The morphosyntax of the periphrastic future under negation in Cypriot Greek

In Cypriot Greek, the negated future is marked by an element, tha, which appears instead of the expected present tense copula and a selected subordinating element. This paper documents the distribution of this item for the first time, and presents an analysis in Distributed Morphology that analyzes tha as a portmanteau morpheme realizing two heads in the context of negation. This analysis requires that spans (or targets of Fusion) can include a verb and the head of its C complement. 1 The future and negation in Cypriot Greek 1.1 The periphrastic future: copula+na Cypriot Greek, like its Standard Modern Greek sister spoken primarily to its northwest across the Mediterranean, has a copula verb that inflects for person, number, and tense, but shows no number distinction in the third person: ∗Our thanks to the reviewers for the Journal of Greek Linguistics, to Karlos Arregi, Anastasia Giannakidou, Marika Lekakou, and Andrew Nevins for valuable feedback on the paper, as well as to audiences at the 12th International Conference on Greek Linguistics and at the University of Chicago, where earlier versions of this work were presented. All Cypriot Greek data reported in this paper without further attribution come from the second author, who is a native speaker from Limassol. Our thanks also to the ten additional speakers who provided judgments in a written questionnaire.

(1) Cypriot Greek copula ime 'be' Cypriot Greek has a two-way tense distinction morphologically represented: past and non-past (see Holton et al. 1997 for more detail on the identical standard Greek system), and two aspects: imperfective and perfective.It marks the future2 periphrastically, using a construction that the (here invariant) 3rd person copula with a clause headed by the subordinating particle na3 (traditionally called 'subjunctive', and which we will assume is a complementizer for the time being) and a verb in the non-past: (3) En be.NONPAST.3na na pao.go.PERF.NONPAST.1sg'I will go.' The particle na is found in a range of contexts, including as the head of selected control and raising CPs; Cypriot Greek, like standard Greek, lacks infinitives, and uses na-clauses in their stead: (4)  (Terzi 1999b:230) As also pointed out in Roussou and Tsangalidis (2010), it is unclear whether en na is a single lexical item or not.We claim that each of the elements in en na has an independent use: en as a verb4 and na as a (clausal) subordinator.The null hypothesis for the syntactic structure of the periphrastic future which combines these two elements, therefore, is to put them together in the usual way, with the verb en selecting a CP headed by na, as illustrated in the following simplified structure (where we suppress the representation of verb movement to T, as well as of other elements possibly present in the clause): The na that occurs in this location has the usual properties found with naclauses elsewhere in the language.First, it can select for what we follow Holton et al. 1997 in calling the dependent form of the verb; the dependent verb form is the perfective nonpast, and it is always selected for-it occurs only under certain particles, like na, and in certain embedded constructions (see Giannakidou 2009 for the claim that the dependent verbal form in the perfective nonpast cannot be identified with the utterance time of the context).This was illustrated for the usual na in (4) and (5) above, where pao and dho are respectively the dependent forms (the perfective nonpast), and in (3) for en na.
Second, as with all na-clauses in Cypriot Greek, when the embedded predicate in the na-clause is negated, the negator that appears under na is the 'subjunctive' negator men, as in (7), glossed NEG.SUBJ (see Chatzopoulou 2012 for the history of the Standard Modern Greek cognate min), not the indicative en (see section 1.2 for more on en).
( The appearance of men in this context is expected if the na that we see in (8) is the usual na found elsewhere the language, such as that in (7).
There are two pieces of evidence that make it clear that the sequence en na is not a fixed or lexicalized expression of a single head, tense or otherwise. 5First, the copula part can appear independently in the past tense as illustrated in ( 9 This is entirely expected if en is merely the nonpast 3rd person copula: itan, also found as ito (Vassiliou 2002) or itun, is the regular 3rd person past tense of the copula.
