependent-that are suspended across a large syntactic expanse.
As mentioned in Section 3.3, if there are any overt arguments or adjuncts within vP that do not move out of TP, then, on the XP/X 0 Analysis, the verb word is not even a constituent-as in (31)  intensely son.gen death.acc grieve-th-impf.sjv-3sg 'when he was intensely grieving over the death of his son' 23 A drawback of this analysis is that the Vocabulary Items in (28) and (30) are somewhat stipulative. As a reviewer notes, however, there may be an alternative analysis that overcomes this issue to some degree. On this analysis, the φ-probes are not on Voice but on an Agr head that takes TP as a complement. If TP is linearized to the left of Agr, or comes to immediately precede Agr via movement, then Voice and Agr will be adjacent, and-on the stretching-based approach to Vocabulary Insertion-can therefore interact in the necessary ways (e.g., the Voice-Agr stretch can be realized by the portmanteau -minī in 2pl passive forms). If this Agr head is missing in nonfinite environments, the φ-lessness observed in these environments follows.
The detailed elaboration of this intriguing alternative, and a thorough comparison with the analysis in the text, must be left for future work.
In (31), the verb word maerēret 'was grieving' is not a constituent; rather, [ vP fīliī mortem maer-ē-] 'grieve over the death of his son' is. However, it is also possible in Latin for the vP to contain no overt material other than V and v; this happens whenever (a) no other material is generated within vP, or (b) some other material is, but it is null (as in the case of null pronouns), or (c) overt material other than V and v is generated within vP, but it all scrambles out of TP.
In any of these situations, the string referred to here as the verb word is the reflex of a constituent, albeit a very large one: TP. An example (adapted from Danckaert 2012: 12) is given below: On the XP/X 0 Analysis, then, when the Latin verb word is a constituent, it is a maximal projection (an XP-specifically, a TP); it is never a complex head. This leads to the following prediction: (33) Prediction A 1 : XP-movement The verb word should be able to move as an XP, skipping intervening heads-i.e., movement of the verb word should not be subject to the Head Movement Constraint (HMC, Travis 1984).
Sections 5.1.1-5.1.2 will show that the finite verb can move to the left of (thereby skipping) the left-peripheral heads ut 'so that', cum 'when', sī 'if', and =ne 'ynq', bearing out the prediction in (33). 24 Section 5.1.3 will argue that, by contrast, the verb word cannot move as a head.

Test cases A 1(a-c) : Movement past intervening Fin heads
The complementizers ut 'so that', cum 'when' (among other meanings), and sī 'if' are argued convincingly by Danckaert (2012: 107) to be Fin(iteness) heads in a Rizzi (1997)style left periphery: 24 The prediction in (33) is based on the traditional assumption that phrasal movement, but not head movement, can skip intervening heads. This assumption has been questioned, however. Harizanov & Gribanova (2019) argue that what has traditionally been known as head movement is in fact two distinct phenomena: genuine syntactic head movement and postsyntactic amalgamation. On their analysis, syntactic head movement is subject to exactly the same locality constraints as XP-movement, and is thus free to skip intervening heads (it is not subject to the HMC  (36) Ovid, Ars Amatoria, I. 99 Spect-ā-tu-m veni-unt, veni-unt [ spect-e-nt-ur ut watch-th-sup-acc come-3pl, come-3pl [ watch-prs.sjv-3pl-pass so.that ipsae ] … self.f.nom.pl ] 'They come to watch, they come to be watched themselves…' ('…they come so that they may themselves be watched…') On the XP/X 0 Analysis, whenever plōret 'she may cry' or spectentur 'they may be watched' is a constituent, it is a TP, so it is unsurprising that these verbs can move as XPs, skipping the intervening head ut. 25 For explicitness, the derivation of the bracketed clause in (36) is briefly laid out below. First, the following structure is built: 25 Such derivations, in which the verb word moves to the left of Fin, are made use of only in poetry, not in prose (Danckaert 2012: 275;2017: 75). This does not threaten the argument, however. Danckaert (2012: 275), citing Schünke (1906), reports that "the fronting of a finite verb [past Fin -Authors] is well attested with numerous authors of poetic texts, and occurs in various periods. Moreover, it is compatible with diverse types of metrical patterns and with different types of genres." This virtually eliminates the possibility that the relevant structures are actually unacceptable (*). Furthermore, we find a sharp contrast between, e.g., (36) (which strikes us as acceptable, if somewhat marked) and the analogous English sentences in (i-c)-(i-f), which are completely unacceptable even in poetry (cf. the controls in (i-a)-(i-b)).
