Additional File

The additional file for this article can be found as follows:

Notes

1. See Bybee (1985) and Rice (2000) for related investigations. [^{^}]
2. We will refer to the root of the verb (i.e., the head of the complement of v) as V, but our analysis will be unaffected if “V” is actually a categoryless root (Marantz 1997). [^{^}]
3. Assuming iterated head-raising with left-adjunction (cf. Pollock 1989) or lowering with right-adjunction (cf. Embick & Noyer 2001). A wider range of possible morpheme orders is predicted if affix-specific linearization (Harley 2013) is possible. Even if it is, though, syntactically higher heads should always surface outside syntactically lower heads within the morphological word. [^{^}]
4. This question is of interest not only because apparent counterexamples to the MG need to be scrutinized, for the reasons discussed above, but also because the development of Latin into Romance provides an excellent laboratory in which to learn how morphosyntactic change happens (Ledgeway 2012a; b; Danckaert 2017a, a.m.o.)—but this requires an accurate understanding of the synchronic morphosyntax of the “start” and “end” states. [^{^}]
5. Embick (2000: 213–214) argues convincingly that -erā is not an incorporated form of esse ‘be’. [^{^}]
6. For Latin verb paradigms, see Gildersleeve & Lodge (2000) or any other standard Latin grammar (and the Appendix). Attested examples larger than a single verb form are drawn from the Latin linguistic literature. Also given below are a few supplementary constructed examples (marked ^[C]). [^{^}]
7. The 2SG form laudēris ‘(that) you may be praised’ has an alternate form laudēre. For convenience, every subjunctive verb is translated here as if it were the main predicate of a rationale clause (e.g., ‘so that I may be praised’), but the interpretation of Latin subjunctive verbs is context-dependent (Lakoff 1968, a.m.o.). [^{^}]
8. Following Heck & Müller (2007), we use the following feature notation: [F:□] = unvalued feature; [*F:□*] = unvalued feature that probes for a value; [*F: ] = 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•] (omitted in the trees; notation from Heck & Müller 2007), which attracts the closest element bearing the categorial feature [CAT:v] (i.e., vP) to [Spec,TP] (following the view that movement can be driven by a feature of a c-commanding head; see Chomsky 1995: 297; Lasnik 1995; McCloskey 2001, a.m.o.). This idea could perhaps be made somewhat more explanatory by positing that vP-movement is how Latin satisfies the EPP, thereby relating it to superficially different phenomena such as English subject movement (see Bailey 2010; Danckaert 2012; 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. [^{^}]
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-ā-v-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. [^{^}]
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/X⁰ 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. [^{^}]
17. This complement is shown in (22) as an AspP containing Asp and Voice heads, but little hinges on this; the complement could in principle be as small as VoiceP or vP (though, in the latter case, the complement of a would have to move to [Spec,aP]—i.e., complement-to-specifier movement would have to be possible, contra Abels 2003). See fn. 21 for more on our analysis of -t and a comparison with an alternative. [^{^}]
18. This portion of the structure is not compatible with the Final-Over-Final Condition (Sheehan, Biberauer, Roberts & Holmberg 2017). How problematic this is and whether it suggests that this part of (22) should be revised somehow are questions that, for space reasons, must be left for future work. [^{^}]
19. A question arises here as to how the auxiliary esse ‘be’ in analytic structures like (22) fits into the analysis. Since it agrees (e.g., est is 3SG), and on the XP/X⁰ Analysis, the clause-level ϕ-probes are on Voice, the natural analysis is (i), with es-, the auxiliary’s “root,” a V or v taking as its complement the participial AgrP (whose internal structure is given in (22)):

    (i)	[TP [vP [AgrP Anna laudāta] es-]2 Asp1+T[Tense:PRS, Mood:IND] ______1 -tVoiceACT ______2 ]

