Two ways of encoding location in Greek: Locative applicatives and prepositions

Cross-linguistically, oblique theta roles such as location can be encoded by both adpositions and applicative morphemes. In this paper we argue that Standard Modern Greek (SMG), a language that encodes location primarily with prepositions, has a set of morphologically complex predicates that consist of an intransitive verbal root and a locative prefix, and behave like locative applicative constructions. We argue that this prefix is a low applicative head, licensing the addition of a locative DP argument to the intransitive verbal root. Specifically, this applicative head: (i) case-and theta-licenses the added argument, but being void of phi-features, it blocks its cliticization; (ii) is distinct from a homophonous free standing P semantically and syntactically; (iii) is undergoing grammaticalization, as evidenced by the emergence of a novel configuration, in which the locative predicates combine with locative PPs that retrieve semantic and syntactic information of the locative prefix. Our findings show that applicatives may come in various flavors, that a language may use both analytic and synthetic devices to encode location which are not derivationally related, and that lexical/inherent case does not necessarily reduce to a PP structure.


Introduction
Cross-linguistically, oblique theta roles such as location can be encoded by both adpositions and applicative morphemes, namely valency increasing morphemes that attach on the verb (Polinsky 2013). The two devices are illustrated below with English and Kichaga, a Bantu language spoken in Tanzania (Bresnan & Moshi 1990). (1) He is eating food at home.
(2) (Bresnan & Moshi 1990: ex. 3b & 1 respectively) a. N-a-i̋-lyì-í-à m̀-ṛì-nyì k-élyà. foc-1s-pr-eat-ap-fv 3-homestead-loc 7-food 'He/She is eating food at the homestead.' b. N-a-i̋-ly-à k-élyà. foc-1s-pr-eat-fv 7-food 'He/She is eating food.' In the English example (1), the location of the eating event is introduced by the preposition at, whereas in the Kichaga example (2a), it is introduced by the applicative morpheme í. The latter increases the valency of the verb to which it attaches, by enabling the addition of an extra locative object. Thus, whereas the simple verbal predicate in (2b) is a transitive verb followed by a single object (the patient DP kélyà 'food'), the derived verbal predicate in (2a) is a ditransitive followed by two objects: the patient DP kélyà 'food' and, in addition, the locative DP mrìnyì 'homestead'.
We will argue that Standard Modern Greek (SMG), an Indo-European language that encodes location primarily with prepositions, has a restricted set of predicates belonging to a formal register -such as iperiptame 'to fly over' in (3) -that behave like locative applicative constructions. (3) Ena sminos apo F-16 iperiptate tis polis mas. A cluster.nom from F-16 over.fly.3sg the town.gen our.cl.gen 'A cluster of F-16 is flying over our town.' These predicates share a unique cluster of properties: Morphologically, they consist of an intransitive verbal root and a locative prefix that is homophonous with an independently available locative preposition. Syntactically, they take an optionally realized object argument that realizes genitive case, encodes the theta role location, and resists cliticization and passivization.
We argue that the locative prefix of these predicates functions like an applicative head in the sense that it attaches to an intransitive root and licenses the addition of a locative DP argument. Specifically, we argue that the applicative head has the following properties: (i) It case-and theta-licenses the added argument, but being void of phi features, it blocks its cliticization; (ii) It is distinct from its homophonous free standing P, as shown by a range of interpretational and syntactic differences that cannot be accounted for under the assumption that they are the same lexical item; (iii) It is undergoing grammaticalization, as evidenced by the emergence of a novel configuration, in which the locative predicates combine with locative PPs that retrieve the semantic and syntactic information of the locative prefix.
The paper is of descriptive, typological, and theoretical importance: At a descriptive level, it offers a thorough description of a set of complex locative predicates that have received little attention in the Greek literature. 1 In this respect, in addition to enhancing our understanding of how location is encoded in Greek, it opens up the possibility for future cross-linguistic comparisons with complex locative predicates in other languages (see, for instance, Wilhelm 2007 and references therein for German and Svenonius 2004b and references therein for Slavic languages). At a typological level, it complements the inventory of Greek applicatives, which, so far, has been shown to include null applicatives introducing goals, benefeciaries, and maleficiaries (Anagnostopoulou 2003;Georgala 2012;Michelioudakis 2012), and it, furthermore, contributes to existing debates concerning their adpositional nature. In this regard, it compares two analytical possibilitiestheir analysis as incorporated adpositions (Baker 1988;1996;Zeller 2006) with their analysis as functional verbal morphemes (Marantz 1993;Baker 1996;McGinnis 2001;Anagnostopoulou 2003;Pylkännen 2008) -and proposes a number of diagnostics that allow us to differentiate between the two (see also Zeller 2006 andKim 2014, for a discussion along these lines). One of the main conclusions of this comparison is that the same language may use homophonous analytic (i.e., prepositions) and synthetic devices (i.e., applicative morphemes) of encoding location that -despite their homophony -cannot be reduced to the same lexical item. Finally, the paper informs our theory, by showing that applicative heads do not only differ with respect to their structural position (see Pylkkänen's 2008 analysis of low and high applicatives; McGinnis 2008), but also with respect to their featural composition. The latter, we argue, can account for the observed variation of applied objects in case, theta role, and cliticization.
The discussion is structured as follows: After giving background information on the origin and properties of the locative predicates and their arguments (section 2), we provide evidence in support of the claim that the predicates are prefixed by a non-adpositional, applicative-like morpheme (section 3). In section 4, we relate the properties of the locative arguments to the properties of the locative predicates. Specifically, we argue that the lack of cliticization can be reduced to the featural composition of the locative applicative/prefix, whereas the lack of passivization is reduced to the independent fact that our predicates are unaccusatives. Finally, in section 5, we discuss an alternative configuration suggesting that the locative prefix/applicative is undergoing grammaticalization, and in section 6, we conclude the discussion.

The properties of locative predicates and arguments
As mentioned in our introduction, SMG has a restricted set of locative predicates -listed in Table 1 Table 1 is based on an exhaustive search of Babiniotis (1998), a dictionary of SMG. Special mention needs to be made to two complex locative predicates that have been excluded: proiγume 'to precede' and iperexo 'to be better than'. Unlike the predicates under consideration, proiγume has a transitive verbal root. A similar complication is posed by iperexo, as it is not clear to us, whether its root is associated with the transitive/possessive exo 'to have' or with the intranstive/existential exo 'there is'. Given that the intransitivity of the verbal root plays a crucial role for the claim we wish to make, we will have to assume that these two verbs are amenable to a different analysis. 3 The free morpheme iper 'in favor of' has been analyzed both as an intransitive locative preposition (Lechner & Anagnostopoulou 2005) and as a locative modifier modifying a null NP Place (Terzi 2010). In this paper, we will be calling it a preposition for simplicity. However, the analysis that we will put forward is not, in principle, incompatible with its treatment as a modifier.
