Decomposing EPP effects in Greek enclisis

This paper pursues the idea, originally proposed by Landau (2007), that the Extended Projection Principle is PF related on the basis of Greek enclisis. It is argued that the complementary distribution pattern attested with Cypriot Greek finite enclisis derives from the fact that the first head H c-selecting TP has a morpho-syntactic requirement, and a related PF/prosodic requirement subject to an Economy Condition. The former derives merger of an X or XP copy at H, while the latter ensures that only one of the two copies gets spelled-out. Non-finite H triggers obligatory enclisis in both Cypriot and Standard Greek, as it contains only affixal morphemes, which is further supported by Medieval Greek non-finite enclisis. The parameterization of H along with potential implications are also discussed.


Introduction1
The nature of the Extended Projection Principle (epp) (see, e.g., Chomsky 1981 and subsequent literature) has always been controversial within generative linguistics. In this paper, I argue that an approach according to which the epp is only partially morpho-syntactic in nature (see Landau 2007) is in a better position to account for the complicated facts of finite enclisis in Cypriot Greek (cg). An example illustrating the finite enclisis pattern is provided in (1) below: ( As shown in (1), in the presence of clitics, either the verb or a preverbal quantifier has to appear in clause-initial position in order for the sentence to be grammatical. Crucially, though, these two strategies are in complementary distribution, as illustrated in (1d).
In what follows I show that this characteristic pattern, found in all Tobler-Mussafia languages (i.e. languages with finite enclisis), only concerns constituents which merge (for independent morpho-syntactic reasons) within a particular clausal area. This area (highlighted in (2) in bold) immediately extends the xp containing the clitic and the verb: ( In particular, I argue that when a constituent x or xp is merged within the said area, v-merger to a head higher than the cliticization site is blocked (*v). This correlates with proclisis: ( *v I argue that this distribution follows from a phonological/spell-out requirement imposed on an Agreeing functional head h with the following properties: h c-selects tp/inflp (= xp in (2-4) above) and is structurally and functionally defined, although crucially not in terms of feature content (i.e. h is subject to a 'decomposed epp' , following the hypothesis put forward by Landau 2007 that the epp is (also) pf in nature): f2; … f n } and feature projection is subject to f1 > f2 > … f n (5) specifies that h has a morpho-syntactic feature that needs checking. It also specifies that h may come in flavours (i.e. it may contain different types of features), where insertion and projection of each feature is regulated by a feature projection algorithm of some sort (which may or may not be reducible to independent semantico-syntactic restrictions). Besides this morpho-syntactic requirement, there is an additional pf requirement imposed on h, namely its Edge must be realized by material with phonological content: (6) [hp Spec [h' h [+feature] [ yp wp y zp]]] agree Given that the Edge of a projection is taken to include both the head and the spec,2 and assuming that this requirement, being an interface requirement, is Journal of Greek Linguistics 17 (2017) 190-232 restricted by Economy, realization of either the head or the spec must be sufficient, and therefore required (see also Collins 2007;Koopman 2000;Kupula 2011;Nchare & Terzi 2014;Pearson 2005;Pesetsky 1998). This is illustrated in (7).
I argue that it is exactly the Economy restriction in (7) that gives rise to the complementary distribution pattern illustrated in (1) above. I further show that obligatory non-finite enclisis in cg (and Standard Greek (sg)) also involves an Edge Requirement imposed on h, however in this particular case an additional affixal requirement, independently imposed on h within Morphology, blocks the complementary distribution. This fact supports the idea that the pf requirement is prosodic in nature (as morphology cancels out (7) as long as (6) is satisfied). Finally, I argue that finite enclisis is not found in languages like sg due to the fact that h is parameterized. This paper is structured as follows: section 2 gives the theoretical background and introduces the notion of the epp adopted in this paper. Section 3 offers a detailed presentation of finite enclisis in cg (and of enclisis in cg and sg more generally). Section 4 provides an analysis of the data based on epp decomposition. Section 5 concludes the discussion.