(Native English speakers could choose to use (i-c)-(i-f) in poetry, but they would be deliberately deploying ill-formed structures.) These crosslinguistic contrasts would be puzzling if (36) were as underivable as (i-c)-(i-f), but fall out from our analysis, on which (i-c)-(i-f) are underivable but (36)  Crucially, we assume that (non-avant-garde) poetry in a language can differ syntactically from prose in that language only in ways in which (prose registers of) different languages can differ from one another-i.e., only in ways that the theory of syntax independently allows-clearly the most restrictive and hence most interesting hypothesis. On the XP/X 0 Analysis, when the verb tetigit 'he (has) touched' is a constituent, it is a TP, so it is unsurprising that it can cross an intervening Fin head.
In (35) In (40), Iuppiter 'Jove' and annōs 'years' scramble out of TP, but two other nonverbal elements remain within TP: mihi 'to me' and praeteritōs 'past'. Consequently, when the TP moves past sī 'if' (Fin), it carries along not only the verb word but also the latter two phrases, revealing clearly the phrasal nature of this movement. 27 A defender of a Long Head Movement (LHM) analysis of (35)-(39) (cf. fn. 24) might argue that, in (40), referat 'would bring back' undergoes LHM past Fin, and mihi 'to me' and praeteritōs 'past' move to the left periphery independently. But although a topicalization analysis might be plausible for mihi (on the grounds that 1sg pronouns are always discourse-given), there appears to be no motivation for claiming that praeteritōs 'past' in (40) moves to one of the (information-structure-related) positions in the left periphery. It is unlikely to be in [Spec,TopP] or [Spec,FocP], since it is not topical (in fact, (40) is discourseinitial, being the very beginning of a speech) and does not appear to be focused either. 28 Finally, consider (41)-(42) (adapted from Schünke 1906: 75, 72)-two more examples in which TP-movement to the left of Fin carries along nonverbal material. In (41), TP-movement past sī 'if' carries along the internal argument aulaea 'the canopy'. In (42), TP-movement past quoniam 'since' (historically cum/quom 'when, since' + iam 'now, already') carries 27 The … indicates not that overt material was omitted from (40) (none was) but that we are remaining agnostic as to where exactly in the left periphery the TP moves in (40). 28 The LHM analyst might concede this point but respond that, on the analysis in (40) In both examples, the additional material carried along (aulaea, īnsānīre) is discourse-new and apparently neither topical nor focal. This strongly suggests that, in (41)-(42), the verb word is not a complex head that undergoes LHM (with the other fronted material moving to [Spec,TopP] or [Spec,FocP]), but rather, everything to the left of Fin gets there in a single step of TP-movement.
Recapitulating, our claim that movement of the verb word alone past Fin ((35)- (39)) is in fact TP-movement is lent plausibility by (40)-(42), in which the phrasal nature of the movement past Fin is particularly clear.

Test case A 1(d) : Movement past the intervening head =ne
Another test case is provided by the left-peripheral element =ne, which heads matrix and embedded yes/no questions. (We will not determine the precise categorial identity of =ne, which is not directly relevant here.) Consider (43) (adapted from Pinkster 2015: 324): (43) Cicero, Tusculanae Disputationes 5.21 Contrā Brūtum=ne mē dic-t-ūr-um put-ā-s? against Brutus.acc=ynq me.acc speak-ptcp-prosp-m.acc.sg think-th-2sg 'Do you think I will contradict Brutus?' The head =ne makes available an XP-position-specifically, a specifier position, [Spec,neP]-which in (43) is occupied by the moved PP contrā Brūtum 'against Brutus'. That this position is occupied by a full-fledged PP in (43) provides evidence that it is an XP-position rather than a head position. (But see fn. 30 on some less straightforward examples and a possible alternative analysis.) On the XP/X 0 Analysis, the verb word, when it is a constituent, is an XP (a TP). Thus, the XP/X 0 Analysis leads us to expect that the verb word (= TP) should perhaps be able to move to [Spec,neP] (a phrasal position) and, consequently, immediately precede =ne.