These predicates are of Ancient Greek origin and were introduced into SMG through the formal register, known as katharevousa. Some of these predicates are more frequent, and morpho-syntactically more productive than others. For instance, verbs such as iperisxio 'to prevail', ekserxome 'to go out', proiparxo 'to preexist', or epizo 'to survive' are quite frequent (as a preliminary Google search reveals), and may inflect for all person-number combinations, in most tenses, and in both perfective and imperfective aspect. 4 On the contrary, verbs such as afistame 'to abstain from', katisxio 'to prevail', or ipolipome 'to be inferior to' are less frequent, and are fairly restricted morpho-syntactically. 5 Despite the above differences, all these predicates show a unique cluster of morphological, semantic, and syntactic properties: In terms of morphological composition, they all consist of an intransitive verbal root and a locative morpheme, which according to Ralli (2005: 42-47) synchronically functions as a prefix (see also Ralli 2004 for the same point for prefixes of Ancient Greek origin more generally).
The verbal root can be synchronically bound (hence its meaning is derived in association with the prefix), as the root -tero in ipertero 'to prevail', or unbound, as the root iptame 'to fly' in iperiptame 'to fly over '. 6 As to the locative prefix, it is always homophonous to an independently available formal locative preposition. 7 As far as their semantics is concerned, the locative prefix can be interpreted literally, as in epiveno 'to ride on', and/or metaphorically, as in epilamvanome 'to take on'. Typically, a literal vs. metaphorical reading of the prefix correlates with a literal vs. metaphorical reading of the root, giving rise to a gradation of compositionality with the less compositional readings being more common with synchronically bound roots.
Finally, with respect to their syntactic subcategorization, locative predicates take two arguments: A nominative figure/theme and a genitive locative argument (or, in Talmy's 1978;2000 terms, a ground argument), whereby the location expressed by the locative argument is predicated of the moving figure /theme.
With all the verbs, except epiveno 'to ride on', iperaminome 'to defend', ipolipome 'to be inferior', and afistame 'to abstain from', the ground argument is optional. For example, epiveno 'to ride on' in (4) takes the figure/theme argument o proθipurγos 'the Prime Minister' and the obligatorily realized ground argument tu oximatos 'the vehicle ', whereas iperisxio 'to prevail' in (5)  The theta role of figure/theme is complex, in the sense that it conveys both an entity that is located in relation to a stationary location (figure) and an entity that moves or is movable (and therefore ends up being located in relation to a stationary location) (theme). The theta role of ground, on the other hand, may involve both literal and metaphorical extensions (see Jackendoff 1983 for temporal meanings and Beavers 2011 for property scales). In all these uses, the locative predicate relates two entities, so that one entity (the moving/movable entity, or figure/theme) is positioned with regard to some other entity (the fixed one, or ground) in terms of place, time, hierarchy, or more generally some scalar property. 8 For example, in the case of epizo 'to survive', the predicate relates two entities on a time scale. Thus, in (6) below, Janis is located on a time scale with respect to a particular event (namely, an accident), such as Janis still lives after the culmination of the event.
(6) O Janis epezise tu ðistiximatos. The Janis.nom survived.3sg the accident.gen 'John survived from/after the accident.' In the case of iperaminome 'to defend', the scale implicated could be a scale of preference (cf. the meaning of iper 'in favor'). Under this view, Janis in (7) is located with respect to tu nomu 'the law' on the positive side of a scale (assuming a negative-positive scale).

(7)
O Janis iperaminete tu nomu. The John.nom in.favor.of.defend.3sg the law.gen 'John is defending the law.' What is of primary importance to us is that despite the fact that the scale may differ (literal location, temporal location, hierarchy, change), the syntax is shared across all these predicates.
Shifting our attention from the locative predicate to the locative argument, it is clear that the latter shows an equally interesting cluster of properties that sets it apart from other types of objects in Greek: First, it realizes genitive case differing in this respect from both direct objects, which realize accusative (8a), and from indirect objects whose genitive case is traced back to historical dative (8b Second, as already pointed out, it encodes the theta role location/ground, unlike direct objects that are typically themes and patients, and, unlike indirect objects, that are typically goals/sources and beneficiaries/maleficiaries. Third, it resists cliticization (9a) (see Holton et al. 2004) as well as any other pattern that is contingent on cliticization, including Clitic Doubling (CD) (9b) (Anagnostopoulou 2003), Clitic Left Dislocation (CLLD) (9c), Clitic Right Dislocation (CLRD) (9d), and Resumption (9e) (Daskalaki 2008;Daskalaki & Mavrogiorgos 2013 Finally, as illustrated in (11), the locative argument cannot undergo passivization (see Anagnostopoulou 2003).
(11) *To oxima epivenete apo ton proθipurγo. The vehicle.nom on.ride.pass.3sg from the prime.minister.acc 'The vehicle is ridden on by the Prime Minister ' The same applies to indirect/dative objects (12a), although direct/accusative objects are passivizable (12b). 10 One could argue that the ungrammaticality of (9) is due to the fact that the locative object is inanimate and genitive clitics are generally dispreferred with inanimate referents (cf. Terzi 2010). Although an animate genitive clitic renders the sentence slightly better, the resulting sentence is nevertheless ungrammatical, and definitely much worse than the corresponding sentence without a clitic: These properties, in turn, combined with the Ancient Greek origin of the predicates under consideration, pose an interesting question regarding the dilemma the language learnerand the linguist, for that matter -is faced with: What is the most appropriate way to analyze such structures within the SMG system? More specifically: Is the locative prefix syntactically and semantically active? In other words, does it have case and theta assigning properties?
(16) Is it the same lexical item with its homophonous locative preposition?
And, finally: (17) What is the correlation -if any -between the properties of the locative prefix and the lack of cliticization/passivization displayed by the locative arguments?
To anticipate our analysis, we provide evidence indicating that the prefix of the locative predicate is a semantically and syntactically active applicative head (section 3.1) that is synchronically distinct from its homophonous locative preposition (section 3.2). Furthermore, we argue that both the lack of cliticization and the lack of passivization derive from properties of the functional projection of the applicative predicate (section 4). We assume that speakers reach this analysis by learning the specific semantic-syntactic content of the locative prefixes (a Lexicon issue), as well as by using the tools that already exist in the system (including applicative/valency-increasing morphemes, genitive case arguments, roots and prefixes that exist independently in the language, albeit possibly with distinct properties). In section 5, we show that competition with independently available, more productive means of encoding location (such as locative/directional PPs and predicates without a locative prefix) leads to semantic and syntactic bleaching which deprives the locative prefix of these special properties. The end result of this process is that location ends up encoded on the root and/or a PP (depending on the properties of the root), on a par with more productive structures. This process is common in case two or more partially overlapping means of encoding the same syntactico-semantic notion co-occur in the same grammar, and in this respect locative predicates behave in a quite regular fashion.