Theoretical background
Researchers have been disagreeing whether the epp constitutes a theoretical primitive, whether it should be reduced to independent principles such as case or agreement, or whether it should be considered an epiphenomenon and hence abandoned altogether (see e.g. Grohmann et al. 2000;Landau 2007;Lasnik 2003 for an overview). But, what is the epp? Initially, the epp was taken to be a grammatical rule ensuring that a finite clause has an overt subject. Within Government and Binding (gb) Theory, this would translate into the requirement that finite t/agr projects an overt specifier (at least in languages like English): Journal of Greek Linguistics 17 (2017) 190-232 of various possible syntax-pf interface models; see also Landau 2007;Pesetsky 1998;Richards 2010Richards , 2016. Although it is not as yet clear whether it is feasible, theoretically and empirically, to have a purely pf requirement or not, what most researchers agree on is that pf should be as restricted as possible. A second question revolves around any possible realizational interactions between an xp and an x merged at a certain functional head. In particular, it has been pointed out in the literature that cross-linguistically the spec and head of a projection need not (and, by economy, must not) be realized phonologically at the same time (although one of them must be spelled-out; see, e.g., Speas 1995). This restriction is more generally known as the doubly filled Comp filter (see Chomsky & Lasnik 1977); however, it is found in other domains besides the cp, as illustrated in (9-12) below: cp domain (Pesetsky 1998 This pattern involves complementary distribution, strict locality (i.e. the x or xp must spell out f locally), and wide distribution (i.e. it applies (to a certaintypically higher-head) across functional domains).4 In what follows I propose that enclisis in cg and sg also falls within the realm of the epp, as described above. Such an approach groups enclisis together with phenomena such as pro drop, doubly-filled comp filter effects, clitic doubling, stylistic fronting, as well as that-trace effects, in the sense that they all involve a complementary distribution of phonologically realized constituents that are in a local structural relationship.

3
Enclisis in cg and sg 3.1 Overview I start the discussion by giving an overview of (linear) enclisis in cg and sg (for phonological proclisis/enclisis, see Klavans 1985; for cg, see Revithiadou 2006Revithiadou , 2007. For descriptions of cg enclisis, see Agouraki 1997Agouraki , 2001Agouraki , 2010Chatzikyriakidis 2010Chatzikyriakidis , 2012Condoravdi & Kiparsky 2001;Mavrogiorgos 2010Mavrogiorgos , 2013Pappas 2004;Philippaki-Warburton 1995Revithiadou 2006Revithiadou , 2007Rivero 1994;Rivero & Terzi 1995;Terzi 1999a,b). The main generalization to keep in mind is the following: in sg, enclisis strictly correlates with lack of person and/or restricted person specification on the verbal host (see Mavrogiorgos 2010). In cg, on the other hand, enclisis correlates with properties of the left periphery of the clause, and only residually with lack of person and/or restricted person on the verbal host.
More specifically, in sg a clitic (cluster) immediately precedes or follows the verbal host, depending on the finiteness properties of the latter: 4 Moreover, it seems to differ from criterial effects (see Rizzi 1996), where both the spec and the head of a projection are realized overtly (although, crucially, they do not have to). This may be the case because the relevant spec and head do not get spelled out within the same phrase, or because languages with criterial effects allow for feature bundling (hence, the spec and head each realize a distinct feature at the same location). This is an issue open for further research.
Journal of Greek Linguistics 17 (2017)  Proclisis typically arises if some preverbal constituent of a particular type ccommands the finite cl v cluster. Depending on the phrase structure status of the c-commanding constituent, the following sub-cases can be distinguished (see also Mavrogiorgos 2013 for Tobler-Mussafia languages more generally) (preverbal constituents are in italics; clitics are in bold):

3.1.1
The c-commanding constituent is an xp Typically, this sub-type involves either operator phrases and/or stressed phrases (including wh-phrases, verum focus phrases, and contrastive focus phrases).5 Note that if the relevant xp does not c-command the cl v cluster, enclisis obtains: [stressed subject phrase in situ] 'I gave it, not Mary.' also require the presence of a negation particle (negative concord), which immediately precedes the cl v cluster and which independently triggers proclisis (as cases with postverbal negative phrases (ii) reveal): give.pst.3.sg (ii) En mu éðoke típota. 'S/he did not give me anything.' As they are not revealing, negative phrases are not discussed in this paper. On the other hand, (i) poses a potential complication for the approach defended here: if both en and típota merge at h and get spelled out there, this would contradict the Economy Condition in (7). In this respect, note that it is likely that típota is spelled out in the spec of a separate Focus head (cf. that it is obligatorily stressed). Alternatively, we could assume that the non-affixal status of the negative particle enforces spell-out of h in the presence of an overt spec (i.e. Morphology overrides the Prosodic Constraint, the latter only being subject to the Economy Condition (and see also section 4.3)). Which option is the right one is an issue open for research. Note that there are certain preverbal xps that give rise to enclisis. These xps (which include d-linked wh-phrases (19d) (see Revithiadou 2006: 83 (her 4b'))) are typically interpreted as topics and must be cliticized when they function as objects. Note that contrastive topics (19e) are typically stressed, which means that the presence of a stressed preverbal xp is not a sufficient condition for proclisis: 'When Maria saw me …'

