2.1. A crash course on the Greek prepositional system

Lechner & Anagnostopoulou (2005) classify Greek prepositions into transitive (9)–(10) and intransitive (11). Transitive prepositions obligatorily take an accusative DP-complement and can be further subdivided into functional (or light) (9a), which can serve as complements of other prepositions, thus forming complex PPs (9b), and lexical (10a), which cannot (10b):⁶

The light prepositions in (9), particularly se and apo, have a wide range of uses that are associated with Case: se has a dative function, marking goal arguments of ditransitive verbs and some types of high applicatives (Anagnostopoulou 2003; 2005), whereas apo marks source arguments of ditransitives and nominal possessors, gradually replacing the genitive (Anagnostopoulou 2003; 2005). Moreover, apo is found in comparatives (Merchant 2009) and distributives (Michelioudakis 2020).

Intransitive prepositions may appear with or without a complement, which can be either a light PP (11a–b), or a genitive DP (11c):

The intransitive prepositions that participate in the formation of spatial expressions, a list of which is given in (12), are also referred to in the literature as “adverbs” (Theophanopoulou-Kontou 2000), a term commonly found in traditional grammars, and as locative [P]s (Terzi 2010). In fact, these “prepositions” appear in a wide range of environments: not only as parts of complex PPs (11), but also as standalone locative adverbs/pro-forms (13a), verb particles (13b), and nominal modifiers (13c):

The function and the syntactic status of these multi-faceted items are different in each case. Moreover, as is argued in this paper, their syntactic status varies within the spatial domain as well, correlating with the semantic type of the expression. For this reason, they will henceforth be referred to by the more abstract and atheoretic term spatial words. The small caps in the gloss indicate their “primitive” spatial meaning.⁷

2.2. Simple and complex locative PPs

Simple locative PPs are formed by se and the ground DP:

The exact interpretation of these PPs is contextually determined, i.e. it depends on world knowledge and properties of the figure and the ground. For example, a bottle of milk is more likely to be found in a fridge, not on one, whereas a book may occasionally be lying on top of a fridge but is very unlikely to be found in it. The denotational range of simple se-PPs is limited to topological (14a–b) and proximity or underspecified relations (14c–d). In order to disambiguate between topological relations (15a) or express projective ones (15b), complex PPs must be used. The latter can be divided into complex se-PPs (15a) and complex apo-PPs (15b) based on the light preposition marking the ground:

Some spatial words can combine with either se or apo. Depending on the spatial word, the se-compound and the apo-compound may be synonymous, as in (16a–b), or express different spatial relations, as in (17a–b) (cf. Terzi 2010):⁸

Although the two kinds of complex PPs are similar on the surface, previous research has posited the existence of two distinct syntactic frames underlying complex PPs in Greek. Theophanopoulou-Kontou (2000) argued for two distinct constructions, one in which the “adverb” (i.e. the spatial word) can be omitted, and one in which it cannot. This observation, when applied to the locative domain, leads to the following generalization: in complex se-PPs, the spatial word can be omitted (18a), whereas in complex apo-PPs, it cannot (18b):

If the spatial word is omitted from a complex se-PP, as in (18a), the remainder is a simple se-PP, which is still a locative, i.e. the type of the spatial expression is not altered. However, if the spatial word is omitted from a complex apo-PP, as in (18b), the locative meaning is lost. Theophanopoulou-Kontou (2000) argued that the “adverb” is a head if it cannot be omitted (18b). Following Starke (1993), she proposed that the light PP is a case-marked DP, the light preposition itself being a nominal C(omplementizer) head:⁹

Although Theophanopoulou-Kontou (2000) does not discuss constructions in which the spatial word can be omitted, i.e. cases like (18a), in depth, she suggests that in the latter, the “adverb” specifies the meaning of the light-PP, with the light P being responsible for theta-role assignment (e.g. as locative in (18a)).