The locative prefix is an applicative morpheme
Before motivating the claim that the locative prefix of our predicates is an applicative morpheme, it is important to spell out our working definition of applicatives. This is because in the literature, applicative morphemes have been treated both as valency-increasing and as function-changing morphemes (see Haspelmath 2002 for discussion). Under the former definition, an applicative increases the valency of the verb to which it attaches by adding a new object argument (Bresnan & Moshi 1990;Beck 2009). The added argument expresses an oblique theta role -such as recipient, beneficiary, instrument, and location -and typically, though not always, displays direct object properties. Under the second definition, an applicative, rather than altering the number of objects, promotes an oblique/indirect object to a direct object (Peterson 2007).
Our claim here is that the locative prefixes of the Greek verbs under consideration behave similarly to applicatives of the valency-increasing type, in the sense that they license the addition of an object argument. Specifically, they attach to monovalent, intransitive verbs and they give rise to bivalent, transitive predicates that may take an optionally realized locative DP. This is clearly shown by the contrast in grammaticality between (18) and (19): (18) a. O Kostas epofeliθike (tis katastasis). The Kostas.nom on.profited.3sg (the situation.gen) 'Kostas profited from/took advantage of (the situation).' b. O proθipurγos eksilθe (tis vulis). The prime.minister.nom out.came.3sg (the parliament.gen) 'The Prime Minister came out of (the parliament).' c. To aeroskafos iperiptate (tis polis). The airplane.nom over.fly.3sg (the town.gen) 'The airplane flies over (the town).' d. To provlima proipirxe (tis krisis). The problem.nom pre.existed.3sg (the crisis.gen) 'The problem existed before (the crisis).' (19) a. O Kostas ofeliθike (*tis katastasis). The Kostas.nom profited.3sg (*the situation.gen) '*Kostas profited (the situation).' b. O proθipurγos ilθe (*tis vulis). The prime.minister.nom came.3sg (*the parliament.gen) '*The Prime Minister came (the Parliament).' c. To aeroskafos iptate (*tis poleos). The airplane.nom fly.3sg (*the city.gen) '*The airplane flies (the city).' d. To provlima ipirxe (*tis krisis). The problem.nom existed.3sg (*the problem.gen) '*The problem existed (the crisis).' What the above data show is that whereas the locative DPs are perfectly grammatical with the complex predicates (18), they are ungrammatical with the corresponding bare roots (19). This, in turn, suggests that the locative DPs are contingent on the locative prefix for their case and theta interpretation. 11 11 An anonymous reviewer asks whether it is possible for applicative morphemes in SMG to attach to bivalent bases. As is made evident by the comparison between (i) and (ii), this appears to be the function of the prefix pro-in protimo 'to prefer': (i) Timame ton kathijiti mas. Honour.1pl the professor. acc our.cl.gen 'We honour our professor.' (ii) Protimame ti ðoksa tu kerðus. prefer.1pl the fame.acc the fortune.gen 'We prefer fame over fortune.' What the above examples show is that a bivalent root timo 'to honour' (i) may become trivalent when prefixed with pro-(ii). The added argument is a ground, in the sense discussed in section 2, and realizes It could be counterargued, that the added locative DP is an adjunct rather than an argument. This would be consistent with its oblique properties (optionality and lack of cliticization/passivization), which, in Greek, are typically associated with uncontroversial adjuncts. For instance, as shown below, a temporal adjunct such as tu xronu 'next year' in addition to being optional, it fails to be cliticized and passivized: (20) a. (Tu xronu) i epitropi θa ðosi to vravio tis Marias. (The year.gen) the committee.nom fm give.3sg the prize.acc the Maria.gen '(Next year) the committee will give Mary the prize.' b. Tu xronu i , i epitropi θa (*tu i ) ðosi The year.gen the committee.nom fm (*him i ) give.3sg to vravio tis Marias. the prize.acc the Maria.gen 'Next year i , the committee will give (*it i ) Mary the prize.' c. *O xronos tha ðothi tis Marias/ The year.nom fm give.pass.3sg the Maria.gen / sti Maria to vravio. to.the Maria.acc the prize.acc. '*Next year will be given Mary the book/the book to Mary.' A number of considerations, though, render this alternative analysis hard to maintain: To begin with, even though locative DPs are only optionally realized, they are always implied by the complex predicate. Compare, for example, (21a) with (21b): a. O Panaθinaikos θa iperisxisi (tu Olimbiaku). The Panaθinaikos.nom fm prevail.3sg (the Olimbiaku.gen) 'Panaθinaikos will prevail (over Olimbiakos).' b. O Panaθinaikos θa iperisxisi (tu xronu). The Panaθinaikos.nom fm prevail.3sg (the year.gen) 'Panaθinaikos will prevail (next year).' Whereas both the locative tu Olimbiaku 'over Olimbiakos' (21a) and the temporal tu xronu 'next year' (21b) are optional, only the locative DP is implied by the complex predicate iperisxio 'to prevail' and, more precisely, by its locative prefix iper 'over', which somehow suggests the presence of a location/ground. The semantic content of the latter is contextually determined, in the sense that it receives a deictic interpretation. 12 Genitive case. We should point out though, that in Standard Modern Greek, (ii) sounds kind of obsolete and the Genitive DP tu kerðus 'the fortune' is most commonly replaced by a PP (apo to kerðos 'from the fortune'). Why the applicatives under consideration are synchronically more frequent with intransitive roots and whether the same was true in previous stages of the language is an interesting question for further research. We thank the anonymous reviewer for pointing out this to us. 12 In this respect, locative predicates remind us of ditransitives such as stelno 'to send' (i) and of comparative adjectives such as psiloteros 'taller' (ii) that also allow deictic implicit objects (a recipient and a ground, respectively): (i) Stilame (tu Kosta) to paketo. Sent.1pl (the Kosta.gen) the package.acc 'We sent Kostas the package.' (ii) O Janis ine psiloteros (tu Kosta). The Janis.nom is taller.nom (the Kosta.gen) 'Janis is taller (than Kostas).' Implicit arguments are also licensed by consumption verbs such as troo 'to eat' (Bhatt & Pancheva 2006): (iii) O Janis efaje. The Janis.nom ate.3sg 'John ate.' However, as correctly pointed out to us by an anonymous reviewer, in the case of consumption verbs the implicit argument is existentially quantified. That is, it does not necessarily refer to a specific entity (on the distinction between deictic and existentially quantified implicit arguments, see Lasersohn 1993).