3.1.3
Two or more c-commanding constituents: An important question is what happens when there are more than one preverbal constituents present. The following patterns apply: (i) when both an x and (at least) one xp precede the clitic (cluster), it is the closest c-commanding constituent that determines the position of the clitic (cluster) in relation to the verbal host: 'S/he said that it was you that gave the book to him, not Mary.' In (21a) a complementizer immediately precedes the clitic (cluster) and triggers proclisis. When a preverbal subject topic intervenes between the same complementizer and the clitic (cluster) (cf. 21b), enclisis obtains. When the Journal of Greek Linguistics 17 (2017) 190-232 intervening phrase is a stressed pronominal subject (as in 21c), proclisis obtains.
What these examples show is that the proclisis effect of a complementizer can be cancelled out in the presence of an intervening phrase that normally triggers enclisis.
The same principle applies when (at least) two xps precede the clitic (cluster): it is the xp closest to the clitic (cluster) which determines the position of the clitic. Consider the following sentences: (22)  While a preverbal wh-phrase triggers proclisis (22a), a preverbal topic phrase triggers enclisis (22b). The crucial piece of data is (22c): here a wh-phrase triggers proclisis even though it is preceded by a preverbal topic, which suggests Journal of Greek Linguistics 17 (2017) 190-232 that it is the xp closest to the clitic-cluster (in this case, the wh-phrase) that determines the position of the clitic.7 The same effect is illustrated with preverbal topics (22d-f), which trigger enclisis, and with an intervening stressed phrase (22g), which triggers proclisis.
To summarize, in this section I have shown that, unlike sg, in cg the position of the clitic before or after the verbal host typically correlates with the spell-out position of certain preverbal cp-related xps and/or Xs (and only residually with the person specification on the verbal host).