Terzi (2010) takes a different approach and proposes that spatial words are modifiers of a silent Place noun, as in (20). Terzi (2007; 2010) and Botwinik-Rotem & Terzi (2008) suggest that the role of the light prepositions se and apo is to check Case on the ground DP because the spatial word fails to do so, thus dissociating se and apo from theta roles:

However, in (20), complex se-PPs are not distinguished from complex apo-PPs in terms of syntactic structure. Everything else being equal, this predicts that both constructions should behave alike. Section 3 shows that this prediction is not borne out, with complex se-PPs differing from complex apo-PPs in a number of ways.

The account developed in Section 4 builds on Theophanopoulou-Kontou (2000)’s core idea that the syntactic status of the spatial word may vary and incorporates elements of Svenonius (2008; 2010)’s analysis of projectives. Specifically, it is proposed that, in complex se-PPs, the spatial word is an adjunct to a KP headed by se, whereas in complex apo-PPs, the spatial word is a head taking a KP complement headed by apo. This head is related to the vector semantics of projective expressions (Svenonius 2008; 2010), which for this reason are syntactically larger than non-projective ones. Moreover, the evidence discussed below supports Terzi (2007; 2010)’s view that se and apo are case-markers that have no semantic contribution.

3.1. Measure-phrase and projective modification as diagnostics for projective PPs

Zwarts & Winter (2000) argued that MP modification is contingent upon the formal property of upward vector-monotonicity (VMON↑). Intuitively, upward vector-monotonicity means that, in a given spatial relation, if the figure moves away from the ground along an axis described by the preposition, the truth-value of the expression is preserved. Figure 2 illustrates this: behind the house, there are two objects that are located on the same axis: a tree and a doghouse, the tree being closer to the house. Because behind is a VMON↑ preposition, if (21a) is true, (21b) is also true:

Figure 2

a house (H), a tree (T), and a doghouse (H_D).

Zwarts & Winter (2000) classify all projective prepositions, i.e. above/over, below/under, in front of / behind, and beside, plus outside as VMON↑. Non-projective prepositions such as near, on, at, in, inside, and between are not VMON↑, and thus disallow MP modification. Among the prepositions studied by the authors, outside stands out as the only non-projective preposition which is VMON↑. Therefore, while not all prepositions that are VMON↑ are projective, the opposite is true, i.e. all projective prepositions are VMON↑.¹⁰

Another property of projective expressions is that they can be modified by the projective modifiers straight and diagonally, which are sensitive to axis information. According to Zwarts & Winter (2000), projective modification is possible only when a unique axis can be determined. This requirement is trivially satisfied by projective prepositions, such as above, which defines just one up axis, while the non-projective outside defines six (in a 3D space). However, outside can be modified by projective modifiers if it satisfies the unique axis determination constraint. For instance, when someone is waiting outside the building, the latter can be a paraphrase of in front of the building (cf. Mador-Haim & Winter 2015: fn. 10). In this case, projective modifiers are acceptable. Note, moreover, that there is a sharp contrast with regard to projective modification between outside (22a) and (the unambiguously topological) out of (22b):

The rest of this subsection uses MP and projective modification as diagnostics to detect projective PPs in Greek.

Table 1 provides a list of possible combinations of spatial words with se and apo, and their respective locative meanings:

A quick look at Column 3 is suggestive of the projective nature of complex apo-PPs: no combination of a spatial word with apo seems to have a topological or force-dynamic meaning. Instead, one mostly finds projective denotations with all possible frames of reference, e.g. brosta/piso apo ‘in front of / behind’ that use either the intrinsic or the relative frame, and voria/notia apo ‘north/south of’ that make use of the absolute frame.¹² On the other hand, Column 2 shows that complex se-PPs have non-projective denotations, such as topological (mesa se ‘in’, pano se ‘on’), force-dynamic (kontra se ‘against’), interpolation (anamesa se ‘between, among’), and proximity (konta se ‘near’) ones. There seem to be two apparent exceptions, namely brosta se ‘in front of’ and dipla se ‘by, next to’, which, as is shown below, behave like non-projectives according to the diagnostics.