A second difference between locative DPs and adjunct DPs concerns their distribution. As shown in (18)-(19) above, locative DPs are sensitive to the morphological composition of the selecting predicate. This kind of restricted distribution, which correlates with properties of the verbal predicate, is typical of DP arguments, but atypical of DP adjuncts. For example, the temporal DP adjunct tin Kiriaki 'on Sunday' is grammatical both with the complex predicate ekserxome 'to come out of' (22a) and with the bare verbal root erxome 'to come', contrary to the locative DP which is only available with the complex predicate (22b). (22) a. Θa ekselθi tu nosokomiu tin Kiriaki. fm out.of.come.3sg the hospital.gen the Sunday.acc 'He will come out of the hospital on Sunday.' b. Θa elθi (*tu nosokomiu) tin Kiriaki. fm come.3sg (*the hospital.gen) the Sunday.acc 'He will come on Sunday.' Further evidence corroborating the argument status of locative DPs comes from the observation that, unlike adjuncts, they have their case and theta role determined by the predicate they complement -and more precisely by its locative prefix. For example, whereas the complex predicate ekserxome requires a locative DP surfacing in genitive case (23a), it does not impose any restrictions on the case morphology of the temporal adjuncts tin Kiriaki 'on Sunday', which realizes accusative, and tu xronu 'next year' (23b), which realizes genitive.
(23) a. Eksilθe tu nosokomiu /*to nosokomio. Out.came.3sg the hospital.gen /*the hospital.acc 'He came out of the hospital.' b. Θa ekselθi tu nosokomiu tin Kiriaki /tu xronu. fm out.come.3sg the hospital.gen the Sunday.acc/the year.gen 'He will come out of the hospital next year.' Accordingly, whereas the spatial interpretation of the genitive DP in (24) varies depending on the predicate it complements -it is a locative, when it complements the locative predicate ekserxome 'to come out of' (24a) and a recipient, when it complements the ditransitive ðino 'to give' (24b) -the interpretation of the temporal adjunct DP tin Kiriaki 'on Sunday' remains constant. This shows that the locative argument bears a specific theta-role, which modifies the complex predicate (and, obviously, the prefix part, given our arguments earlier -see examples (18) & (19)), whereas a locative adjunct optionally modifies the whole event.
To the above arguments, we may add the evidence coming from the do so test, originally discussed by Lakoff andRoss (1976 [1966]) and widely applied since then (Schütze 1995;for Greek, see Anagnostopoulou 2003;. According to this test, the do so pro-form may replace a VP, provided that the latter is headed by a non-stative verb. However, whereas the internal arguments of the VP are obligatorily deleted, its adjuncts can be exempted from deletion. This is so because adjuncts are attached outside the VP, whereas arguments are sisters to the V head. What is of relevance for our purposes is that, once again, locative DPs pattern with arguments rather than with adjuncts. Thus, the examples below show that whereas it is possible to exempt a temporal adjunct DP from deletion (25), not deleting a locative DP leads to ungrammaticality (26).
O Olimbiakos iperisxise tu Panaθinaiku to Savato The Olimbiakos.nom prevailed.3sg the Panaθinaikos.gen the Saturday.acc ke to iðjo ekane ke i AEK tin Kiriaki. and the same did.3sg and the.nom AEK the Sunday.acc 'Olimbiakos beat Panathinaikos last Saturday and AEK did so last Sunday.' O Olimbiakos iperisxise tu Panaθinaiku The Olimbiakos.nom prevailed.3sg the Panaθinaikos.gen ke to iðjo ekane ke i AEK (*?tu PAOK). and the same did.3sg and the.nom AEK (*?the.gen PAOK) 'Olimbiakos beat Panathinaikos and AEK did so (*over PAOK).' In view of the above properties, we will conclude that the locative DPs under consideration are arguments licensed by the presence of a locative applicative morpheme (i.e., by the locative prefix) and that even when they are not phonologically realized, they are syntactically represented and pragmatically recovered. 13

The locative prefix is distinct from the locative preposition
Having established that locative prefixes show an applicative like behavior -and thus are semantically and syntactically active -we may now move on to our second research question concerning the relation between the locative applicatives and the homophonous locative prepositions.
There are two main approaches in the literature regarding this question. On the one hand, Baker's (1988;1996) transformational analysis treats applicatives as incorporated adpositions. Within this analysis, therefore, applicatives and adpositions can be reduced to the same lexical item. A different view is taken by Marantz (1993), McGinnis (2001), and Pylkkänen (2008), among others, who -differences aside -treat applicatives and adpositions as distinct lexical items (the former being analyzed as v-heads). 13 It is important to acknowledge that some of the predicates under consideration have obligatorily intransitive uses with different semantic and syntactic properties. Compare, for example, the intransitive epikrato in (ia) with its transitive variant illustrated in (ib): (i) a. Epikrati panikos *(tis lojikis). Prevail.3sg panic.nom *(the logic.gen) 'Panic ensues.' b. Epikrati o panikos (tis lojikis). Prevail.3sg the panic.nom (the logic.gen) 'The panic prevails over the logic.' The syntax and semantics of the intransitive epikrato is different in two ways: First, it does not license an overtly realized ground (compare (ia) with (ib)). Second, it does not seem to involve comparison. These differences suggest that we are dealing with two different predicates that are amenable to two different analyses: An intransitive predicate with a syntactically inactive prefix (this is the case with epikrato in (ia)), and a transitive predicate with a syntactically active prefix (this is the case with epikrato in (ib)). In the first case, the prefix epi-is empty (or part of the root) and does not license a ground argument. In the second case, the prefix epi-is syntactically active and licenses a ground argument, which is present in the syntax, independently of whether it is phonologically realized or not. We would like to thank an anonymous reviewer for bringing to our attention the obligatorily intransitive use of epikrato 'to ensue'.
Given that the locative prefixes and the locative prepositions are homophonous, the null hypothesis would be that they are the same lexical item and that an analysis along the lines of Baker is more suited for the data under consideration. However, as shown below, a number of considerations render this analysis hard to maintain. 14 To begin with, the locative prefix and the locative preposition show several semantic differences that are surprising under the assumption that they are the same lexical item. For example, there are cases where the meaning of the preposition is more restricted compared to the range of meanings associated with the prefix. This is for instance the case with the preposition iper and the prefix iper-. In Ancient Greek, the preposition iper could take on both the literal interpretation 'over' and the metaphorical one 'in favor of'. The same was possible for its homophonous prefix (Liddell Scott Jones 1996). In SMG, though, the situation is different. Whereas the prefix iper-can take on both literal (27b) and metaphorical interpretations (27a), the preposition iper has retained only its metaphorical interpretation 'in favor of' (28). An additional problem for the treatment of locative prefixes as locative prepositions derives from the morphosyntactic differences between the two elements. As we have already seen, the locative prefixes always introduce an optional DP that realizes genitive case and resists cliticization. Their corresponding locative prepositions, though, show less homogeneous patterns. First, whereas most of them assign genitive case to the argument they license (29 a-d), there are two prepositions (ipo 'under' and apo 'from') that assign accusative (29 e-f). 15 (29) a. minima ek tu proeðru message from the president.gen 'Message from the President' b. Ine epi tis oðu Skufa. Is on the street.gen Skufa.gen 'It is on the Skufa street.' 14 It is important to clarify here that Baker's approach does not exclude, in principle, the co-existence of various types of applicatives (adpositional and non-adpositional ones). In this respect, our data do not constitute an argument against his theory overall; rather, the claim is that the locative predicates discussed here are not amenable to such an analysis. 15 The only examples where ipo 'under' and apo 'from' assign genitive are in the idiomatic expressions ipo malis 'under the armpit' and apo stiθus 'by heart', respectively. However, these PPs, unlike the PPs considered in the main text, are fossilized expressions and, consequently, unlikely to be manipulated by morphosyntactic rules such as incorporation. Third, most of the locative prepositions (31) differ from locative prefixes in that they require obligatory DP arguments. Specifically, of the locative prepositions we are examining, only iper 'in favor of' has optionally realized arguments (31f).