3.2
Patterns in the data: the epp-like distribution of finite enclisis8 3.2.1 Preverbal xps and modality/negation Xs In this section I argue that the cg data described earlier have the epp distribution (see sections 1 and 2), in that enclisis only shows up whenever the specifier or head of a certain functional head is not independently spelled out.
By way of reminder, cg has proclisis when some preverbal operator and/or stressed xp (23a), or some preverbal Comp/Modal/Neg head (23b) c-commands the cl v cluster, otherwise it has enclisis (23c): On the other hand, a Topic xp gives rise to enclisis, and the same applies to certain complementizers: There are two main interesting facts regarding (23) and (24): first, the proclisisenclisis alternation seems to correlate with various types of features, which do not constitute a natural class, morpho-syntactically and/or semanti-Journal of Greek Linguistics 17 (2017) 190-232 cally.9 Second, all these features are typically considered to be projected above the phrase where the cl v cluster is located, i.e. at the cp-tp border. Leaving the first issue aside for the moment (but see section 4 below), it would be useful to see whether the projections hosting the relevant features form some kind of contiguous domain or not. In this regard, Rizzi's (1997) cartographic enterprise (and much subsequent work after that) offers us a useful localization tool.
In particular, Rizzi (1997) has pointed out that cross-linguistically Topics are generally merged higher than preverbal focused/stressed phrases and/or whphrases. It has been shown by various researchers (see e.g. Iatridou 1991; Anagnostopoulou 1997; Roussou 2000, among many others) that this also applies to sg, and the same seems to be the case also for cg, given the data presented in section 3.1 (and see the following sentences which illustrate the fact that the reverse order is ungrammatical):10 (25)  To illustrate with one example, there is no single operation that we know of which is contingent on the presence of a (non-wh) comp or a wh-operator. 10 Note that d-linked wh-phrases behave like other preverbal topics in that they trigger enclisis and may co-occur with other topics (iii-iv). However, the fact that a d-linked whphrase cannot co-occur along with a wh-phrase (i) (as opposed to non-wh topics, which can (ii)), suggests that at some point during the derivation the d-linked wh-phrase must be in WhP/Foc:  (27), in combination with the data presented in section 3.1, leads to the following descriptive generalization (to be further revised): Generalization 1: (28) Merger of a constituent (x or xp) above the c, v cluster and up to TopP triggers proclisis.
In other words, in cg the proclisis-enclisis alternation appears to correlate with the structural cut-off point between old and new information. In the following section the position of complementizers is discussed.
3.2.2 Complementizers As it has been pointed out earlier, complementizers in cg may trigger proclisis (29a), although some trigger both proclisis and enclisis (29b), while some trigger only enclisis (29c)  Starting from the non-optional cases in (a) and (c) above, the issue here is that two distinct complementizers, presumably merged in the same structural position (i.e. c), trigger either proclisis (a) or enclisis (c). This may suggest that the position of the clitic before or after the verb is not related to the structural posi-Journal of Greek Linguistics 17 (2017) 190-232 tion of the complementizer, clearly a counter-intuitive conclusion. Fortunately, it has been shown that some embedded/dependent clauses exhibit more rootlike properties as opposed to others (see, e.g., Emonds 2004;Iatridou & Kroch 1992), a difference plausibly linked to the higher vs. lower position of the complementizer introducing the embedded clause in each case. In other words, not all complementizers seem to be merged in the same structural position. This conclusion has been independently reached by the cartographic approach on the basis of various empirical phenomena, including recomplementation phenomena in languages that allow them (e.g. Spanish, European Portuguese, Italian or English varieties), but also with respect to cases where preverbal topics or foci may precede a (low) complementizer. correlates with variation in the structural position of the clitic (and therefore with proclisis-enclisis). This would be a welcome result, as in the previous section we saw that the structural position of preverbal xps, and xs other than complementizers (e.g. the negation head) does correlate with the proclisisenclisis alternation. So, what is the evidence for this hypothesis?
Although there is much we do not know about the left periphery of cg clauses, we may assume that cg and sg are similar in this respect. Starting from cases of obligatory enclisis (29c), what is interesting about the complementizer an dʒe is that it is composed of two independently existing homophonous morphemes, namely the complementizer an ('if') and the conjunction dʒe ('and').
Whereas an typically correlates with proclisis, dʒe always triggers enclisis. On the other hand, an dʒe (on its own) triggers enclisis (and the same applies to other complex complementizers containing dʒe; see also Agouraki 2001;Chatzikyriakidis 2012). Why should this be the case? Assuming that an dze is a conjunction-like element might provide an answer to this question, as bona fide conjunctions are associated with enclisis when they conjoin clauses (31) The standard intuition is that the conjunction in (31) does not count as being part of the second clause (and see Schütze 2004 and references therein for second position cliticization). However, there are reasons to believe that an dʒe in (29c) is a complementizer, and not a conjunction. For example, (29c) is rendered ungrammatical if one deletes an dʒe or an, but not if one deletes dʒe. Moreover, dʒe on its own may assume a complementizer-like use, in which case it triggers enclisis and cannot be deleted (while keeping the reading constant): Given the high comp status of an dʒe and the fact that it always triggers enclisis, we may assume that its high structural position is causally linked to clitic positioning.
To see if this is indeed the case, let us move on to (29a), which involves a factive pu, repeated here for convenience: 11 Note that a topic phrase preceding an dʒe is ungrammatical, unless it is construed as a main clause topic (in which case it cannot belong to the clause introduced by an dʒe): This is an example of a complementizer which (on its own) triggers obligatory proclisis.12 This should follow from its low (< TopP) structural position. An immediate problem for this hypothesis is that pu obligatorily precedes preverbal topics: One could assume that pu in (35a) checks some feature below TopP before moving on to some pre-topic position, and that its copy is sufficient to trigger proclisis. However, this would predict that we should get proclisis independently of the nature of the constituents preceding the clitic cluster, contrary to fact: (36) shows that a preverbal object topic triggers enclisis in the presence of pu. If pu had originated below tu Kósta (cf. the copy of pu), and if a copy were a sufficient condition to trigger proclisis, we would expect proclisis to be possible, contrary to fact. This suggests that the spell-out position of pu is the crucial factor. Moreover, it suggests that proclisis in (35a) is due to the presence of negation rather than to the presence of (a copy of) pu. The immediate question 12 Other complementizers of the same type include relative pu ('that'), interrogative and conditional an ('if'), as well as complementizers introducing temporal adjunct clauses (e.g. ama ('when'), pu ('when'), or prin ('before')).