    In (i), the highest Voice head, which is in the extended projection of es- ‘be’, is active, and T is present, exactly as in (present-tense) copular clauses; the passive and (aspectually) anterior properties of (i) are contributed by the participle. A rough gloss would be ‘is [Anna having-been-praised].’ Two remarks are in order here. First, on the XP/X⁰ 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). But there is no incorrect prediction here, because the relevant categorial features—[CAT:a], [CAT:Agr], and [CAT:V] or [CAT:v] (for es- ‘be’)—are fundamentally syntactic features. Nothing in the theory forces them to yield a stative interpretation; rather, what denotations the lexical items in (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. [^{^}]
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:

    (i)

    [ModP -ūrMod [AspP -t-/-s-Asp [vP v [VP V DP]]]]

    (adapted from Embick 2000: 219)

    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) and (391–392)):

    (ii)

    [Adj0 [n0 [Asp0 [V (TV)] -tAsp0] -ūrn0] -uAdj0] …

    (TV = theme vowel)

    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. [^{^}]
22. The parentheses in (28-a) are shorthand. Asp_PFV should be mentioned in this Vocabulary Item iff it is projected in Perfect Passive Participles (this was left open in fn. 17). 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 (2018: 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 generally—may 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. [^{^}]
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. [^{^}]
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). Postsyntactic amalgamation, by contrast—which can be either Raising or Lowering—is much more local, being restricted to structurally adjacent heads. (For discussion, see Dékány 2018 and Newman 2018a; b.) The notion of relatively nonlocal head movement is somewhat controversial (see Matushansky 2006, §5.1.3). Thus, Arregi & Pietraszko (2018: 8ff.; to appear: 12ff.) argue that at least some cases of what Harizanov and Gribanova analyze as relatively nonlocal syntactic head movement are actually subject to strict, HMC-like locality, despite appearances. Angelopoulos & Sportiche (2018: 3, 32) take the strong view that all head movement is subject to the HMC. Likewise, Svenonius (2018: 10) “take[s] the HMC to be essentially empirically correct”; see also Buell, Sy & Torrence (2014: 77) for essentially the same view. This debate cannot be settled here, but if, as Harizanov and Gribanova argue, syntactic head movement exists and is no more local than XP-movement, then the Latin verb word’s ability to move past intervening heads will not decisively favor the XP/X⁰ Analysis over the Complex-Head Analysis, though it will certainly remain compatible with the former. See, though, (40) and (45) below, where we argue that the TP housing the verb word moves across an intervening head and carries along nonverbal XPs inside it (and thus must be moving as an XP rather than as a complex head): since these cases must involve TP-movement, it would be redundant to posit that the verb word is a complex head that can move relatively nonlocally. [^{^}]
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) is derivable. (For convincing evidence of the usefulness of crosslinguistic acceptability comparisons, see Bošković to appear, §5.)

    (i)	a.	They come [so that they may be watched themselves].
    	b.	They come [so that they may themselves be watched].
    	c.	*They come [they may be watched so that themselves].
    	d.	**They come [may be watched so that they themselves].
    	e.	*They come [they may be watched themselves so that].
    	f.	*They come [they may themselves be watched so that].

    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. That is, we adopt the Development Hypothesis (on which see Fabb 2010) for syntax. In this case, we propose that, in the poetic but not in the prose grammar of Latin, Top can optionally bear a movement-driving feature [•T•], which attracts TP to [Spec,TopP], a standard case of attraction-driven movement (fn. 11). The two grammars thus differ in what movement-driving features they allow to appear on Top, much as (on one attraction-driven approach to movement) T bears [•D•] obligatorily in English, but only optionally in Dholuo (Cable 2012). [^{^}]
26. Zyman (2019) argues for a generalization he dubs Phasal Antilocality: “For a phase YP, movement from within a constituent at the edge of YP must cross an XP dominating YP” (Zyman 2019: 2). He further argues that the set of phases comprises not only CP and vP but also TP. If all this is correct, then ipsae ‘themselves’ cannot move from within the vP in [Spec,TP] to [Spec,ScrP] as shown in (38), because this movement would not cross a maximal projection dominating the TP phase. Instead, it must be that (a) there is more structure between TP and Fin than (38) shows, and the scrambling of ipsae actually does cross some XP dominating the TP phase, or (b) ipsae, prior to vP-movement, undergoes intermediate movement to the TP phase edge (thereby circumventing Phasal Antilocality) before scrambling further. Which of these possibilities is correct, if either, is not directly relevant here. [^{^}]
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), it is not clear where the TP moves or what effect this movement has on information structure, if any. The point is well taken, but since FinP is left-headed in Latin, the conclusion that (a constituent containing) referat ‘would bring back’ undergoes some sort of movement to the left of Fin is virtually inescapable. This is not true of the undermotivated claim that praeteritōs ‘past’ in (40) moves to the left periphery independently—an alternative to which we have just laid out. [^{^}]
29. It might be objected that it is a weakness of the XP/X⁰ 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/X⁰ Analysis that its defenders must posit scrambling in certain sentences. Crucially, too, there are sentences in which (on the XP/X⁰ 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) (adapted from Devine & Stephens 2006: 236):
    1. (i)