(31) a. minima ek *(tu proeðru) message from *(the president.gen) 'Message from *(the President)' b. Ine epi *(tis oðu Skufa). Is on *(the street.gen Skufa.gen) 'It is on *(the Skufa street).' c. i pro *(tu Aθo) perioxi the before *(the Aθo.gen) area.nom 'The area before (Athos)' d. ipo *(to eðafos) under *(the ground.acc) 'Under (the ground)' e. apo *(to nosokomio) from *(the hospital.acc) 'From *(the hospital)' f. iper (tu nomosxeðiu) in.favor.of (the bill.gen) 'In favor of (the bill)' The above set of morphosyntactic differences resists a straightforward explanation within an analysis that treats the prefixes as prepositions: If the locative prefixes and the locative prepositions were the same item, we would expect them to impose the same morphosyntactic restrictions on their arguments. As shown above, this prediction is only partially born out.
Finally, to the above counterarguments, we may add the absence of a grammatical input. Specifically, under an analysis that treats the locative prefix as an incorporated preposition, one would expect the existence of a grammatical analytical structure that could serve as the input to incorporation. 16 Synchronically, this is true only for two of our predicates: proiparxo 'to preexist' and ekserxome 'to come out of'. To illustrate with the former verb, the VP proiparxo tis krisis 'to exist before the crisis' (32a) has a perfectly grammatical analytical variant consisting of the bare verbal root iparxo 'to exist' and the PP pro tis krisis 'before the crisis' (32b): (32) a. To provlima proipirxe tis krisis. The problem.nom pre.existed.3sg the crisis.gen b. To provlima ipirxe pro tis krisis. The problem.nom existed.3sg before the crisis.gen 'The problem existed before the crisis.' However, this is not the case with the remaining complex predicates. For instance, as shown in (33), the analytical counterpart of the VP epiveno tu oximatos 'to ride on the vehicle' -consisting of the verbal root veno 'to develop' and the locative PP epi tu oximatos 'on the vehicle' -is, synchronically, unacceptable: Once again, this is hard to explain under the preposition incorporation analysis: if the prefix were incorporated from inside a PP complement, the underlying structure should be acceptable, contrary to fact. 16 An anonymous reviewer points out that a language may have obligatory incorporation for a class of cases, which would explain the lack of alternation between free and incorporated forms. We think that this depends on one's theoretical assumptions. Under such an analysis, we would have to accept that incorporation may have an ungrammatical input. Furthermore, we would have to explain why the incorporation of a free form (a P) into another free form (a V) is obligatory in these particular cases.
To conclude, a number of empirical problems suggest that the locative prefixes cannot be treated as locative prepositions. It is important to clarify, though, that our conclusion holds for SMG. Whether or not the two items are historically related and whether or not in Ancient Greek locative applicatives were prepositional in nature requires a thorough investigation of Ancient Greek predicates and prepositions that is beyond the scope of this paper. 17

Analysis: The locative prefix and the locative DP argument
In the previous section, we provided empirical evidence in support of the claim that the locative prefix of Greek complex predicates is an applicative head, homophonous though distinct from the corresponding locative preposition. Our aim in this section will be to develop an analysis that in addition to capturing the "valency increasing" effect of the applicative morpheme, accommodates the "curious properties" of the added argument summarized in (14)  To this end, we will first motivate our assumptions and we will, then, show how our analysis derives the properties listed in (34).

Assumptions
We will be following Pylkkänen's (2008) account of low applicatives in assuming that the locative applicative under consideration is originally introduced as the head of an ApplP that complements the Root. Evidence that it behaves like a low applicativerather than as a high applicative attached above the verb phrase -comes from the observation that it denotes a spatial relation between two individuals (a figure/theme and a ground) and that, in addition, it introduces an argument that resists secondary predication (35): (35) *Iperisxise tu andipalu tis eksandlimenu. Prevailed.3sg the opponent.gen her.cl.gen exhausted.gen '*She prevailed over her opponent exhausted.' In both respects, it differs from high applicatives, which, in Pylkkänen's (2008) typology, express a relation between an individual and an event, and introduce arguments that are amenable to secondary predication.
Furthermore, we will be assuming that the Root and the applicative head combine via conflation (see Hale & Keyser 2002), due to the affixal nature of the applicative head and/ or the Root. 18 With respect to the exact categorial status of the Appl head, two possibilities have been suggested in the literature: It could be a v-head or a p/P-head. Under the p/P analysis, the applicative head is a Place Prepositional head, further embedded within a p-head, as illustrated in (36) The role of the p-Head would be to introduce the figure/theme argument and to case license the ground argument (see e.g. Jackendoff 1983; van Riemsdjik 1990; den Dikken 2010; Koopman 2000/2010; Svenonius 2010; Terzi 2010). Even though both analyses are compatible with the essentials of our proposal, it appears to us that the second one is not supported synchronically by independent evidence. Thus, it is not clear why the presumably prepositional applicative is always affixal 20 and why it fails to license any additional material despite the presence of multiple specifiers. In this respect, we may compare (37a) with (37b).
(37) a. *To aeroplano iperiptate ekato metra tis polis. The airplane.nom over.fly.3sg hundred meters.acc the town.gen 'The airplane flies a hundred meters above the town.' b. To aeroplano petai ekato metra pano apo tin poli. The airplane.nom fly.3sg hundred meters.acc above from the town.acc 'The airplane flies 100 meters above the town. ' The examples show that the applicative iper-'over' fails to license the degree phrase ekato metra 'a hundred meters' (37a), differing in this respect from the free standing Path PP pano apo tin poli 'over the town', which, having a complex structure, may license the degree phrase in question (37b).
Summing up, in view of the above evidence, we will be assuming that the locative prefix is a low vAppl head -rather than a p/P head -that combines with the root via conflation.