Journal of Greek Linguistics 17 (2017)  There are at least two plausible ways to deal with this issue. One alternative would be to assume that pu is always merged/spelled-out high, but whether enclisis or proclisis obtains depends on whether pu is contained or not within the same prosodic domain as the clitic (cf. that cg clitics cannot appear at the edge of some prosodic domain). If the presence of a preverbal topic blocks this particular prosodic phrasing (a straightforward assumption, given what we know about the prosody of preverbal topics), it would explain why enclisis obtains in (36) (there is no host for the clitic within the prosodic domain containing it), as well as why proclisis obtains in (35a) (the negation particle and the clitic are merged/spelled-out within the same prosodic domain, independently of pu). A second alternative would be to claim that pu always merges low, triggering proclisis. Any difference would follow from the amount of structure pu merges with in each case: whereas in (37) pu merges with a clausal chunk that only contains {Cl, v} (presumably, a tp/inflp, given standard assumptions on cliticization, and see also section 4), in (35) and (36) it merges with a bigger clausal chunk (namely, one that contains {Top, Cl, v} (presumably, a TopP)). Assuming this is correct, a complementizer like pu, which triggers proclisis when on its own, would be free to merge with clausal chunks of variable sizes (including a tp/inflp). On the other hand, a (high) complementizer like an dʒe, which triggers enclisis when on its own, would be free to merge with any constituent besides tp/inflp. Why this would be the case is far from clear. What strikes me as important is that both alternatives propose more or less the same thing, but also face a similar problem. In particular, both argue that a proclisis trigger (which includes complementizers like pu) needs to form a contiguous domain with the clitic (cluster), although the nature of this domain differs depending on the alternative (syntactic vs. prosodic). In any other case, enclisis obtains. The problem that both seem to be facing is that whereas pu is able to count as part of the prosodic or clausal chunk that contains inflectional information (including the clitic) (either via prosodic restructuring, or due to merger with tp/inflp), this is not the case for an dʒe, which seems to be separated from it by some sort of barrier. This is an unavoidable conclusion, and it does correlate with clitic positioning. In addition, it is compatible with the intuition that whether Journal of Greek Linguistics 17 (2017) 190-232 some constituent counts as a proclisis trigger or not ultimately reduces to hierarchical structure (namely, its structural position), which was the essence of our working hypothesis (under the assumption that syntactic structure is directly related to prosodic structure).13 For all these reasons, I treat pu, and similar complementizers, as being able to merge below TopP.
That the structural position of a complementizer is crucial for clitic positioning becomes most clearly evident in the case of optional complementizers. An example of an optional complementizer was provided in (29b), which is repeated here for convenience: In cg, optional complementizers include among others óti and pos 'that' , or epiðí and γiatí 'because' (see also Agouraki 2001;Chatzikyriakidis 2012;Revithiadou 2006). Their common property is that they seem to be compatible with both proclisis and enclisis. One way to capture these facts is by postulating that optional complementizers may be either low or high: a low complementizer (or, given the discussion in the previous paragraph, a complementizer that may merge with tp/inflp) triggers proclisis, as opposed to a high complementizer (or, alternatively, a complementizer that cannot merge with tp/inflp), which triggers enclisis. This hypothesis is consistent with the overall empirical picture, given that low-only complementizers trigger proclisis, as opposed to high-only ones, which trigger enclisis.
One piece of evidence in favour of this analysis is the fact that high vs. low complementizers also correlate with semantic/pragmatic differences (besides prosodic/phonological differences). This suggests that they involve distinct morpho-syntactic features (a fact that increases the chances they are merged in distinct positions).14 Consider the following pair of sentences from Asturian, a language that behaves like cg as far as finite enclisis is concerned: 13 This conclusion would be further supported, if at least some of the complementizers that trigger proclisis could be shown to allow preverbal constituents, including topics and foci. This is an issue for further research. 14 In this respect, see Pesetsky & Torrego (2001), who propose a double c analysis for Spanish embedded questions (on the basis of featural content), which among other things Journal of Greek Linguistics 17 (2017) (50))] According to Fernández-Rubiera, enclisis in (39a) correlates with a high Comp (which does not block v-movement to CFin), which in turn correlates with epistemic modality (i.e. the speaker asserts that the reported content is true) and enclisis. (39b), on the other hand, correlates with a low Comp (which blocks v-movement to CFin), which in turn also correlates with epistemic modality (in this case the speaker does not assert the truthfulness of the reported content) and with proclisis. Similar semantico-pragmatic effects have been reported for embedded v2 in Germanic languages (see, e.g., Meinunger 2006;Truckenbrodt 2006), and they have been linked to the root vs. non-root distinction. More importantly, they seem to hold also in cg. In this respect, consider the following sentences:15 The effect is identical with Asturian: the óti Comp that allows for enclisis correlates with assertion of the truthfulness of the reported content by the speaker, as opposed to the óti Comp that allows for proclisis (where the content is only reported but not asserted). This fact, which has not been previously reported for cg, clearly suggests that the two complementizers differ in semanaccounts for the lack of that-trace effects. Interestingly enough, both cg and sg have h (see section 4.3), lack that-trace effects, and have obligatory v-to-t movement. Thanks to an anonymous reviewer for pointing out this link to me. 15 Thanks to Christina Giannapi for these judgments.
Journal of Greek Linguistics 17 (2017) 190-232 tic terms, and by default (given our current understanding of grammatical models) also in syntactic terms (including features and possibly positioning, or alternatively the portion of structure projected above the cl v cluster). This analysis is supported by independent prosodic evidence: whereas in (40a)  Where does this discussion leave us regarding cg? It seems that preliminary investigation strongly suggests that complementizers for which we have independent evidence that they are merged low in the clause (i.e. below TopP but higher than tp/inflp) trigger proclisis, while complementizers for which we have evidence that they are merged higher in the clause (i.e. above TopP) trigger enclisis. This gives us the following (revised) empirical generalization: Generalization 1':