This expectation is met. Sentences in which =ne is immediately preceded by the verb word are amply attested, an example being (44) (adapted from Pinkster 2015: 324): (44) Cicero, 2nd Philippic, 2.45 Scī-s=ne mē dē rēbus mihi nōtissimīs know-2sg=ynq me.acc about things.abl me.dat extremely.well.known.abl dīc-e-r-e? speak-th-prs.inf-act 'Do you know that I am speaking of matters with which I am fully familiar?' On the XP/X 0 Analysis, scīs 'you know' in (44) moves as a TP, so there is no obstacle to its crossing the intervening head =ne.
Someone who was skeptical about the XP/X 0 Analysis might claim that the verb word undergoes head movement to =ne in (44) rather than XP-movement across =ne. However, this would require an additional stipulation, namely that =ne determines not only a landing site for phrasal movement, but also a landing site for head movement. This stipulation is avoided on the XP/X 0 Analysis, which unifies verb + =ne and XP + =ne sentences: the former are analyzed as a subcase of the latter, and thereby reduced to an independently observable phenomenon. (See Lee 2006: 57-58 for a similar argument from San Lucas Quiaviní Zapotec.) In (44), the (infinitival) clausal complement of V scrambles out of (or is extraposed from) TP before TP moves to [Spec,neP]. Consequently, the only overt material inside TP is the verb word. 29 But if, as argued here, Latin allows TP-movement to [Spec,neP], this process should be able to carry along nonverbal material. This expectation is met, as shown by (45) (adapted from Devine & Stephens 2006: 237), which on our analysis is derived as in (46). (The preceding context for (45) is 'And so that they will not be taken by surprise by me all of a sudden, these are the things I will ask'.)  (45), nāvem 'a ship' and populō Rōmānō 'to the Roman people' are topicalized independently (i.e., they do not analyze nāvem populō Rōmānō dēbeant 'they owe a ship to the Roman people' as a constituent in (45)). However, they do not argue for this view, and it would be difficult to: all the overt constituents in (45) convey discourse-new information, and more generally, it is hard to see, given the preceding context, how one could motivate the claim that the two internal arguments undergo movement related to topicality. On the XP/X 0 Analysis, by contrast, the fact that these constituents precede the verb + =ne sequence in (45) receives a straightforward explanation: they do not scramble out of TP, so they are carried along when TP moves to [Spec,neP]. 30 29 It might be objected that it is a weakness of the XP/X 0 Analysis that its advocates must posit scrambling out of TP (prior to remnant TP movement) in (44), (35), and (39). This objection has little force, however, because scrambling is a prominent and amply attested feature of Latin syntax (see fn. 32 and Section 5.3), so it is hardly a liability for the XP/X 0 Analysis that its defenders must posit scrambling in certain sentences. Crucially, too, there are sentences in which (on the XP/X 0 Analysis) little or nothing scrambles out of TP before TP moves to the left periphery, so the phrasal nature of the movement can be seen clearly ((40)-(42), (45)). 30 The sentences containing =ne 'ynq' that are most straightforward syntactically are those in which the element that moves to [Spec,neP] immediately precedes =ne in its entirety (e.g., (43)). However, the fronted element can also surround =ne, as in (i)  Because =ne presumably cannot move into its own specifier, and there is no independent motivation for claiming that it undergoes sideward head movement into the moved XP before that XP is merged in [Spec,neP], we take it that sentences like (i) are derived by postsyntactic encliticization of =ne to the first morphophonological word in the fronted XP in [Spec,neP].
Since the placement of =ne is at least partially sensitive to prosodic constituency, one might try to account for it largely or entirely in terms of prosodic constituency. We cannot construct and evaluate such an alternative analysis here, but if one turns out to be possible, that may undermine the argument from =ne that the Latin verb word can only move as a phrase-though not our other arguments to this effect.