Derivation
Having motivated our assumptions, we are now in the position to discuss the details of the proposed derivation, which we depict in (38) with the sentence Proeðrevi i Maria tu sineðriu 'preside.3sg the Maria.nom the conference.gen ': 18 Recall that the prefix is only homophonous with the free standing P. Because it always appears attached on a root, we can safely assume that it has an inherent affixal property. 19 Depending on the semantics of the predicate, the postulation of an additional Path Prepositional head would also be possible. 20 Compare, for instance, the grammaticality of the affixal variant of iper 'over' in iperiptame 'to fly over' with the ungrammaticality of the free variant in *iper tis poleos 'over the city'. (38) In the emerged configuration, the DPground is introduced in the complement position of the applicative head, whereas the DPfigure/theme is introduced in its Specifier. This has a number of advantages. First, we capture straightforwardly the spatial relation that exists between the DPfigure/theme and the DPground (see Svenonius 2004a;. In particular, the DPground expresses a stationary entity (i.e. an entity whose location is fixed in relation to a particular landmark) in relation to which the path, site, or orientation of the DPfigure/theme (a moving or movable entity) is set. This (locative and/or directional) relation is mediated via the Appl head. Second, we can account for the fact that the DPfigure/theme asymmetrically c-commands into the DPground (but not the other way round) (see Anagnostopoulou 2003;  Third, by positioning the two arguments in the same locality domain, we immediately explain why permutation/A-scrambling of the DPground across the DPfigure/theme is possible: If β c-commands α, and τ is the target of movement, then β is closer to τ than α unless β is in the same minimal domain as Specifically, it corresponds to (41a), where both the DPground (XP) and the DPfigure/ theme (Spec1) are contained within the same minimal domain, and, consequently, the former can move across the latter. In addition to the above advantages, the suggested configuration explains the properties of locative arguments listed in (34). Let us consider them in turn: (i) Case and Theta Role: We propose that the case and the theta role of the DPground are assigned by the applicative head, which is itself specified for (lexical) case (namely, genitive) and theta role (ground/location). The genitive case is lexical in the sense that, rather than being related to a structural relation such as Agree, it is related to selection by a particular set of lexical predicates: those containing the locative prefix/ applicative head. As to the DPfigure/theme, we propose that each component part of the complex theta-role figure/theme is assigned at a distinct specifier (following in this Ramchand 2008). In particular, the DPfigure/theme receives the figure interpretation in [Spec ApplP], which allows it to be interpreted as an entity located in relation to the DPground. It then moves to [Spec, RootP], where it receives the theme interpretation, which allows it to be interpreted as the moving/movable entity that undergoes the action denoted by the predicate. This two-step derivation captures the fact that it is the presence of the locative prefix that licenses the figure interpretation of the theme. When the prefix is absent, then, only a theme interpretation is possible. The DPfigure/theme then moves to the specifier of an intransitive v, 21 where it receives its Nominative case via 21 Note that in terms of event structure this intransitive v may be a BE (stative) or BECOME or CAUSE (eventive) predicate (alternatively, a structure may involve more than one v's, neither of which has accusative case/agentive features). This possibility ( Agree with a T head merged higher up. Note that because the highest v-head is intransitive, it does not carry any case/phi-features that could assign structural (accusative/ genitive) case to any of the two internal arguments (on the different flavors of v crosslinguistically, see Arad 1999;for Greek, see Alexiadou, Anagnostopoulou & Schäfer 2006). 22 (ii) Lack of Cliticization and other clitic patterns: To explain the lack of cliticization (and related clitic patterns such as resumption, CD, CLLD, and CLRD, which are all contingent on the presence of a clitic head merged as the highest D-head within the extended nominal projection -see Daskalaki & Mavrogiorgos 2013), we will assume that the applicative head lacks phi-features. In the absence of phi-features, the Agree relation between the applicative head and the (clitic inside the) locative DP fails, and the impossibility of cliticization comes as no surprise (for the claim that cliticization is contingent on Agree, see Mavrogiorgos 2010;Roberts 2010 In (b), the locative prefix, being void of phi-features (a lexical issue), fails to Agree with the clitic-head, which carries an uninterpretable case feature. As a result, cliticization is ruled out. This hypothesis also explains the fact that direct/accusative and indirect/dative objects can be cliticized, as they both Agree with a v-head carrying phi-features.
(iii) Lack of Passivization: The lack of passivization, we suggest, can be related to the fact that our predicates are unaccusatives. 23 Given that unaccusatives are intransitive [-accusative] predicates, the impossibility of passivization comes as no surprise. Evidence for the unaccusativity hypothesis comes from the observation that all of our predicates licence bare plural subjects, patterning in this respect with unaccusative predicates. 24 This is illustrated below with ekserxome 'to come out of': Example (i) denotes that John is better than Mary now and that this state has been true before (e.g. we may assume that John is a better student than Mary this year and that he was a better student also last year). On the other hand, example (ii) could be felicitous in two situations: (a) If John did something again (e.g. he ran in the local marathon) and as a result he prevailed over Mary (although he has not won over her before); and (b) if John did something yesterday (e.g. he ran in the local marathon) and as a result he prevailed over Mary again (he had won over Mary also last year, when both participated in a baking contest). 22 An implication of an intransitive v is that promotion of the oblique argument to a direct object would still be out, independently of whether the prefix and the free standing P were the same lexical item or not (as promotion is contingent on the presence of a structural case assigning v-head). 23 This is not a surprising correlation from a typological perspective, as there are reported cases of unaccusative applicative predicates resisting passivization. For example, Bresnan & Moshi (1990) report that Chichewa has a small class of unaccusative applicative verbs such as gwera 'to fall into' that contrary to other applicative verbs in the language fail to passivize.
(i) (Bresnan & Moshi 1990: ex 80 (a,c)) a. Mbuizi y-a-gw-er-a m-chi-tsime. 9 goat 9 sb-perf-fall-appl-ind 18-7-well 'The goat has fallen into the well.' b *M-chi-tsime mw-a-gw-er-edw-a' ndi mbuzi. 18-7-well 18 sb-perf-fall-appl-pass-ind by 9 goat 'The well has been fallen into by a goat.' (43) Ekserxonde vuleftes tu ktiriu tis vulis. Out.come.3pl members.of.parliament.nom the building.gen the parliament.gen 'Members of parliament are coming out of the parliament building.' Admittedly, once we take semantic criteria into consideration, the picture gets less clear. On the one hand, most of our predicates belong to semantic classes that have been shown to have an unaccusative syntax cross-linguistically (see e.g. Levin & Rappaport-Hovav 1995). That is, they are either verbs of existence (proiparxo 'to pre exist', epizo 'to survive', ipertero 'to prevail', iperisxio 'to prevail', epikrato 'to prevail') or verbs of inherently directed motion (ekserxome 'to come out of'). 25 On the other hand, there are three predicates whose semantic class has been linked to unergative syntax (epiveno 'to ride on' is a manner of motion verb, iperiptame 'to fly over' is a maintain spatial configuration verb, while proeðrevo 'to preside' is an agentive activity verb) and two predicates that are reflexives (iperaminome 'to defend' and epofelume 'to profit from'). 26 Independently of whether we decide to analyze these problematic verbs as unergatives or unaccusatives, what is of relevance for our purposes, is that under either analysis, the root is embedded under a [-accusative] v. Significantly, this line of reasoning is compatible with the fact that passivization in Greek is only possible with transitive predicates with a [+accusative/+agentive or +causative] v (see Tsimpli 1989;Anagnostopoulou 2003).