(41) Merger of a constituent (x or xp) above the cl v cluster and up to TopP triggers proclisis. This also includes complementizers merging at some position lower than TopP
If a low Comp may be merged either above or below a FocusP/WhP (but lower than TopP; see also Roussou 2000), one gets the following structure (where one may assume, for expository reasons, that xp = tp/inflp; c3 = CForce; c1 = CFin and TopP = Left Dislocated Topic/ld): In this case, v-merger to a head higher than the cliticization site is blocked (*v), and proclis obtains: 16 It is not clear to me where énna would be in this structure. One possibility that comes to mind is c1 (the position of na), as it is an invariable particle realizing modal inflection. This give rise to the following abstract generalization: Generalization 1'':

(45) Realization of a spec or a head within a particular area within the high tplow cp domain blocks realization of the verb within that same domain.
What is the best way to analyse this distribution? For example, does the complementary distribution pattern involve the spec and head of a single syntactic head? Why are only heads within a particular domain subject to this pattern? How can we capture the fact that distinct morpho-syntactic heads are subject to the same pattern? Finally, why should the spell-out position of an xp interact with the realization of the verb (i.e. of a head)? These issues are taken up in the following section.

epp as a pf edge requirement
Let us repeat the generalization formulated in section 3.2.2 above. What the data show is that finite enclisis in cg involves activation of the high tplow cp domain, in the following manner: when a head or phrase is merged and/or spelled out within this domain, proclisis obtains. Otherwise, we get enclisis. In other words, any single feature merged and/or spelled out within this particular domain is sufficient to trigger proclisis, independently of its precise position within that domain. If we assume that enclisis involves the realization of (a copy of) the verb in some position higher than the cliticization site, then it is simpler to assume that the complementary distribution pattern observed with finite enclisis derives from some restriction applied to a single projection. Given that phonological realization is conditioned by the presence Journal of Greek Linguistics 17 (2017) 190-232 of a morpho-syntactic copy, it follows that the restriction must be pf in nature (see also Landau 2007 for a detailed discussion, and also the analysis below), as syntax does not prevent the presence of both copies at the same projection (although it does not force it, either). This is the claim I am going to make in this section. Before I move on to the analysis, let me first present my assumptions.

4.1
Assumptions I make the following four assumptions regarding cliticization and the creation of verbal copies within the clausal extended functional projection.