The inability of the verb word to move as a head
On the XP/X 0 Analysis, the verb word is often not a constituent. However, if there is no other overt material within TP, then the verb word is the reflex of a (phrasal) constituent-TP-and can therefore be expected to undergo XP-movement. All of Section 5.1 so far has argued that this expectation is met.
However, although the verb word sometimes corresponds to a phrasal constituent (TP), no operations can assemble its parts into a (complex) head on the XP/X 0 Analysis-which therefore makes the following prediction: (47) Prediction A 2 : No head movement The Latin verb word should be unable to move as a head.
This prediction is testable. As foreshadowed in (32), and discussed further in Section 5.3, we take nōn 'not' to be a Σ head (Laka 1990), and, crucially, to be-when it is projectedthe lowest head above the verb word. (For detailed empirical arguments that nōn is a head rather than a specifier, see If the verb word can undergo head movement to nōn, (49) should also be acceptable. (As is standard, we expect that, if the verb word can undergo head movement to nōn, it will be linearized to the left of nōn within the resulting complex head.)

Prediction B: The putative vP-movement should carry along vP-adjuncts
On the XP/X 0 Analysis, a central part of the derivation of Latin verb forms is vP-movement to [Spec,TP]-which gives rise to the following prediction: 31 (50)

Prediction B: vP-adjuncts
The putative vP-movement should be able to carry along an adverbial left-adjoined to vP-i.e., it should be possible for a vP-adverbial to immediately precede the core vP in surface syntax. 31 31 We do not claim that only the XP/X 0 Analysis gives rise to this expectation about constituent order. Bailey (2010), without drawing firm conclusions, considers the possibility that a synthetic finite verb in Latin surfaces in T (owing to head movement), and the remnant vP moves to [Spec,TP], deriving the default SOV order. On this analysis too, the putative vP-movement should be able to carry along a vP-adjunct. (As shown in Section 6, though, Complex-Head Analyses like this struggle to explain anti-mirroring in passive forms without ad hoc stipulations.) The goal of this section, then, is to show that a particular expectation that the XP/X 0 Analysis gives rise to ((50)) is met (i.e., the analysis is consistent with certain observable facts about Latin constituent order), not to argue that only the XP/X 0 Analysis can account for the relevant observations.
The reason for this is straightforward. If, as the XP/X 0 Analysis contends, vP obligatorily moves to [Spec,TP] in Latin (when T is in the structure), this operation should be equally applicable when the maximal vP is formed by adjunction of an adverbial to a smaller vP containing all the arguments. (It is this latter, smaller vP that we call the core vP.) The expectation in (50) is met, as shown in (51)  ? that.acc seek-th-pfv-pst-3pl '…why didn't they go to that region just as [they did] then, directly and openly?' 32 It might be objected that, Latin word order being flexible, the constituent order in (51) reveals little about its derivation. But this would be conceptually and empirically misguided. Conceptually, "flexible word order" is an observation, not an explanation. Thus, the acceptability of (51) is not explained by saying that Latin has "flexible word order"; rather, the observation that Latin has "flexible word order" is to be explained by a principled analysis of the mechanisms giving rise to it, as is well appreciated (Walker 1918;Marouzeau 1922;Ostafin 1986;Polo 2005;Salvi 2005;Devine & Stephens 2006;Bailey 2010;Longobardi & Roberts 2010, §3;Spevak 2010;Danckaert 2012;2017a;b;Gianollo 2016, a.m.o.). Empirically, Devine & Stephens (2006) argue convincingly that Latin word order is not "just random" (as was already noted in Marouzeau 1922: 1), but rather "a default order clearly emerges from the data," namely (i) (adapted from Devine & Stephens 2006: 79)-departures from which are derived by movement, which often has information-structural effects. This strongly suggests that the approach to Latin word order argued for above on conceptual grounds is the right one.