Second, we find coordinated na-clauses under en (and under itan, shown in (25) below): (10) a. En be.NONPAST. 5Precisely the same reasoning applies to the English modal expression be to, which has undergone a different path of development and has slightly different properties, but which bears obvious similarities to the Cypriot Greek locution.See Huddleston et al. 2002:113 andKayne 2013 for brief discussion.The fact that Greek also has such an expression shows that Kayne's proposed generalization about the distribution of this construction cannot be correct: he analyzes Abby is to appear tonight as Abby is MEANT FOR to appear tonight and claims that only a language like English that has a case-assigning complementizer like for will be able to have this construction.Greek lacks any such case-assigning complementizer.Kayne's reasoning is based on the idea that there is no way to assign the attested modal semantics to be to in its individual parts (and therefore the modality must come from unpronounced elements).But this reasoning strikes us as unfounded: the same question arises with any multi-morphemic idiom or partially compositional phrase, including call up 'muster', call out 'challenge', spill the beans 'reveal the secret', etc.Any solution to the issues raised for compositionality in such phrases (see e.g.Kobele 2012 andHarley 2014) will apply to be to and en na as well.If en na were a single head or fixed lexical expression, then coordination of such na-clauses alone should not be possible. 6he verbal form selected by na in the periphrastic future also behaves as expected for such forms in Cypriot Greek with respect to the placement of clitics.Though object clitics in Cypriot Greek are enclitic in a number of matrix and embedded clauses (see Terzi 1999b; Mavrogiorgos 2013, to appear among others), as illustrated in (11), clitic arguments of dependent verbs under na are proclitic, as seen in ( 12). ( 11 The fact that en en is not found in (17) is not due to mere haplology (even syntactically conditioned haplology, as in Salzmann 2013 and Merchant 2014), given that (15a) is well-formed.
What we find instead for the negated future is surprising and shows an interesting fact of syntactic micro-variation between two closely related varieties.It has been observed before that "Cypriot Greek lacks the future particle [tha] of Standard Greek" (Terzi 1999a:110  The appearance and identity of this tha is the puzzle to be solved.
2 Spanning: Vocabulary Items that realize more than one node The element tha is in complementary distribution with the usual future marking strategy en na.The most parsimonious analysis therefore is to take the two sets of elements to be differing, competing realizations of the same nodes, one realized under negation, and the other otherwise: allomorphs, in other words, expanding the use of that term slightly.9 Such an analysis can be implemented in a theory such as Distributed Morphology (see Arregi and Nevins 2012and Bobaljik 2015and Embick and Noyer 2007) in which a single generative system is responsible both for word structure and phrase structure and derivation of complex objects is syntactic.Distributed Morphology (Halle and Marantz 1993) incorporates hierarchical structure into morphology by positing that the input to morphology is syntactic structure.Traditional features (or feature bundles) are distributed over nodes, which in turn are subject to Vocabulary Insertion, the process that replaces or realizes morphosyntactic featurally specified nodes as particular morphemes.In our analysis, we posit the following Vocabulary Items (abstracting away from person features here for simplicity): (20) a.
By the Elsewhere Principle, insertion of the Vocabulary Items in (20a,b) will be preempted by the more narrow context of insertion in (20c).
The unusual property of the entry in ( 20c) is that it realizes three nodes, not just one.It is for this reason that we adopt a theory that allows Vocabulary Insertion to target multiple nodes with a single rule: these nodes must form what is known as a span (Svenonius 2012;Merchant 2015;Bennett et al. 2015), and our innovation here is to claim that tha is a portmanteau morpheme that in inserted for T, the copula root, and C simultaneously. 10The resulting structure is as follows: (21) NegP q q q q q q q ▼ ▼ ▼ ▼ ▼ ▼ ▼ Neg TP We note here that it is not crucial that the root of the copula be categorized as V, as we have done here, nor is it crucial that the copula has not moved to T. Either of the routine assumptions that the root is categorized by a separate V node or that V moves to T are compatible with our analysis; they simply require slight reformulations of the Vocabulary Items in (20).Indeed, we could leave the copula entirely in situ and make the presence of en vs. itan contingent on whether the neighboring Tense node is specified as present or past, as in ( 22).This kind of analysis merely extends the logic of spanning from the insertion of Vocabulary Items to their conditioning environments (as argued for in Merchant 2015): such an alternative would need to take the sequence of nodes that constitute the conditioning environment together, forming themselves a span of nodes.11( 22) Whichever implementation is ultimately to be preferred, this kind of local allomorphic analysis makes a prediction: if the adjacency between en and na is disrupted, tha should fail to appear.There are in fact two such cases, both of which bear out this prediction of our analysis.
First, the copula allows for a missing complement: dropping the CP sister to the copula will bleed the rule in (20c).Under our analysis, the environment for the insertion of tha is not met (due to the lack of C) (compare 23b and 23c), and we expect to find instead an emergence of the unmarked effect (McCarthy and Prince 1994).The following data show that this expectation is met: ( FUT (Intended: 'Giannis will go out tomorrow, but Maria will not.') We can see that the missing na-clause is due to ellipsis, not Null Complement Anaphora, because it is possible to extract from the missing complement.In (23a), the preverbal subject i Maria in the matrix clause has moved there by an A-movement from the embedded clause (see section 3.1 below 'The car, Kostas was going to wash, and the motorbike, Maria was (going to wash).' The fact that ellipsis blocks allomorphy that depends on a triggering element internal to the ellipsis site is a simple ordering effect: ellipsis bleeds the allomorphic rule in (20c) by removing part of the structural description of the rule.This interaction is thus fully parallel to those studied in Bennett et al. 2015 for Irish, for example.
The second piece of evidence that the appearance of tha involves locally conditioned allomorphy over a span comes from the behavior of en na and tha in coordinations.As seen in ( 10) above and in (25) here, na-clauses can be coordinated under the copula, with the futurate meaning maintained for both conjuncts.But in the negated present, we find an asymmetry: the conjunct closest to negation shows the expected replacement of en na by tha, but any nonlocal conjunct does not.In other words, when the span targeted by (20c) is interrupted, the tha allomorph is impossible in the conjunct that isn't adjacent to Neg-instead, we find the regular na again: We assume that the coordination of the CP complements to en is a balanced (symmetrical) coordination with the expected morphology on both conjuncts (see Johannessen 1993), represented here for convenience as a ternary branching structure. 13Vocabulary Insertion targets a span of nodes that includes the C head of the closest conjunct, in an apparent violation of the Coordinate Structure Constraint: (27) NegP q q q q q q q ▼ ▼ ▼ ▼ ▼ ▼ ▼ Neg TP VP q q q q q q q ( ( ( Note that an analysis based on head movement followed by Fusion would have difficulty accounting for these data: head movement is subject to the Coordinate Subject Constraint, so the requisite complex head consisting of T, √ BE, and C 1 (which would form the input to the Fusion operation) cannot be formed in the syntax.
If this analysis is correct, we must countenance an expanded domain for spans, allowing T-V-C in ( 21) and ( 27) to count as a span (or merely V-C, if the rule in ( 22) is correct).This conclusion is at odds with the definitions of spans that restrict spans to extended projections, such as the following, from Merchant 2015: (28) Let T be an ordered n-tuple of 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 span.(Every node is a trivial span.) b.For any n > 0, if t k is a span, then t k , . . ., t k+n is a span.
(29) Spanning Insertion Hypothesis: A span and only a span can be targeted for Vocabulary Insertion.
Instead, it would appear at first glance that we must define a span to include a contiguous string of elements after Linearization.By locating the point of the derivation that spanning is sensitive to after Linearization, this proposal also explains why no Coordinate Structure Constraint violation is registered in ( 27): the CSC is a constraint on syntactic (or semantic) representations, not on strings.The definition in ( 30) is much weaker than that in (28), since it eliminates the requirement in that the elements be in an extended projection.14 (30) Let T D be the unique totally ordered n-tuple of terminal nodes t 1 , . . ., t n over the elements in a derivation D that satisfy the Linearization statements generated by D.15 a.For all k = 1 . . .n, t k is a span.(Every node is a trivial span.) b.For any n > 0, if t k is a span, then t k , . . ., t k+n is a span.
That the conditions on insertion of Vocabulary Items must in some cases be made sensitive to linear adjacency is a conclusion argued for on independent grounds by Arregi and Nevins 2012 and Ostrove 2015 (though see Moskal and Smith 2016 for an opposing view).
At this point in our investigation, we have little evidence that mere adjacency is not the best model for these data, though allowing such string adjacency to be the sole conditioning factor for allomorphy is widely thought to overgenerate (see Svenonius 2012 for discussion).Another possibility for analyzing the coordinate structures would be to assimilate the spanning insertion condition to that found for closest conjunct agreement (as in McCloskey 1986, Munn 1999, Villavicencio et al. 2005, and Haegeman and van Koppen 2012, among many others).These in turn could privilege the first conjunct for structural reasons: it could be that the first conjunct CP is the true and only complement to the copula, and noninitial conjuncts are mere adjunct CPs to the first conjunct.These latter would show na, therefore, because they are not in the extended projection.This move raises a number of obvious difficulties (for extraction, subcategorization, and agreement), but these are the usual difficulties in handling conjunction to begin with.
Nevertheless, given the wide range of predicted but absent phenomena that mere linear adjacency would allow, it seems most prudent to find a middle ground.What we need is to define spans as consisting of all tuples of nodes that are in the set of adjacent terminal nodes and that stand in a (possibly transitive) selection relation.This is easily done: (31) Let T ≺≺ be the unique set of ordered pairs of terminal nodes t i , t j over the elements in a derivation D such that t i immediately precedes t j .Let T S be the set of all pairs of nodes in D t i , t j such that t i selects t j .a.For all k = 1 . . .n, t k is a span.(Every node is a trivial span.) b.For any n > 0, t 1 , . . ., t n is a span iff for all 1 This definition picks out a subset of those nodes that select other nodes: the subset of such nodes that also stand in the immediate precedence relation.This means that when a verb selects a CP and immediately precedes it, V, C will be a span.Likewise when a v selects and immediately precedes a V.But when a v selects a specifier to its left (say, a DP), v, D will not form a span: although v selects D, it does not precede it.Likewise for potential selection relations that hold between a selecting head and a head remote from the selecting head, as was the case in the coordinations in (27) above: √ BE selects C 2 , but does not immediately precede it.Since √ BE, C 2 / ∈ T ≺≺ , these two elements do not form a span.16An additional prediction is made by the present analysis: because T-V-C must form an uninterrupted span to surface as tha under negation, the presence of a marker on the left conjunct should make tha impossible.This is the case in balanced coordinations, such as those involving ute ... ute 'neither ...nor' under negation (see Giannakidou 2007 for further discussion of the properties of ute).In such a situation, the unmarked na should appear.These predictions are also borne out: ( The example in ( 33) is particularly significant: it shows that the appearance of tha for en na under negation is not due to some incompatibility of negation with en na, either syntactic or semantic: in (33), the regular sentential marker en does occur with en na, yielding the expected meaning.This remarkable reappearance of en na under negation is predicted by our analysis: because the copular en is not adjacent to na in (33), the allomorphic rule in (20c) will not apply, and instead we find the usual exponents of these morphemes.

On the properties of the copula
The nature of the interactions between the higher verb, realized as en or itan, and the surface subject, are tangential to our concerns in this paper, so we will confine ourselves to only a few remarks.
It appears that apparent preverbal subjects in the matrix clause are moved there from the embedded clause via one of the strategies that underlie the robust word order permutations that Greek enjoys: the movement that derives much of the attested variation appears to have A-movement-like properties despite not targeting a position associated with agreement in their own clause (see Alexiadou and Anagnostopoulou 2002 for raising out of na-clauses, and Spyropoulos and Revithiadou 2007 for a discussion of preverbal subjects).
For example, subject idiom chunks from clausal idioms like the one in (34) do not appear in control clauses (35a) or as cross-clausal topicalized phrases ( 35b Second, nominative anaphors as in (39) allow backward binding into subjects of raising, but not control, predicates (see Landau 2013 for extensive additional discussion and diagnostics).This shows that the anaphor can be reconstructed to and bound in its base position.In (39), o eaftos tu is in the subject position of the main clause and it is bound by the pronoun in the na-clause; this is consistent with the supposition that o eaftos tu A-moves out of the embedded clause to the matrix clause.appeal.3s'He will like himself.' Two other typical tests that distinguish A-from A ′ -movement, the presence of weak crossover effects and the licensing of parasitic gaps, yield less than reliable results in Cypriot Greek, and so we do not report the data here.Briefly, quantificational subjects of en na clauses appear to be able to bind pronouns that are inside matrix adjunct.In other words, such subjects do not trigger weak crossover effects.We would normally conclude that they reach their surface positions by A-movement (over matrix adjuncts), expanding their A-binding domains.But it is difficult to interpret the results of this diagnostic because it is not clear whether even clearly topicalized phrases from embedded clauses trigger weak crossover effects in all cases in the first place.It is also unclear whether Greek has a set of gaps with the profile of parasitic gaps in languages like English.
Given the parallels to raising predicates, then, it comes as something of a surprise that in these constructions, the copular verb always appears in the 3rd person (recall from (1) that the 3rd person shows no number distinction in this verb).( 40 While unusual, this would not be the only verb in Greek to show this behavior: impersonal verbs like prepi 'must' and bori 'is possible' have a similar pattern.There are two obvious analytical paths for accounting for the lack of agreement on en in (40).
First, the locality domains for Agreement and for movement could be different (Potsdam andPolinsky 2008, Keine 2016): in Greek, apparently, the embedded CP na-clause, which does not assign nominative case, allows for EPP-driven (•D•driven) A-movement from its subject to the higher subject position.The ability of agreement to probe into the CP could be limited, perhaps because CP itself has φ-features.The necessity for default inflection would follow on this analysis from the fact that the higher T node, where the probing φ-features are located, is separated from the potential controller of agreement by a clause boundary, CP.Movement of the DP into the higher domain does not alter this fact, since the probe only agrees downward.
The second possibility again would take it that Agree to value the φ-features on the higher T can only probe T's c-command domain, but would derive the lack of agreement from an interaction of movement and Agree.If movement occurs before Agree, then movement of the subject above T would bleed Agree.Such an analysis can be implemented in a system like that of Georgi 2014 and Martinović 2015, who study such opaque interactions in detail: the movement feature on the head that agrees would precede the agreement feature: •D•, * φ * .The advantage of this approach is that it could code on individual lexical items a differing order of Move and Agree triggers, allowing us to understand the difference between en/itan and the aspectual verbs studied by Alexiadou and Anagnostopoulou 2002, which do agree with their derived subjects, as a lexical difference, specified in the lexical entries for the different verbs.
In either case, any theory of default values should capture the fact that what surfaces is 3rd person (for example, the interaction of default values with Agreement failures in Preminger 2014).It it not important for our purposes to choose between these alternatives, and we leave adjudication between them to future work.

Is clause union an alternative?
There is in principle another possibility for analyzing the allomorphic competition between en na and tha, one that would preserve the definition of span built on the notion of extended projection, and not weaken it to mere transitive selection (modulo the immediate precedence condition).This alternative would require two changes to our analysis: first, we could adopt the position of many researchers (such as Giannakidou 2009, Philippaki-Warbuton 1994, andothers) that na is in a lower clause-internal projection, such as Mood, and not in C, and does not require embedding under a C. Second, we could claim that the copula en/itan in these structures selects for MoodP directly, bypassing the CP layer, and that this truncated complement phrase is the realization of a restructuring context.As in Germanic and Romance restructuring phenomena (Wurmbrand 2004), the selecting V and the lower V would be in the same domain for certain purposes, including allomorphy (a phenomenon familiar from German for example, where one subset of restructuring contexts licenses the Infinitivus Pro Participio, a locally conditioned variant realization of a verb under the perfect auxiliary).On this analysis, na and en would not be in different domains, and our definition of span requires no revision along those lines: instead, we could analyze the entire V-na-V complex as involving only one extended projection (see also Grano 2012 for an in-depth discussion of some Greek restructuring verbs).
On this analysis, we would have structures such as the following.

Mood
x 8 x 8 x 8 x 8 TP q q q q q q q ▼ ▼ ▼ ▼ ▼ ▼ ▼ tha T VP q q q q q q q ▼ ▼ ▼ ▼ ▼ ▼ ▼ (pro 1sg ) pao In this case, we would therefore not find a CP layer, even though na-clauses in other contexts project CPs, as can be seen by the appearance of a clitic-leftdislocated DP between a selecting verb and na in the following examples.The en na construction, however, appears to have an embedded CP layer: there is a landing site of at least marginal acceptability for clitic-left dislocated phrases between the matrix verb and na: 17 But a significant remaining question on such an analysis is why the higher T cannot agree with the lower subject.If a clause boundary intervenes (a CP, as in (21) above), standard theories of locality of agreement will correctly predict that the embedded subject only triggers agreement on the embedded verb.If we collapse the clauses, it would seem, ceteris paribus, that the matrix T would now be in the same phase as the lower subject, and hence able to value its φ-features via Agree with the embedded subject.This use of the + na is also reported in Roussou and Tsangalidis 2010 as a reduced form that maintains the na-complement.Markopoulos 2008 notes the emergence of the na in the 14th century AD as a construction used for expressing volition-related meaning, and he treats the as the product of the loss of the unstressed word-final /i/ of the volitional verb theli 'want' following the loss of intervocalic /l/.Similarly, the 2nd person singular in Standard Modern Greek also exhibited a similar reduction, from thelis to thes, in the same period, a fact suggesting that it belonged to the same pattern as the the na construction.where na na ta them mbasomen?pass.through.1pl'If they will go through the hole, where will we pass the horses through?' (Cypriot Greek, Mahairas, 509) Synchronically, the is hardly used in any other context and speakers do not necessarily associate it in the aforementioned construction with a volitional reading.As the example in (46a) shows, it is no contradiction to not want to do something but to nevertheless predict or plan for it to happen; the fact that the in precisely the same context gives rise to a contradiction, as indicated in (46b), shows that the is not merely a reduced form of thelo and is compatible with the indicating futurity in the same way as en na (though we will have to leave a close investigation of their semantic differences to future work).( 46 Note finally that the fact that the na can be felicitously embedded under negation makes it unlikely that the failure of en na to similarly appear under negation is due to its meaning or to something idiosyncratic about na in these contexts: it is due to the fact that there is a more specific morpheme, tha, which is competing for precisely the copular+na span under negation, and which, by the Elsewhere Principle, pre-empts the appearance of en na.

An additional point of comparison with Standard Modern
Greek tha The assumption that the Cypriot Greek future periphrastic construction is built on a biclausal structure, involving a na-clause, comes as no surprise when considering the development of the future particle tha in Standard Modern Greek.Historical work on the development of the future particle tha also suggests that more complex structures were involved, similar to the one we argue for here.Joseph andPappas 2002 andRoberts andRoussou 1999 argue that Standard Modern Greek tha involves a redeployment of the volitional verb thelo 'want' and the infinitive, the latter replaced by the head ina.A use of the volitional verb and the na-clause to give a future reading has also been documented for medieval Cypriot Greek in Aerts 1983 in (47) and a similar development to today's use of en na has also been previously assumed (Chatziioanou 1999:92, Menardos 1969).The Cypriot Greek periphrastic en na can profitably be compared to the Standard Modern Greek future particle tha (see Tsangalidis 1998 and Giannakidou and Mari to appear for extensive discussion of Standard Modern Greek tha).In all contexts in which we saw en na appear above, Standard Modern Greek would use the particle tha: like en na, tha takes both perfective and imperfective nonpast verbal complements in its futurate use (while also taking the past in its epistemic use).Both en na and tha can appear, for example, in protases of conditionals: (48) a.The fact that tha can appear in the Cypriot Greek conditional protasis in (48b) is unexpected, unless the conditional head an in Cypriot Greek bears the relevant conditioning feature NEG that the Vocabulary Insertion rule in (20c) above requires.We are not in a position to pursue this further here, other than to note the cross-linguistically variable appearance of so-called 'expletive' negation in conditionals (see Yoon 2010 for discussion).
This comparison of the uses of Standard Modern Greek tha and Cypriot Greek en na serves strengthen the supposition that these two elements play parallel roles and have parallel distributions in the two language varieties.

Conclusion
The variation of the Cypriot Greek periphrastic future en na∼tha presents a puzzle for standard spanning theory, where spans are restricted to extended projections.We analyzed en na as being just what it seems to be: the copula followed by the 'subjunctive' subordinating particle na; tha replaces both of these under clausemate negation.In other words, tha is a portmanteau form realizing the copula and the embedded head that normally surfaces as na: concretely, tha realizes a (T-)V-C span.In one sense, tha is an allomorph of en na.
Cypriot Greek tha therefore represents a morpheme whose environment for Vocabulary Insertion cannot be stated within a single extended projection.For this reason, we revised the definition of span to make it sensitive to mere selection (including across a clause boundary).The facts from coordination, in particular the ability of the closest conjunct to license tha across its edge, and the fact that CP ellipsis bleeds tha, required a further, more radical revision: spans are sequences of selecting heads that also immediately precede one another.
Constraints on possible morphemes can come from the kinds of linearizations that are possible, from cyclic constraints on the generation of those statements, or elsewhere, but one conclusion is inescapable: eppur si spane!

17
For reasons that are unclear to us at present, cases in the present tense are much worse: going to fire Yiannis tomorrow.'

Greek
Cypriot speakers also find examples like (44) acceptable.At first sight, this seems to involve a contracted form of the volitional verb thelo (to the; see Joseph and Pappas 2002 and Pappas 2001 for relevant discussion); but this the is not found anywhere else in the language.not going tonight.' will stay, let him stay.' 23) a. O the Merchant 2013)ncies can also reach into the missing CP (unlike what is possible in cases of Null Complement Anaphora; seeMerchant 2013).12 32) a. *En If this reasoning is correct, we have a case of A-movement out of an embedded finite clause, a conclusion in line with that of Alexiadou and Anagnostopoulou 2002 for certain aspectual predicates.Two additional tests from Alexiadou and Anagnostopoulou 2002 yield the same result.First, as they discuss, pseudocleftlike structures as in (38) do not allow the na-clause complement of raising verbs to be the pivot.
) a. En