    1. Cicero, In Verrem 2.4.101
    1. Iīs = ne
    2. those.DAT = YNQ
    1. rēbus
    2. things.DAT
    1. manūs
    2. hands.ACC
    1. adfer-r-e
    2. lay.on-PRS.INF-ACT
    1. nōn
    2. not
    1. dubit-ā-stī…?
    2. hesitate-TH-PERF.2SG
    1. ‘Did you not hesitate to lay your hands on those things…?’
    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. [^{^}]
31. We do not claim that only the XP/X⁰ 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/X⁰ 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/X⁰ Analysis can account for the relevant observations. [^{^}]
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; 2014; 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/X⁰ 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 Anna is’ = ‘Anna {was/has been} praised’). On the XP/X⁰ Analysis, the natural hypothesis here is that this order is derived from (ii) by leftward movement of Anna followed by leftward movement of the remnant participial AgrP [Anna laudāta].

    (ii)	[TP [vP [AgrP Anna laudāta] es-]2 Asp1+T[Tense:PRS, Mood:IND] ______1 -tVoiceACT ______2 ] (from fn. 19)

    This hypothesis is independently motivated by sentences like (iii) (judgment from Samuel Zyman, p.c.):
    1. (iii)

    1. Spanish
    1. [Alabada _{______1}
    2. [praised.F.SG
    1. en
    2. on
    1. la
    2. the
    1. tele]₂,
    2. TV
    1. Ana₁
    2. Anna
    1. no
    2. not
    1. fue _{______2}.
    2. was
    1. ‘Praised on TV, Anna wasn’t.’
    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. [^{^}]
33. No adverbial can intervene between Voice and the rest of the verb word, suggesting that there is no left-adjunction to VoiceP in Latin, but only to vP (e.g., for manner, instrumental, and comitative adjuncts), in keeping with a suggestion made on partially independent grounds in Section 4.1. [^{^}]
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). The important question of what drives scrambling in Latin must, for space reasons, be left for future work. [^{^}]
36. As noted by Danckaert (2017a: 38ff.), the highest, and only the highest, verbal element in a clause is unable to precede nōn ‘not’ (qua clausal negation), as shown in (i) (adapted from his (67b) and (72c)):
    1. (i)

    1. Gaius, Institutes 3.32
    1. Praetor
    2. praetor.NOM
    1. hērēdēs
    2. heirs.ACC
    1. fac-e-r-e
    2. make-TH-PRS.INF-ACT
    1. ⟨*potes-t⟩
    2. ⟨*can-3SG⟩
    1. nōn
    2. not
    1. ⟨potes-t⟩.
    2. ⟨can-3SG⟩
    1. ‘The praetor cannot appoint heirs.’
    On Danckaert’s analysis, this is because the verb word is a complex head and thus cannot cross the intervening head nōn, owing to the HMC. This explanation is unavailable on the XP/X⁰ Analysis, on which the verb word is not a complex head. There is, however, a straightforward way to capture this effect on the XP/X⁰ Analysis. Suppose, essentially following Grohmann (2003: 74ff.), that the highest projection in each of the three clausal domains (lexical, inflectional, and complementizer—Rizzi 1997: 281) functions as a phase (a dynamic phasehood analysis—Bobaljik & Wurmbrand 2005, a.m.o.), and that complement-to-specifier movement is banned by antilocality (Abels 2003). Then, when nōn (= Σ) is projected, the verb word (= TP complement of Σ) cannot move to [Spec,ΣP] because of antilocality, and cannot move higher in one fell swoop because ΣP (= the inflectional domain) is a phase. Thus, the verb word cannot come to precede nōn at all. This holds only for the highest verb/auxiliary, as desired: a lower verbal projection can move to [Spec,ΣP] (and possibly higher), since complement-to-specifier antilocality is irrelevant in this case. On this analysis, ordinary affirmative clauses (lacking nōn) lack Σ; in these clauses, the highest projection in the inflectional layer is TP, which can therefore move to the left of Fin (Section 5.1) unhindered by antilocality. We leave further testing of this analysis, and alternatives, for future research. [^{^}]
37. Relatedly, the XP/X⁰ Analysis makes a prediction about OV structures like (i) (adapted from Danckaert 2017a: 70). (We analyze the -r in present infinitives like facere ‘to make’ as a portmanteau of present T [which presumably bears a [–FIN] feature in this case] and Asp, and the final -e as an active Voice head; compare active laud-ā-r-e ‘to praise’ with passive laud-ā-r-ī ‘to be praised’. See the Appendix.)
    1. (i)

    1. Cicero, In Verrem 3.211
    1. [_TP [_vP
    2.  
    1. PRO
    2.  
    1. innumerābilem
    2. uncountable.F.ACC.SG
    1. pecūniam
    2. money.ACC
    1. fac-e]₁-r-e
    2. make-TH-PRS.INF-ACT
    1. _{______1} ]
    2.  
    1. ‘to make an enormous amount of money’
    Whenever the subject in a structure like (i) either is null (as in (i)) or scrambles out of TP, the object-plus-verb-word string should pass constituency tests, since, on the XP/X⁰ Analysis, it is the reflex of a constituent (TP). Danckaert (2017a: 68–73), who focuses on OV sequences in which the verb is infinitival (as in (i)), shows that this expectation is met: such OV sequences can be coordinated, preposed, elided, and pronominalized, and can antecede relative pronouns. [^{^}]
38. Embick uses the abbreviated notation “√-v-Th” (Th = theme vowel) rather than explicitly indicating the hierarchical structure he posits for this complex head; this abbreviation is reproduced in (61). [^{^}]
39. Embick exemplifies this structure with imperfect indicative passive forms like (i):
    1. (i)

    1. laud-ā-bā-t-ur
    2. praise-TH-IMPF-3SG-PASS
    1. ‘he/she/it was being praised’
    —hence the choice of imperfective Asp, past T, and 3SG features in (62). Our (62) reproduces Embick’s (22) exactly; we take it that his feature bundles [imperf], [past], and [3sg] are not to be interpreted as literally the daughters of Asp, T, and Agr, respectively. [^{^}]
40. To be fair, accounting for the anti-mirroring in passive forms is not Embick’s (2000) main goal; in fact, he states that “[t]he suggestion here merely highlights the problem, which may be resolved mechanically” (Embick 2000: 199). As argued above, though, the apparent need for a stipulative mechanical solution is an artifact of the assumption that the Latin verb word is built exclusively by operations on heads—an assumption that the XP/X⁰ Analysis profitably dispenses with. [^{^}]
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/X⁰ 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 X⁰ category in (63) (except D_R, which is both minimal and maximal). [^{^}]
43. Modulo certain minor postsyntactic idiosyncrasies (fn. 13). [^{^}]
44. Crucially, the vP-movement is not restricted to the anti-mirroring forms, so it is not a stipulation tailor-made to deliver the correct morpheme orders in just these forms: it plays a role in deriving the morpheme orders in the mirroring forms also. [^{^}]

Acknowledgements

References