If complex apo-PPs are projective and hence VMON↑, they are expected to allow MP modification. This holds for all compounds except for konta apo ‘not far from’, the canonical uses of makria apo ‘far from’, and gyro apo ‘around’ which are discussed next. Examples (23)–(24) illustrate:

An interesting case is mesa apo, in which the spatial word mesa ‘in’ combines with apo to form a projective. Unlike English inside, locative mesa apo is not compatible with containment within the ground, but rather denotes containment in an enclosed region beyond but adjacent to the ground, roughly translating as ‘on the other side of’ or ‘behind’. It can be modified by a measure phrase (24):¹³

The next diagnostic is projective modification. As is shown in (25), all the PPs in (23) and (24) can be modified by the projective modifiers isia ‘straight’, and diagonia ‘diagonally’, except for the absolute frame voria/notia/anatolika/ditika ‘north/south/east/west of’, with which diagonia is deviant.

All the complex apo-PPs so far examined have passed the tests and can thus be classified as projectives. There are a few apo-compounds, however, that do not qualify as projectives. The complex apo-PPs konta apo ‘close to, not far from’, accepted by some speakers,¹⁴ and makria apo ‘far from’ likely involve path denotations, as they are ambiguous between linear (air-distance) and non-linear (zig-zag) readings. The fact that their English counterparts include the preposition from, a source preposition, is suggestive. This assumption is further confirmed by the ill-formedness of projective modification:

MP modification is never allowed with konta apo, but is acceptable with makria apo,¹⁵ in which case, however, the inherent degree specification disappears (also note the different gloss):

Moreover, a contrast is found between konta apo and the standard proximity compound konta se. Konta se displays a sensitivity to the relative size of the figure and the ground (28a), which is not present with apo:¹⁶

Lastly, the complex apo-PP class contains route, or “extended” prepositional compounds, such as gyro apo ‘around’ (29a). Recall from (5) that paths may also denote the location or extension of the figure along the ground (Zwarts 2003; Svenonius 2010). As will be discussed in Section 3.3, the ground is marked by apo in non-projective route constructions. For this reason, some complex apo-PPs are ambiguous between a projective reading and a locative route (“extended”) reading (29b):

Moving on to complex se-PPs (Table 1: Column 2), we see that it is a smaller inventory. All its members disallow MP and projective modification. First, consider the topological compounds for containment (30a) and contact (30b). As expected, MP modification is out:

Note that the compound preposition mesa se in (30a) differs from the projective mesa apo of (24), in that in the latter, the figure is not contained in the ground, but in enclosed area beyond it, whereas in (30a), the figure is contained in the ground.¹⁸ However, there are a few cases in which MP modification with mesa se is felicitous:

These examples likely involve embedded paths (cf. Svenonius 2010). Firstly, in (31a), the MP 5 xiliosta ‘5mm’ measures out the extension of the figure into the ground, while (31b) has a zig-zag reading on which one has traveled a 100km distance from the border without being at a linear distance of 100km from it at the moment of utterance. Secondly, the MPs in (31) cannot be omitted: if the MP 5 xiliosta ‘5mm’ in (31a) were left out, the clause would mean that the whole nail is stuck inside the wall (32a–b). Thus, (31a) does not entail (32b). This is never the case with projectives: in (23), all MP-modified constructions entail their un-modified variants.

Turning to the rest of the se-compounds, the force-dynamic kontra se ‘against’, and the proximity-denoting konta se ‘near, close to’ are not expected to license a MP either, and indeed, they do not. The same goes for the interpolation compound anamesa se ‘between, among’:

As was mentioned above, there are two se-compounds that at first sight seem to have projective denotations: brosta se ‘in front of’ and dipla se ‘by, next to’. However, both fail to pass the measure phrase and projective modification tests (compare (34a–b) with (23a) & (25a)):

Furthermore, in some cases, brosta/dipla se seem to lack a projective component altogether. In examples (35a–b), there is a sharp contrast between brosta/dipla apo, i.e. the “real” projective prepositional compounds, and their se-counterparts: only the se-variants are felicitous. Interestingly, the same effect obtains in English, if one uses the counterpart projective prepositions in front of and beside instead of by:

Here, the se-variants seem to merely denote proximity rather than projective geometry. The difference between brosta se and dipla se in (35a) and (35b), respectively, is that with brosta, but not with dipla, the house has to be right on the seafront, not just close to the sea. As to why the projective readings are impossible even in English, a tentative explanation would be that the sea is conceptualized as an unbounded space, and consequently no front/back/left and right boundaries can be defined (this is not the case for under and above, however, as under/above the sea are fine). Moreover, brosta/dipla se pattern along the rest of the complex se-PPs with respect to the syntactic facts discussed in the next subsection.

Lastly, projective modification is impossible with se-compounds. Modification by isia ‘straight’ and diagonia ‘diagonally’ is acceptable in a few cases, in which, however, straight/diagonally modify the figure’s orientation, not its location:

Summing up, complex se-PPs universally disallow MP and projective modification, while complex apo-PPs allow both, modulo a few compounds that have path denotations. These are route prepositions such as gyro apo ‘around’, which would fall into Svenonius (2010)’s “extended” class, and makria/konta apo ‘far/ not far from, close to’, which likely involve source paths. The class of locative proper complex apo-PPs otherwise contains only projectives. On the other hand, complex se-PPs are “bounded”/non-projective according to the diagnostics. Their denotation range excludes projective relations, but is not uniform, including topological, proximity, force-dynamic and interpolation relations.

3.2. Evidence for different syntactic structures

Recall from Section 2.2 that Theophanopoulou-Kontou (2000) argued that the possibility to omit the spatial word correlates with its syntactic status. If it is a head, it cannot be omitted, but if it is a modifier, it can. As was shown in (18), omission is possible in complex se-PPs but not in complex apo-PPs. Thus, the spatial word is an adjunct in non-projectives (37a), and a head in projectives (37b). While the possibility to drop the spatial word is not compelling evidence for its adjunct status on its own, this section presents additional evidence that this analysis is correct.

The first piece of evidence comes from movement of the light PP. Complex se-PPs allow topic or focus-related scrambling of the functional-PP to the left of the spatial word (38), while complex apo-PPs do not (39):

A similar effect is found with wh-extraction:

Under the assumption that the spatial word is a head in complex apo-PPs, the ill-formedness of scrambled (39b) and wh-fronted (41b) apo-PPs in projectives can be attributed to illicit preposition stranding. As the examples below illustrate, Greek does not allow P stranding. This is shown for the light (42) and lexical (43) transitive prepositions of (9)–(10):

Stranding is also ungrammatical in non-spatial complex PPs:

An anonymous reviewer comments that examples (39b) and (41b) become grammatical if modified by (ligo) pio piso/kato lit: ‘(little) more back’. The modified examples are reconstructed below:

I suggest that these constructions involve comparatives in which the apo-marked DP is not the ground, but rather the standard of comparison, which is marked by apo in both clausal and phrasal comparatives (see Merchant 2009). This is why non-fronted variants of typical projective expressions degrade radically when modified by (ligo) pio:

The modified version of (46a) has an odd reading in which both the cat and the bed are below a third object, with the cat at a greater distance from it. Similarly, the modified version of (46b) has a reading in which both the paintings and the piano are behind something else. Thus, when pio ‘more’ is present, the projective reading is often unavailable because the apo-PP is not interpreted as the ground. For the same reason, if piso ‘back’ in (45a) is substituted with kato ‘down’, as in (47), the reading in which the motorcycle is under the car is unavailable:

Crucially, apo-PPs in comparatives can be moved, as the following example shows:

Therefore, while spatial constructions involving comparatives seem to be orthogonal to the present discussion, they are interesting and should be looked at in future research.

Moving on to the second piece of evidence supporting the different syntactic status of the spatial word in complex se-PPs vs complex apo-PPs, we observe that they have different pro-forms. In the case of complex se-PPs, if the referent is known, the spatial word may be used on its own for anaphoric identification of the ground, as in (49):

This is not the case with complex apo-PPs, where omission of the functional apo-PP does not result in an anaphoric projective pro-form, but in a different interpretation. In (50b), piso means ‘in/at the back’, rather than ‘behind it’. In order to license an anaphora, the special pro-form apopiso must be used (50c):

All complex apo-PPs have a pro-form of the type in (50c), which is treated in Section 4 as incorporation of the light preposition apo into the spatial word (83). This is only possible in complex apo-PPs because the spatial word is a head. By contrast, non-projective pro-forms (49b) can be viewed as elliptical structures.

3.3. Directional expressions

The third piece of evidence that the spatial word is a head in the projective complex apo-PPs and an adjunct in the non-projective complex se-PPs comes from directional expressions. Directional expressions built on complex non-projective PPs pattern along simple PPs, whereas directional expressions built on projective PPs follow a distinct pattern.

All locative PPs in Greek are homophonous with their goal counterparts because Greek is a V(erb)-framed language in the sense of Talmy (2000). Recall from (14) that in simple PPs, the ground is marked by se. Consequently, a se-PP may either express a locative (51a) or a goal argument (51b).²⁰ The goal interpretation is licensed by path verbs such as pigeno ‘go’. Manner verbs like xorevo ‘dance’ do not admit a goal reading (52).

Simple source and route arguments are marked by apo. To determine whether an apo-PP stands for a source or a route, one has to look at the verb:

Thus, there is an one-to-many relation between the light P and the spatial types it can express. Although the specific interpretation is co-determined by the verb, the choice between a se-PP and an apo-PP itself is not an inherent selectional property of the verb. Both verbs of (53) may alternatively combine with se-PPs, giving rise to goal readings, as (54a–b) illustrate:

Complex non-projective PPs follow the same pattern as simple ones, i.e. goals are homophonous with their locative counterparts (55a–b), but in sources and routes, the light preposition changes to apo (55c–d).

Thus, the locative complex se-PPs of (55a) become complex apo-PPs in source and route environments (55c–d), with an additional instance of apo optionally preceding the spatial word. Crucially, the path is reflected (via the se/apo shift) to the right of the spatial word mesa, which modifies location. This is surprising, as there is significant consensus in the literature that path prepositions are syntactically higher than locative ones (Jackendoff 1983; Koopman 2000; den Dikken 2010; Svenonius 2010; Pantcheva 2011), as schematized in (56):

Under the assumptions: (a) that the order in (56) is (at least underlyingly) universal; and (b) that mesa se is a complex locative preposition, the expected order for Greek would be (57), which is ungrammatical:

In the grammatical examples (55c–d), the path is expressed both before (optionally) and after (obligatorily) the locative modifier. The optional instance of apo in non-projective sources and routes is a piece of evidence for the higher Path position, which is the locus of path interpretation. The lower instance of apo is then best viewed as a K(ase) head, agreeing with the spatial environment. In the same vein, se is also to be treated as a K head not tied to locative semantics (pace Theophanopoulou-Kontou 2000, who views it as a theta marker) because it disappears in certain types of path environments. This blends in well with the case-marking behavior of se and apo in other domains.

The assumption that the higher instance of apo is the real Path preposition is supported by paths built on projectives. Projective goal PPs are homophonous with their locative counterparts as expected (58a–b), while in sources and routes, a second apo obligatorily occurs before the locative compound (58c–d). The fact that this instance of apo is obligatory is expected because, in projectives, the ground is already marked by apo, and as such, realizing Path is the only way to disambiguate goals from sources.

In projectives, therefore, the ground is invariably marked by apo in all environments, while in paths built on non-projectives, be they simple or complex (55), the marker changes to apo in source and route environments. This is predicted if the spatial word is a head in projectives (locative complex apo-PPs) and an adjunct in non-projectives (locative complex se-PPs). The se/apo alternation can then be treated as a reflex of a Path head in a local relationship with a K head. In projectives, this is not possible because the spatial word intervenes. The details of the analysis are laid out in the next section.

4.1. Non-projectives

Section 3 presented evidence that the spatial word in a complex se-PP such as (59) is an adjunct, while the light preposition is a case marker. I propose that locative se-PPs are syntactically reduced. Specifically, simple se-PPs are bare KPs, while non-projective complex PPs are KPs modified by a spatial word (60):

In (60), KP is the projection headed by the case-markers se and apo. Svenonius (2010) suggests that K is the locus of eigenplace, the function that shifts the ground individual to the region it occupies in space (Wunderlich 1991). The Greek data does not support this view because there is no strict correlation between either se or apo and eigenplace across all complex PPs. In non-projectives, specifically, we saw that the choice between se and apo depends on the path environment (cf. (55)). Hence it would be unnatural to assume that K is tied to a semantic function such as eigenplace, while at the same time its realization is sensitive to the path context. However, eigenplace is independently motivated in the semantics. Therefore, we may assume that the type-shifting happens in the semantics only, as in Zwarts & Winter (2000)’s original proposal. Accordingly, K in (60) stands for the identity function.

Spatial words take the ground’s eigenplace and return a set of regions, which can be the ground’s interior, surface, neighborhood and so on. Under this analysis, vectors are not syntactically represented. This explains why (complex) se-PPs do not license MPs, and why there are no se-projectives. However, vectors could still be needed to define some relations between regions. I leave this question open for future research.

The figure is finally introduced by small p, a relational head which locates the figure in the space denoted by KP. I assume that there is only one kind of p across all locative expressions. This is a slight modification of Svenonius (2010), who proposes that containment and contact are alternative relational meanings contributed by p. Recall from (30) that in Greek, containment and contact are expressed by the compounds mesa se ‘in’ and pano se ‘on’, respectively. The spatial words mesa and pano are treated here as restrictive modifiers of the ground’s eigenplace because there is no evidence that they differ syntactically from the rest of the se-compounds. Note that an alternative analysis in which mesa and pano are modifiers of pP instead of KP would entail that the figure is introduced below them, which is unwarranted. Therefore, under the proposed analysis, spatial words in non-projectives are uniformly XP-modifiers of eigenplace. This also means that the denotations for containment and contact are geometric.

The definitions and a sample derivation are given below (abstracting away from the verb’s contribution):

Going through the steps in (63), the couch is first type-shifted from an individual to a region via the eigenplace function (63a–c). The non-projective modifier pano then takes the eigenplace and returns a property of regions that qualify as the couch’s surface (63d–e). Finally, small p introduces the figure, i.e. the cat, and existentially closes the region variable (63f–h). The resulting (63h) states that there is a region r such that the location of the cat is r, and r is the surface of the couch.²³

I assume that locative KPs can alternatively merge with Path, with the figure being introduced higher. KP is then interpreted as a start, end, or transitory location, according to the type of the Path head (goal vs source vs route). The se/apo alternation itself is a reflection of the relationship between K and either of these higher Path heads. Because in the present analysis p/Path and K are sisters, this can be modeled in terms of selectional features, but it can also be viewed as case assignment / feature valuation. For the present purposes, nothing hinges on this. The following examples illustrate: p and Path_goal take se (64), while Path_source and Path_route take apo (65):

I suggest that Path_goal has no exponent in (64) because it is licensed by path verbs, possibly via a checking/Agree operation between V and Path. Path_source/Path_route can be instead lexicalized by apo (65), but realization is optional in non-projectives because path is reflected on K. Moreover, realization of Path_source/Path_route is possible only in complex PPs because in the absence of a spatial word, there would be two adjacent occurrences of apo, i.e. * apo apo, which we may assume is blocked at PF.

The analysis of Greek non-projective complex PPs advanced here is similar to Pesetsky (2013)’s account of directional constructions in Russian. Pesetsky takes the locative preposition P not to be part of the extended projection but rather to be an adjunct to the ground DP (66). For Russian, this ensures that the DP is not assigned oblique case by P, but remains nominative (or is assigned the accusative by DIR_V if the ground is animate, feminine, or pronominal):

The counterpart of Pesetsky’s P in the present account is the spatial word qua non-projective modifier in (60), which also does not project. In both accounts, this allows the case of the ground DP to be unaffected by the locative modifier and remain sensitive to higher heads. Furthermore, in both languages, the locative modifier may appear as a head in a different construction. In Greek, the spatial word is a head in projectives, and an adjunct in non-projectives (in path and locative constructions alike), whereas in Russian, it is an adjunct in paths, and a head in locatives (projectives included).

Lastly, the adjunct analysis of the spatial word in non-projectives accounts for the ability of the ground KP to move without the spatial word (38)–(40). Example (67) shows this for scrambling/fronting of the se-PP, where KP₁ raises to a higher topic position:

4.2. Projectives

The analysis of Greek projective PPs builds on Zwarts & Winter (2000)’s vector account and assumes a decomposition similar to Svenonius (2008; 2010). Unlike in the non-projective PPs discussed in the previous section, the spatial word in projectives is part of the extended P projection. The structure of a projective PP like (68) is given in (69):

The following types and variables are used:

Firstly, the ground is type-shifted from an individual into a region via the eigenplace function (71), as in non-projectives.

The Project head is responsible for the vector semantics of projective expressions. There are different instantiations of Project corresponding to the varying directions up, down, front, back, left, right, etc.²⁴ These are realized by the respective spatial words. (73) illustrates this for piso ‘back’, which returns the vector space for ‘behind’. The entry for Project uses Zwarts & Winter (2000)’s definition. Measure phrases adjoin at the ProjectP level, restricting the length of the vectors. Thus, a measure phrase like deka metra ‘ten meters’ has the entry in (74).

The Region head shifts the vector space back to a region. The semantic entry for Region in (75) is a modification of Zwarts & Winter (2000)’s loc^– and returns the set of projected regions instead of introducing a figure and predicating its location.²⁵ This has the advantage that the projected region can be the argument of either a relational small p head introducing the figure, or a Path head, as was the case with non-projectives.²⁶

Note that RegionPs and non-projective KPs have the same semantic type, i.e. <<p, t>,t>, which is why both may combine with either p or Path. This opens the possibility that Region is nominal in its categorial features. While silent nouns have been proposed in previous analyses of Greek spatial expressions (Terzi 2010), they were not tied to projectives. In the current proposal, spatial words in non-projectives modify the ground’s eigenplace within the same syntactic projection, whereas in projectives, there are two regions, the ground’s eigenplace, on one hand, and the projected area, on the other, which project separately (in the syntax). A consequence thereof is that the ground DP in projectives is deeply embedded, and thus its case is unaffected by the higher Path or p heads. The ground is invariably marked by apo across all environments, which can be modeled as a selectional property of Project, or alternatively as inherent case assignment. The examples below illustrate:

Unlike in non-projectives, spatial words in projectives cannot be stranded because they are heads, specifically, instantiations of Project. Consequently, KP cannot move without pied-piping Project (cf. (39b)–(41b)), as is shown in (81):

Lastly, the proposed analysis provides a way to account for the projective pro-forms discussed in (50c), repeated below as (82). In (83), an empty/defective D moves and incorporates into Project via K in a roll-up fashion:

4.3. A note on insertion

In the analysis above, it was shown that the syntactic and/or categorial status of spatial words may vary. For example, in the projective compound pano apo ‘above’, pano realizes Project, while in the non-projective compound pano se ‘on’, it serves as the exponent of a topological modifier. Further uses of spatial words as verb particles and nominal modifiers were illustrated in (13). Even though the inventory of spatial words is small, being best described as a semi-closed class, their distribution is wide and too pervasive to be reduced to accidental homonymy. Instead, what seems to characterize spatial words in Greek is contextual allosemy (Marantz 2013). Allosemy refers to the property of having multiple meanings, while contextual implies that the choice among these meanings depends on the context. In a late-insertion framework such as Distributed Morphology, spatial words like pano can be treated as exponents of different nodes. The lexical information associated with each spatial word is represented only once in the lexicon as a root. Accordingly, the representation of pano in the lexicon would be √up. Following Embick & Marantz (2008), roots are acategorial and are categorized in the syntax. Project, which is a special type of a P head associated with projective semantics, is then a complex head which categorizes spatial roots such as √up. The semantic interpretation depends on the root and the syntactic context. In the context of Project, √up will be interpreted as ‘above’, while in the context of an adjectival categorizer, it will be interpreted as ‘upper’. As each functional head also bears its own features, this approach would allow for the different nodes to be realized by more specific vocabulary items. In Greek, however, there are no exponents associated with these features. For this reason, only the spatial word is realized in projectives, non-projectives, and spatial adjectives alike.