(i) subject > direct object > indirect object/oblique argument > adjunct > goal/source phrase > nonreferential direct object > verb Questions arise here as to how particular constituent orders are derived on the XP/X 0 Analysis. For example, in analytic passives (fn. 19), the order Perfect Passive Participle > internal argument > esse 'be' is possible (Danckaert 2014): [C] Laudāta Anna est ('having.been.praised  In (iii), the internal argument Ana moves out of a participial projection, which subsequently undergoes remnant movement. (The empty category ______2 must result from movement, not ellipsis, since Spanish lacks post-auxiliary verb phrase ellipsis -Zagona 1988: 162.) We do not claim that leftward movement of remnant participial AgrPs in Latin is necessarily topicalization, as the remnant movement in (iii) is, but only that the general type of derivation proposed here for [C] Laudāta Anna est is independently (crosslinguistically) motivated. Determining the precise nature and landing sites of the two movements posited here for [C] Laudāta Anna est must, for space reasons, be left for future work.
Since one of the fundamental properties of adjunction is iterability (Hornstein & Nunes 2008: 60), it should be possible to adjoin adverbials to vP iteratively, and the XP/X 0 Analysis predicts that the largest vP formed by iterated adjunction should be movable to [Spec,TP] as a unit. When this happens, the core vP should be immediately preceded by a sequence of two or more vP-adverbials. This is indeed possible, as shown in (53)  . attribute-th-perf.sjv-3sg '…that he credited me in the Senate, not once but often and at length, with having saved the empire and the world.' Our claim that the adjuncts in (51)-(53)  Summarizing, the XP/X 0 Analysis leads us to expect that it should be possible for the core vP to be immediately preceded by one or more vP-adverbials in surface syntax, and this expectation is met. 33

A note on leftward argument movement out of vP
Section 5.2 showed that, as predicted by the XP/X 0 Analysis, it is possible for one or more vP-adverbials to immediately precede the core vP, with the latter containing all the (overt) arguments. However, arguments can also precede vP-adverbials. In (55) (adapted from Devine & Stephens 2006: 104), the object mortem hominis necessārī 'the death of a close friend' precedes the vP-adverb graviter 'grievously': Cicero, Epistulae ad Familiares 11.28.2 quod mortem hominis necessārī graviter fer-ō because death.acc person.gen close.gen grievously bear-1sg 'because I take the death of a close friend grievously' Likewise, in (56) (adapted from Devine & Stephens 2006: 102), the object cohortēs 'the cohorts' and the PP argument ex castrīs 'out of the camp' precede the vP-adverb celeriter 'quickly': Caesar, De Bello Gallico 7.49 ut cohortēs ex castrīs celeriter ēdūc-e-re-t that cohorts.acc out.of camp.abl quickly lead.out-th-impf.sjv-3sg 'that he should lead the cohorts out of the camp quickly' The XP/X 0 Analysis, then, requires that (57) hold: (57) In Latin, arguments generated within vP can move leftward out of vP (thereby coming to precede left-adjuncts to vP).
But the hypothesis in (57) 34 is very well motivated, and on grounds independent of the relative ordering of arguments and vP-adjuncts. Consider the relative ordering of arguments and the particle nōn 'not', which we analyze as a Σ head (Laka 1990;cf. Danckaert 2017: 51ff.). The arguments can follow nōn, as in (58)  But the arguments can also precede nōn, as in (59)  This flexibility of argument positioning vis-à-vis nōn is standardly interpreted as indicating that arguments in Latin are able, though not required, to move leftward out of vP (Devine & Stephens 2006: 88-89). 35 The flexible positioning of arguments with respect to 34 We take argument movement out of a moved vP to be unproblematic in principle, since movement out of moved constituents is sometimes possible: (i) [Which building] 2 did it seem that [on pictures of ______2 ] 1 , she had relied ______1 , but [on models of ______2 ], she hadn't?
In (i), both instances of of are stranded (suggesting that which building moves across the board from their complement positions), and both instances of on are lexically selected by rely, demonstrating that the on-PPs move from within the corresponding VPs. 35 We argue, then, that the subject, like other arguments, can scramble out of the moved vP in [Spec,TP].