4.1.1
Assumption 1: clitics Clitics in cg (and sg) target the periphery of vP (to check case and/or a Topic feature), and the periphery of tp/inflp (to check deixis and/or person; see e.g. Anagnostopoulou 2003;Mavrogiorgos 2010;Terzi 1999a,b): Assumption 2: c-selection of functional heads A functional head locally c-selects (the head of) its complement (potentially via agree or feature share, given available feature matrices in Lexicon List 1). This gives rise to one or more morpho-syntactic dependencies/chains (cf. (47) for a standard case; see also Landau 2006Landau , 2007Pesetsky & Torrego 2001;Zwart 2001), where a chain may be phonologically realized by one or more items (this depends on morphological factors, as e.g. the available vocabulary items (see, e.g., Embick 2000), or on the morphological/affixal properties of the head of the chain and/or its position at a prosodic edge-see Lasnik 1995;Platzack 2010;Richards 2016;Zwart 2001, and section 4.3 below).
4.1.3 Assumption 3: head movement Head movement is contingent on agree among heads. While nothing precludes the creation of a higher copy via internal merger (in which case we would expect to get interpretative effects-see Lechner 2007;Roberts 2010), I assume that in the default case the actual movement part is delegated by pf (e.g. through an 'affixal/morphological' requirement on a c-selecting headsee Landau 2006 on Hebrew v(p)-fronting). As discussed later on in section 4.3, this assumption is empirically supported by non-finite enclisis.
Journal of Greek Linguistics 17 (2017) 190-232 4.1.4 Assumption 4: content is separate from function and structural position The final assumption is the most important but at the same time the most controversial one. In particular, the first head h c-selecting tp may come in flavours, in that it hosts some (interpretable) morpho-syntactic feature f compatible with this particular domain (e.g. a modal feature, a negation/affirmation feature, or an operator feature): b. h = {f1; f2; … f n } and feature projection is subject to f1 > f2 > … f n Only a single feature at a time may be inserted in h (modulo the c-selection feature, which is not interpretable, as opposed to f ). Feature insertion is regulated by a feature insertion algorithm of some sort (see 48b), which, hopefully, could be reduced to independent semantic-syntactic properties. Crucial for our analysis is the fact that not any feature can be inserted in h: topic features (or any other 'higher' feature ('higher' in the sense of Rizzi 1997) that does not belong to the core of the clause) cannot appear in h. Despite featural variation, the structural position of h remains constant. The same applies to its function, which is that of marking the edge of a morpho-syntactic domain prosodically and semantico-pragmatically (cf. that h carries interpretable features realized by stress or an overt lexical item, and constitutes the cut-off point for the theme-rheme division). The dissociation of content from function and structural position (see Ritter & Wiltschko 2009) allows us to take into account the variation we have seen, which cannot be accounted for on the basis of a single morpho-syntactic property (e.g. the relevant features do not form a natural class), and at the same time to account for the predictable interaction between h (a prominent structural position) and finite enclisis. Moreover, and as discussed in section 4.3, it opens the way for a principled account of crosslinguistic variation with respect to finite enclisis. The (morpho-)syntactic requirement of h reduces to the presence of an unvalued morpho-syntactic feature f (inserted in h via 48), which agrees with a matching xp/x and leads to the creation of a copy at h (with subsequent interpretational and pf realization effects). Since not any feature may be inserted in h, this captures why not any xp or x may participate in the complementary distribution pattern discussed in section 3.

Analysis
The Edge Requirement ensures that (the valued feature in) h be realized by (the overt copy of) an x or xp merged at h (see also Collins 2007;Landau 2007;Nchare & Terzi 2014), where h extends tp (see Roberts 2010 for x and xp being able to extend the root): The Edge Requirement is a pf requirement: (a) it marks the edge of tp (via phonological realization), possibly for linguistic (e.g. prosodic) and/or extralinguistic (e.g. processing) factors, and with pragmatico-semantic correlates (cf. the theme-rheme division);17 (b) it is satisfied locally (i.e. by a copy at h) and once (only the spec or the head of hp may be realized); (c) if hp cannot be realized overtly, pf spells-out the copy of t on h (via c-selection) as a last resort. If the phonological exponent realizing h is affixal, obligatory head movement ensues (no complementary distribution pattern; see next section).18 Let us now see how the Edge Requirement derives the finite enclisis patterns in cg (irrelevant details are omitted): 4.2.1.1 (Preverbal) wh-phrase: proclisis An unvalued [wh] feature is inserted in h and agrees with the wh-phrase. The latter is merged in spec hp (possibly for independent pf reasons; see Richards 2016), and spelled-out at h (in which case the Edge Requirement is trivially satisfied, t is spelled out in situ, and proclisis obtains): 17 For similar ideas from v2, see Mohr (2009);Zwart 2005. 18 An anonymous reviewer asks how the analysis proposed here differs from the prosodic account offered by Revithiadou (2006Revithiadou ( , 2007. Although both analyses assume that pf has a filtering role only, according to Revithiadou any differences between proclisis and enclisis triggers is related solely to their distinct prosodic properties (contrary to the analysis defended in this paper). That an identical syntax-distinct prosody account does not work is made evident by both optional complementisers (see section 3.2.2) and by dlinked wh-phrases (see sections 3.1 and 3.2.1) (the very same empirical phenomena used by Revithiadou herself), which have distinct syntactico-semantic properties that naturally map onto distinct prosodic structures. All the above cases involve an x or xp spelled-out at h. If nothing gets merged and/or spelled-out at h, enclisis obtains. This is illustrated below with Left Dislocated topics and verb-initial sentences.