(iv) Optionality: The final aspect of our analysis that needs to be discussed concerns the intuition that the ground DP is implied, even when it is not phonologically realized. This, we suggest, could be attributed to the fact that the DPground is syntactically projected independently of its phonological realization (for a discussion on null Grounds in English see Svenonius 2004a and references therein; for a discussion of implicit arguments and of the debate surrounding their syntactic representation, see Bhatt & Pancheva 2006;Landau 2010).
All in all, in this section we have proposed an analysis which takes the locative prefix to be an applicative affix that introduces an optional DPground. In structural terms, the ground is in the complement position of the applicative head, while the figure/theme DP is in its specifier. In featural terms, this applicative head is special in that it does not carry phi-features (hence the lack of cliticization), but assigns lexical/genitive case to its ground complement. On the other hand, the figure/theme DP is assigned nominative case via AGREE with the higher T head.
Before concluding, it is worth making a digression to motivate an aspect of our analysis that could, in principle, be amenable to a different explanation than the one adopted here. We are referring to our account of agreement failure.
(i) ??Tinos tha epofelithi o aðelfos tu neu nomu jia ta kokina ðania? Whose fm profit.3sg the brother.nom the new law.gen for the red debts.acc 'Whose brother will profit from the new law on bad mortgages?' 25 It is not clear to us which semantic class ipolipome 'to be inferior ', afistame 'to abstain', and proistame 'to preside' belong to. The bound root -istame denotes spatial configuration (= 'to stand', either with a maintain position reading or a simple position reading -the former is considered in the literature to correlate with unergativity, the latter with unaccusativity), while -lipome is a bound deponent root with an obscure meaning within the Modern Greek system. 26 Note that the status of reflexives as unergatives or unaccusatives is far from settled in the literature, as some consider them unaccusatives, some unergatives, and some mixed (see e.g. Alboiu, Barrie & Frigeni 2004;Chierchia 2004;Reinhart & Siloni 2004).

Agreement failure
In theory, agreement failure could be attributed either to the functional projection of the verbal predicate -this is the view taken here -or to the functional projection of the nominal argument. Under the former view, AGREE (in the sense of Chomsky 2001) fails, because the functional projection of the verbal predicate lacks an appropriate probe with phi features (Daskalaki & Mavrogiorgos 2013). This is illustrated in (44)

*AGREE
In order to decide between the two possibilities, we need to provide independent evidence for/against the postulation of a null P. To this end we will discuss the evidence coming from three commonly cited diagnostics: wh-extraction, locative inversion and coordination. As it will become clear, all three diagnostics suggest that locative arguments in Greek pattern with DPs rather than with PPs. We will begin with the evidence from wh-extraction. It has been argued that sub-extraction out of a null PP is blocked and that this explains the impossibility to extract a goal argument in English (see e.g. Baker 1988): (46) *Who did you send a letter?
The idea here is that goal arguments are headed by a null P, which prevents its DP complement from being extracted. It follows, that if the locative arguments under consideration were PPs they should resist wh-movement. Given that the prediction is not borne out (47), the postulation of the null P cannot be maintained: 28 27 This theoretical possibility builds on a long tradition in the literature maintaining that datives -and possibly genitives -are always c-commanded by a (null/overt) P (see Emonds 1985;Bittner & Hale 1996;Řezáč 2008;Caha 2009, among many others). The presence of P has been used to explain various opacity effects observed with oblique arguments, and has led various researchers (see e.g. McFadden 2004;Landau 2009;Baker in press) to propose the reduction of inherent and/or lexical case to the presence of P. As we will see later on, an absolute reduction is not supported by the particular constructions reported in this paper. This suggests that locative lexical case in SMG cannot be reduced to an underlying PP structure. 28 At first glance, Free Relatives (FRs), that is embedded A'-movement constructions that inherit the syntactic category of the moved pronoun, challenge the conclusion that locative arguments pattern with DPs. Thus, one could argue that the reason why (i) is ungrammatical is because the FR pronoun opjon is subextracted out of a null P.
(i) *Sinexarika opjon iperisxises. congratulated.1sg who.acc prevailed.2sg 'I congratulated whoever you prevailed over.' A closer inspection of the data though reveals that its ungrammaticality is more likely related to the case conflict between the matrix predicate 'to congratulate' -which requires accusative -and the relative predicate 'to prevail' -which requires genitive. We know that different languages resolve case conflicts in FRs in different ways. What is of relevance for our purposes is that in Greek, the FR pronoun has to realize the case required by the matrix, whereas the case required by the relative predicate, if genitive, has to be resumed by (47) Tinos iperisxise o Nikos? who.gen prevailed.3sg the Nikos.nom 'Who did Nick prevail over?' Let us move on to the evidence from locative inversion. Landau (2009) has argued that locative PPs undergo overt/covert raising to [Spec, TP] in order to check a related feature on T. As a result of this, they display subject properties cross-linguistically. This observation suggests that if genitive grounds in Greek were PPs, they should display subject properties on a par with locative PPs across languages.
Once again, the prediction is not borne out. Greek genitive grounds, pre-verbal and post-verbal alike, fail a series of subjecthood diagnostics, including the ability to control PRO within an absolute construction. This is clearly shown below, where the ground argument tu Petru, independently of whether it is preverbal as in (48a) Landau's (2009) reasoning, the fact that genitive locative arguments do not display subjecthood properties suggest that they cannot be headed by a null P. Unfortunately, the evidence from this diagnostic is weakened by the observation that in SMG, overt PPs do not display subjecthood properties either (49) More helpful is the evidence coming from co-ordination. Thus, Jaeggli (1982), building on Vergnaud (1974), shows that coordinated PPs may not serve as the head of a relative clause which functions as a derived collective predicate. Based on this diagnostic, Anagnostopoulou (2005) shows that Greek goals/beneficiaries introduced by se 'to' are indeed PPs. This is because a conjunction of se goals can only marginally be assigned a group interpretation (50a). By contrast, a conjunction of DPs, headed by se, is well-formed under the same interpretation (50b).
(50) (Anagnostopoulou 2005: ex. 125a-b) a. ??Estilan γramata ston anðra ke sti γineka Sent.3pl letters.acc to.the man.acc and to.the woman.acc pu zusan mazi. that were.living.3pl together 'They sent letters to the man and to the woman who were living together.' b. Estilan γramata ston anðra ke ti γineka Sent.3pl letters.acc to.the man.acc and the woman.acc pu zusan mazi. that were.living.3pl together 'They sent letters to the man and the woman who were living together.' What is of interest for our purposes is that Greek genitive grounds behave on a par with DPs with respect to this diagnostic: Stus aγones epikratise tu anðra ke tis γinekas In.the games.acc prevailed.3sg the man.gen and the woman.gen pu zusan mazi. that lived.3pl together 'In the games, she prevailed over the man and the woman who were living together.' In view of the problems associated with the postulation of a null P, we will conclude that the properties of Greek locative arguments (genitive case, ground theta role, and lack of passivization) are related to the featural composition of the locative applicative. 30

PP Alternates
In our discussion so far, we have focused on locative predicates taking locative DP arguments and we have argued that they are prefixed by a semantically/syntactically active applicative head. In this section, we will consider the emergence of a novel, less formal configuration, in which the same predicates are followed by locative PPs rather than by locative DPs. In these cases, we argue, the prefix of the locative predicate has undergone a significant degree of grammaticalization, in the sense that it has lost its syntactic (case and theta assigning) and semantic (spatial) properties (see Dimela & Ralli 2012 for the properties of grammaticalization in Greek prefixation; see also Narrog & Heine 2011 for an introduction into the various aspects of grammaticalization phenomena). Grammaticalization has resulted in the creation of a novel verb with different syntax and less transparent semantics. In what follows, we will first present the data and we will then provide evidence in support of our hypothesis.
As already mentioned, all the predicates discussed in section 2 allow their locative DP to be replaced by a prepositional PP. The two possibilities are illustrated with proeðrevo 'to preside over' in (52), whereas a complete list of the attested possibilities is given in Table 2. (52) a. Proeðrevi tu simvuliu. Preside.3sg the meeting.gen 'She presides over the meeting.' b. Proeðrevi sto simvulio.
Preside.3sg at.the meeting.acc 'She presides at the meeting.' For each one of the locative predicates, Table 2 provides the prepositional variants reported in Babiniotis (1998) -a well-regarded dictionary of SMG -as well as those ones attested in recent online documents (primarily newspapers). The latter ones were gathered through a Google search and are illustrated with complete (numbered) examples in the Appendix (X denotes unavailability). 31 The hypothesis that the prefix of predicates combining with PPs has undergone grammaticalization is supported by a number of considerations. First of all, cross-linguistically, overtly realized applicative heads introduce, to the best of our knowledge, DPs. Thus, arguing that in examples such as (52b) above, the locative prefix is an applicative that licenses a locative PP would be odd from a typological perspective. 32 But even if we set the cross-linguistic perspective aside, there is language specific evidence suggesting that the prefix of predicates combining with PPs is syntactically/ semantically impoverished compared to the prefix of predicates combining with DPs. First, the locative PP shows a wider distribution than the locative DP. As illustrated in (53) below, the PP apo ti vuli 'from the Parliament' is equally grammatical with the complex predicate ekserxome 'to come out' (53a) -that consists of the prefix ek-'out' and the verb erxome 'to come' -and with the simple predicates erxome 'to come' (53b) and vjeno 'to come out' (53c). to combine with locative DPs. This is, for instance, the case with kataferome 'to express oneself negatively against', which at least etymologically consists of the prefix kata 'towards/against' and the verb ferome 'to bring myself': (i) a. ??*Kataferete tis kivernisis. Against.bring.himself.3sg the government.gen b. Kataferete kata /enandion tis kivernisis. Against.bring.himself.3sg against(formal) /against(informal) the government.gen 'He expresses himself negatively against/attacks (verbally) the government.' the expression of the locative relationship by means of independently available analytic expressions (i.e. PPs). 36

Conclusions
To conclude, SMG, an Indo-European language which normally uses a P predicate to express location, has a restricted set of morphologically complex predicates that belong to a formal register and that behave like locative applicative constructions. In particular, these predicates involve an overt locative prefix which, when added to an intransitive root, introduces an optional locative DP argument with lexical genitive case and the thetarole Ground. We have argued that the locative prefix cannot be identified with an independently available homophonous free P (although the two might be related at earlier stages of the language, an issue for further research). This fact combined with the lack of an underlying PP input and the fact that the root is not an accusative case assigner makes a P-incorporation/promotion analysis extremely unlikely. Regarding certain morpho-syntactic properties of the locative argument (such as lack of cliticization or passivization), we argued that these are related to the featural make up of the applicative prefix (and not to the structural/PP status of the locative argument). Finally, we showed that locative predicates may also combine with a PP argument or adjunct (depending on the predicate), which constitutes a novel, informal, analytical environment, and which correlates with loss of semantic and syntactic content from the locative prefix (via grammaticalization).
These findings are of typological interest as they show that SMG, in addition to the relatively well studied type of null applicatives that encode the theta roles of goal, maleficiary and beneficiary (Anagnostopoulou 2003;Georgala 2012; Michelioudakis 2012), has a restricted set of overt locative prefixes that display an applicative-like behaviour. How exactly these prefixes compare with overt locative applicatives that are productive in African and Native American languages (see, e.g. Bresnan & Moshi 1990 for Bantu languages; Beck 2009 for Salish) and how common it is for languages to combine overt analytic (adpositions) and synthetic devices (applicatives) of encoding location are interesting questions that require a systematic cross-linguistic comparison. Given though that both devices have the same function, it seems reasonable to hypothesize that languages that make an equally productive use of both them are less common than languages that only use one of the two, or both but in different degrees.
In addition to its typological interest, our study has interesting theoretical implications. Thus, it suggests that applicative morphemes do not only vary in terms of their structural position within the clause (high vs. low), but also in terms of their featural make-up (see e.g. Anagnostopoulou 2003; Greek low applicatives; Arad 1999 for flavors of v cross-linguistically). This is compatible with the Chomsky-Borer conjecture that linguistic variation is located in the functional Lexicon, and it follows directly from the nature of AGREE.
Equally importantly, our analysis implies that not all oblique arguments taking inherent and/or lexical case (also known as semantic case) can be analysed as PPs. In particular, we argue that lexical case is assigned to an argument when the latter is selected by a particular flavor of v head, an assumption that is fully compatible with structural approaches to case (e.g. Marantz 1992). Of course, this does not entail that there are no instances of inherent and/or lexical case which may be reducible to the categorial status of the casemarked DP. Rather, it means that what renders a DP structurally case-marked or not (in terms of its morpho-syntactic behavior, such as passivization, cliticization, agreement, etc.) may be related to more than one structural strategy (e.g. an applicative morpheme, a P/p predicate, or a K morpheme). In this case, the question that arises is whether all these strategies are reducible to a single underlying strategy or not, and if yes in what way exactly.
Accordingly, the development of novel uses for the locative predicates shows that both applicative morphemes and P/p predicates (and K morphemes, for that matter) contain semantic/syntactic information, which may license a spatial argument. What varies is the morphological realization (which is language specific, and clearly influenced from diachrony), but also the domain in which such a morpheme belongs to (e.g. nominal domain, verbal domain, or an independent domain projected by an additional predicate). 37 In this regard, further empirical and theoretical evidence may be provided by a cross-linguistic comparison of locative morphemes (including, affixes, particles, case, and P/p-predicates), as they seem to differ morpho-syntactically and semantically in various respects (see e.g. Svenonius 2004b for aspect and case in Slavic prefixes; Wilhelm 2007 for German particles).