We assume, as is standard in the Latin generative literature, that the subject does not obligatorily move to [Spec,TP]. There is thus no need to posit, problematically, that T can attract the subject from within the vP in its own specifier (and into a second specifier of TP vP-adverbials is standardly taken as evidence for the same conclusion (Devine & Stephens 2006, §1.6). 36 In permitting arguments to optionally move leftward out of vP, then ((57)), the XP/X 0 Analysis falls fully in line with the current understanding of what Latin syntax allows. 37 6 Comparison with other analyses of anti-mirroring forms As alluded to above, a significant advantage of the XP/X 0 Analysis is that it accounts for anti-mirroring forms such as passive subjunctives (e.g., laud-ē-t-ur '[that] he/she/it may be praised'), with their apparently peculiar morpheme ordering, without any stipulations specific to these forms. This is because, on the XP/X 0 Analysis, anti-mirroring forms have exactly the same syntactic derivation as mirroring forms.
This section considers two other analyses of anti-mirroring in Latin passive verbs (Embick 2000 andCalabrese 2019) and shows that they, by contrast, are forced to make passive-specific stipulations to account for the anti-mirroring in these forms-providing an argument against these analyses and in favor of the XP/X 0 Analysis. Embick (2000: 196-199) assumes that the Latin verb word is assembled by head movement. On his analysis, there is successive head movement from the root to v to Asp to T, producing a complex T head. In the (postsyntactic) morphological component, an Agr(eement) head is added within this complex T head, yielding (61): It might be objected that the XP/X 0 Analysis posits vP-movement to [Spec,TP], which could also be considered a stipulation. Crucially, however, this vP-movement is not claimed to occur only in anti-mirroring passive forms, nor is it posited exclusively to derive the anti-mirroring in passive forms: rather, it plays a key role in deriving the morpheme orders in antimirroring and mirroring forms alike. The empirical benefits of the proposed vP-movement, then, are considerably wider-ranging than those of Embick's morphological operations-not to mention that the XP/X 0 Analysis makes correct syntactic predictions that are not shared by Embick's (largely postsyntactic) analysis without further assumptions (Section 5). 41 Calabrese (2019, §3.4.4.2) also develops an analysis of the anti-mirroring in passive forms.

Calabrese (2019)
Calabrese proposes that the passive morpheme (what was analyzed in Section 3.2 as a Voice PASS head) is actually a reflexive-like clitic ("D R ") originating in [Spec,VoiceP] (in passives of transitives) or in direct object position (in the case of unaccusatives with passive/nonactive morphology-i.e., deponent unaccusatives). On his analysis, the verb word is assembled largely by head movement, which causes it to surface in Mood, but also partially by certain postsyntactic operations that are not directly relevant here.

(63)
Derivation of passive anti-mirroring according to Calabrese (2019) 41 Embick (2000: 201) actually ends up proposing that synthetic finite verbs not only are complex heads but in fact surface in Asp; on this analysis, deriving constituent orders such as low adverb > object > verb word (Section 5.2) would presumably require remnant vP-movement to [Spec,AspP] or [Spec,TP]-bringing the analysis closer to the XP/X 0 Analysis, but complicating it further in the process. 42 Our (63) omits structure and operations that are not directly relevant here. For clarity, we have added a 0 to the label of every X 0 category in (63) (except D R , which is both minimal and maximal).
It might be objected that this derivation is rather complex, since D R moves three times. However, the first two movements may be independently motivated, since they are modeled after analogous movements that, according to Calabrese, the Italian clitic si undergoes. (Calabrese's motivation for giving a unified analysis of D R and si is that Latin passive/nonactive clauses and Italian si-constructions have a similar range of interpretations.) Evaluating Calabrese's analysis of si is beyond the scope of this paper; what is important here is the last movement step in (63).
If D R underwent only the first two movements in (63), it would surface at the left edge of the verb word, whereas in fact it surfaces at the right edge. To solve this problem, as mentioned, Calabrese posits that it is an idiosyncratic property of D R that it adjoins to Agr. This analysis, then-like Embick's-can derive the morpheme order in anti-mirroring passive forms only by making at least one stipulation specific to the passive morpheme.

Summary
Both Embick's (2000) and Calabrese's (2019) analyses resort to stipulations specific to the passive morpheme to derive the morpheme order in anti-mirroring passive forms.
The XP/X 0 Analysis, by contrast, accounts for this anti-mirroring order with no passivespecific stipulations whatsoever (Section 3). On the XP/X 0 Analysis, the anti-mirroring order in these forms simply follows from their syntactic derivation, which is shared with all other synthetic finite verb forms in Latin, including those whose morpheme order is mirroring rather than anti-mirroring. This provides a significant argument for the XP/X 0 Analysis and against the two alternatives just considered.