4.2.1.4
Clitic left dislocated topics: enclisis As illustrated in (55) below, a Clitic Left Dislocated topic is merged in a position higher than h, presumably due to the fact that [top] cannot be inserted in h (this empirical fact is not well understood but seems to hold across many languages): 19 In the case of min, which is obligatorily selected by na, the simplest hypothesis would be to assume that it is generated in a separate modal head c-selecting h {neg} . Thanks to the editor for pointing out this issue to me.
Journal of Greek Linguistics 17 (2017) 190-232 morphology. As far as I know, there is no Tobler-Mussafia language having a true imperative/non-finite non-affixal head that blocks enclisis (although suppletive heads, such as subjunctive or deontic modality negation particles, do have this function (see Isac & Jakab 2004), in which case we would expect them to show root properties). This means that imperative/non-finite morphology in h will normally trigger enclisis. The second locus involves the Edge Requirement and where it applies (h or some other head). In particular, if h = t in morpho-syntactic and/or prosodic terms, we would get a language that lacks finite enclisis (cf. sg and most standard Romance languages). In this case, t would be predicted to have h properties, either morpho-syntactically and/or prosodically, with repercussions on the (a-vs. a-bar) nature of constituents merged in [Spec tp]. Although there seems to be evidence that t may indeed have morpho-syntactic and prosodic properties located in h in languages like sg (see, e.g., Alexiadou & Anagnostopoulou 1998 on Greek a'-subjects; Richards 2016 on t being the edge of a prosodic domain in null subject languages), additional research is required to understand what the exact matching of morphosyntax and prosody is, and how it fits into the present theory. This is beyond the purposes of the current paper, so I leave it open for future research (Mavrogiorgos, in progress). Summing up, in this section I argued that h seems to have an additional morphological/affixal requirement, as manifested by non-finite enclisis in cg, sg and mg. I also offered some speculative remarks on possible loci of parameterization for h, which may account for the lack of finite enclisis in languages like sg.

Conclusion
The main claim of this paper was that cg and sg enclisis are amenable to a decompositional analysis of the epp. In particular, h, the first head c-selecting tp (presumably the locus of theme-rheme division), contains an unvalued morpho-syntactic feature that agrees with a matching x or xp item. h also has a contingent pf requirement imposed on it (the Edge Requirement), according to which h must be spelled-out locally and only once by a non-affixal (x or xp) item. This derives the complementary distribution pattern attested in cg finite enclisis and mg imperative enclisis, as well as the lack thereof in cg and sg non-finite enclisis. The novelty of this approach is that in order to account for finite enclisis it postulates a pf part for the epp (see also Revithiadou 2006Revithiadou , 2007  available literature, which assumes that the epp is purely morpho-syntactic in nature, potentially applying to any feature within any functional head (see, e.g., Rivero &Terzi 1995, andTerzi 1999a,b for cg;Benincá 2006;Poletto 2014 for Old Italian and Medieval Romance; Fernández-Rubiera 2013 for Asturian; Roberts 2012 for Tobler-Mussafia languages and Slavic languages). One advantage of the current approach is that it derives all the attested patterns on the basis of a wellgrounded hypothesis (see Landau 2007; see also Mavrogiorgos, in progress, for additional evidence in favour of a local pf epp). A second advantage is that the hypothesis that the epp applies to h is simpler and more easily falsifiable, compared to an approach that postulates that the epp is a feature potentially applying to any functional head/feature (as, e.g., Landau 2007).
From an empirical point of view, an equally important contribution is that it sets apart proclisis triggers from enclisis triggers in terms of their hierarchical position in the clausal structure. This approach is novel, and has led to the empirical generalization in section 3.2.2 and to the formulation of an explicit theoretical hypothesis about the epp vis-à-vis clausal architecture. Moreover, it has direct implications for the status of non-stressed preverbal subjects in cg as topics (which could be informative also for sg, which lacks finite enclisis), as they behave like preverbal object topics with respect to enclisis (which, in turn, predicts that they should have additional topic properties, such as referentiality and wide scope, and moreover that subjects lacking these properties would trigger proclisis).21 Further issues arise at this point: how is cg enclisis related to second position cliticization or to v2 phenomena, which also involve a c-t interaction vis-à-vis the epp?; are there other domains with the same epp requirement, and if yes, what are these and what are their properties?22 Although these issues are relevant to the proposed analysis, they cannot be addressed appropriately within the limits of the current paper.