Conclusion
As we have seen, Latin verbs present a morpheme-ordering puzzle: they seem to both obey and disobey the Mirror Generalization (MG), albeit in different inflectional forms (Cinque 1999: 197, citing Calabrese 1985. Given the theoretical and diagnostic significance of the MG, this situation merited careful investigation to determine whether it indicated that the MG was fundamentally on the wrong track. The result that has emerged is that, contrary to initial appearances, the MG is in fact valid for Latin-but it should be understood as a generalization about structures formed by operations on heads. Head movement, being extremely local, readily gives rise (in conjunction with other operations) to Latin verb forms displaying morpheme orders that straightforwardly obey the MG. Phrasal movement, by contrast, is subject to much less strict locality conditions than head movement (or at least the canonical head movement at play in the relevant Latin verb forms), and can consequently bring about apparent violations of the MG. This, we have argued, is exactly what happens in apparently antimirroring Latin verb forms.
Our analysis of anti-mirroring in Latin lends plausibility to the notion that other instances of anti-mirroring crosslinguistically, including instances not yet discovered, also owe their existence to phrasal movement. However, the analysis does not entail that phrasal movement is the only possible culprit for anti-mirroring in any language. Whether a particular instance of anti-mirroring is due to syntactic XP-movement, to postsyntactic operations, or perhaps even to something else is an empirical question.
On the XP/X 0 Analysis, every synthetic finite verb form in Latin, whether it appears to obey the MG or not, has exactly the same derivation 43 -one involving one step of head movement (Asp to T) and one step of phrasal movement (vP to [Spec,TP]).
The analysis satisfies the theoretical desiderata in (12). It is uniform, in that all the synthetic finite verb forms share a single derivation. 44 It is general, in that it extends readily to nonfinite and even deverbal forms (see the Appendix). (In particular, the hypothesis that Latin makes use of vP-movement allows us to make sense of an otherwise seemingly paradoxical instance of anti-mirroring that arises among the participles.) Finally, the analysis has good syntactic fit: it correctly predicts that the verb word, when it moves, should move as a phrase; it makes correct predictions about the placement of vP-adverbials; and it is fully compatible with what is independently known about leftward argument movement out of vP in Latin.
A perhaps surprising consequence of the analysis is that the Latin verb word, though a phonological word, is not always a syntactic constituent. (And when it is a constituent-because there is no overt nonverbal material within vP-it is the reflex of a large phrasal constituent, TP, hence why it moves as a phrase.) The larger picture that emerges from this squishing-based or alexist analysis is one in which phonological words need not always correspond to syntactic constituents, and can in fact mismatch them dramatically. And whether a particular phonological word corresponds to a syntactic constituent or not, it is entirely possible that it is not the reflex of a complex head, but of a linearly contiguous series of morphemes suspended across a potentially vast region of syntactic space Abbreviations abl = ablative, acc = accusative, act = active, arg = argument, [C] = constructed example, dat = dative, EA = external argument, f = feminine, fut = future, gen = genitive, IA = internal argument, impf = imperfect "tense" (= past tense + imperfective aspect), ind = indicative, lit. = literal translation, m = masculine, n = neuter, Neg = negative particle, nom = nominative, O = object, pass = passive, perf = perfect "tense" (perhaps = present tense + perfective aspect), pfv = perfective aspect, pl = plural, plup = pluperfect "tense" (= past tense + perfective aspect), prosp = prospective aspect, prs = present, pst = past, ptcp = participle, S = subject, sg = singular, sjv = subjunctive, sup = supine, superl = superlative, th = theme vowel, ynq = yes/no question particle, 1/2/3 = first/second/third person

Additional File
The additional file for this article can be found as follows: • Supplementary file 1. Appendix to 'XP-and X 0 -movement in the Latin verb: Evidence from mirroring and anti-mirroring'. DOI: https://doi.org/10.5334/gjgl.1049.s1

Funding Information
This material is based upon work supported by the National Science Foundation Graduate Research Fellowship Program under Grant No. DGE-1339067. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation.