Cross-linguistically, there is a long-standing tradition of research (that goes back at least to Thomas 1931) on the alternation between constructions such as the judge’s decision and the decision of the judge, a phenomenon often called genitive alternation/variation, see Rosenbach (2014), Szmrecsanyi et al. (2017), Ackermann et al. (2018) and references therein. This phenomenon has been widely discussed as it bears on the interaction between syntax and morphology, and it offers a window to the factors regulating the inflectional marking of Case and the development of an adpositional alternative. One of the questions raised in the study of this alternation is whether there is one syntactic mechanism deriving both constructions and, thus, the alternation is determined morpho(phono)logically or whether each construction constitutes the realization of a distinct syntactic structure.

In this paper, we contribute to the investigation of the genitive alternation by introducing novel evidence from Greek. Despite the abovementioned long-standing tradition, the genitive alternation in the synchrony of Standard Modern Greek remains a relatively understudied topic that needs to be combined with existing diachronic and dialectal studies to provide a better understanding of the phenomenon. Unlike English, where there is alternation between prenominal and postnominal genitives that differ in form, ‘s-genitive and of-genitive, Standard Modern Greek shows an alternation between inflectional and prepositional forms, both appearing in the post-nominal position. Prenominal genitives are not licit in unmarked contexts in Standard Modern Greek.1 The examination of this alternation in Greek supports the hypothesis that genitive case-marked nominals are introduced into structures via a syntactic mechanism distinct from the one introducing their prepositional alternates.

Realizations of the genitive have attracted much attention in diachronic linguistics as they constitute a prominent and well-documented case of language change in Indo-European languages. Although the genitive was part of an elaborate Proto-Indo-European case system where it had a possessive and a partitive use, nowadays, it is entirely lost (e.g., in Romance) or appears in remnants (e.g., in Germanic), see Luraghi (2003), Hewson & Bubenik (2006), and Kulikov (2011; 2013) among others.

Greek is an Indo-European language that has undergone significant changes in its case system. Mertyris (2014; 2023) shows that the Ancient Greek genitive inherited both uses of its Proto-Indo-European ancestor but added the ablative use. Both the partitive and ablative uses of the genitive were lost in the Medieval period, during which the genitive retained only its possessive use. Meanwhile, the loss of inflectional dative had the genitive take over its uses, only in the Southern variety of Greek; Northern Greek replaced dative with accusative.

In the course of diachrony, it became impossible to express several of the genitive’s uses through inflection and hence these were realized by prepositional phrases instead. This is, for instance, what happened with the partitive function, which was expressed via a prepositional phrase headed by apo even in the Classical period, although the relevant examples in this period are still rare, see more in Seržant (2012), and Conti & Luraghi (2014):

Today, Standard Modern Greek still has an inflectional genitive, despite the fact that several dialects, mainly spoken in the North, have (almost) entirely lost it, and grammars of Heritage Greek speakers do not include it, at least to the same extent as the grammars of native Standard Modern Greek speakers, see Zombolou (2011). In these cases, inflectional genitives have been replaced by PPs headed by apo ‘of, from’, a situation reminiscent of Romance and other Indo-European languages, see Luraghi (2003) and Hewson & Bubenik (2006).

This paper focuses on the synchrony of Standard Modern Greek, where inflectional genitives exist but are becoming rarer as apo-PPs are gaining ground. Our study shows that the replacement of the genitive by apo-PPs is often triggered by paradigm gaps in the declension classes of the nominals but is restricted to specific contexts. This suggests that inflectional genitives are not being replaced by prepositional genitives but rather substituted by prepositional phrases that can convey the same readings. In other words, apo-PPs do not constitute an alternative surface form for inflectional genitives; they are introduced via a distinct syntactic mechanism.

The rest of this paper is structured as follows. Section 2 provides an overview of the genitive alternation in Standard Modern Greek, discussing its various syntactic and semantic properties. Section 3 examines the role of the paradigm gaps in the nominals’ declension system and the role of the language’s history in shaping the alternation pattern. Section 4 compares the Standard Modern Greek distribution with dialectal data, particularly from Northern Greek varieties, to assess the extent to which apo-PPs can be analyzed uniformly across varieties. Section 5 shows how Standard Modern Greek apo-PPs and inflectional genitives differ from each other, while Section 6 focuses on derived nominals and shows that when accompanied by apo-PPs, derived nominals do not support an argument structure comparable to the respective verb but pattern like non-derived, result nominals. Section 7 considers non-derived nominals and presents a syntactic analysis of the alternation, showing that the two constructions stem from distinct syntactic mechanisms rather than from morphological variation. Finally, Section 8 concludes with a discussion on the implications of these findings for diachronic change and cross-linguistic perspectives.

The genitive in Standard Modern Greek (SMG) marks, among others, indirect objects in the verbal domain as well as possessors and internal arguments in the nominal domain, see Holton et al. (2012: 336–351). Nikiforidou (1991) provides an overview of all uses of the SMG genitive and proposes a grouping via a network of metaphors. We will focus here on the genitive in the nominal domain, namely the genitive of possession (2a) and the genitive realizing the argument of derived nominals (2b).

Specifically, the genitive case expresses all types of possession, including ownership (2a), kinship (3a), part-whole relationships (3b), and body parts (3c).

In the meantime, genitive marks the theme argument of a derived nominal (4a). In the presence of a second argument, the theme argument retains the genitive-case marking, while the external argument, the agent, is introduced via a prepositional phrase headed by apo ‘of, from, by’(4b).

Apo-PPs that are morphologically similar to the ones introducing the external argument of derived nominals can also replace inflectional genitives in some of the abovementioned uses, as explained below. The use of such a PP instead of a genitive in SMG is associated with lower registers. Holton et al. (2012: 473) call this a colloquial alternative; see also at the end of Section 3.

Regarding possessives, apo-PPs alternate naturally with genitives in the context of part-whole relationships (5a), as is the case with source/origin relationships (5b). Relations between body parts and animates, though, do not alternate as easily (5c):2

Apo-PPs are unacceptable in kinship (6a) and ownership (6b). As pointed out by our reviewers, the evidence presented in (5) and (6) suggests that the animacy of the possessor creates a blocking effect on the alternation: apo-PPs do not seem acceptable with animate possessors. The rest of this Section supports this generalization. However, additional evidence discussed in Sections 3 (ex. 14, 15) and 5 (below ex. 27) suggests that animacy alone cannot explain the distribution; animate possessors are acceptable in apo-PPs if the relationship expressed is a (body) part-whole one.

As for derived nominals, prepositional phrases with apo can replace genitive theme arguments, as shown in (7). However, although very common in (but not exclusive to) culinary contexts, as a simple internet search confirmed, sentences containing apo-PPs are degraded for some speakers, including one of the authors of this paper and one of its anonymous reviewers. Our take on this issue is presented in Section 6:

Crucially, in the presence of two arguments, only the external one surfaces as an apo-PP (cf. 8a, 8c, 8d), even though the internal one can appear as a PP when in isolation (8b), see Alexiadou (2001) for discussion and references:

A further observation is that the prepositional variant is unacceptable when the possessor or the nominal argument is a pronoun4 (9) or a proper name (10). The sentences in (9) are acceptable under non-possessive interpretations, i.e., only if a thematic relationship is identified between the noun and the PP, e.g., if apo sena ‘of you’ in (9a) denotes someone who sent/donated to vivlio ‘the book’ or if apo mena ‘of me’ in (9b) means that the dance (to xoro) was offered by me. Similarly, the examples in (10) are acceptable if the apo-PPs are interpreted as by-phrases.

Moreover, the resistance in the realization of the genitive as an apo-PP (which is expected in an example like (11) given that the relationship between a mat and a door is not that of part-wholes) reduces when the nominal following apo is in plural:5

More importantly, the abovementioned restrictions on apo-PPs can be (almost) circumvented in cases of paradigm gaps, i.e., in cases where the morphological paradigm of a given noun lacks an inflected form for the genitive; see next section.

Unlike the SMG pattern, dialects spoken in northern Greece, e.g., Grevena Greek (GG), use apo-PPs across the board since the inflectional genitive has fallen completely out of use, see Michelioudakis et al. (2024). In GG, for instance, both arguments of a derived nominal can be introduced via an apo-PP; see more in Section 4:

Before comparing SMG to dialectal Greek, we focus on the restricted alternation within SMG and consider the conditions that regulate it. Paradigm gaps in specific declension classes are crucial in this respect. We take them into account first.

SMG has three genders (masculine, feminine, and neuter) and four cases (nominative, genitive, accusative, and vocative), which delineate a very complex nominal declension system, see Ralli (2000). As shown in Table 1 from Alexiadou & Müller (2008: 121), based on Ralli (2000), SMG has eight declension classes (DCs) and no one-to-one correspondence between gender and declension class. Although the SMG system is characterized by massive syncretism (see, for instance, the nominative-accusative syncretism in all neuter DCs -marked with ‘N’ subscript- in both numbers), the genitive is never syncretic to any other case within each DC. Moreover, although all DCs have the same form for the genitive plural, singular genitive forms show some variation:

Table 1 does not immediately show that the genitive formation involves not only the addition of a specific ending but also a shift in stress, at least in several cases (see Revithiadou 1999 for more on the stress patterns in Greek). As detailed in Mertyris (2014) and Sims (2006), and as shown in Table 2 (adapted from Mertyris 2014: 33, his 2.7), the genitive formation across DCs requires a stress shift–either in both numbers or only in the singular–that moves the stress from the antepenultimate to the penultimate syllable or from the penultimate to the ultimate syllable. Additionally, in some cases within the same DC, no stress shift occurs at all. This variation is already suggestive of a complex and unpredictable system, see also Markopoulos (2018).

Besides the marked6 morphological marking and stress shift, additional factors pertaining to the interaction between derivational and inflectional morphemes or between compound constituents render several nominals unable to form genitives, mainly in the plural; see Mertyris (2019) and Alexiadou (2024). For instance, Mertyris (2014: 93) cites the compound bananófluða ‘banana peel’, for which all possible forms in the genitive plural sound odd, e.g., ??bananófluðon/*bananoflúðon/ *bananofluðón. The derived word for ‘female cook’ majίrisa has also three forms for the genitive plural, namely ??majίrison, *majirίson, ??majirisόn, none of which sounds good enough. However, in principle, these nominals should pattern with the example όra given for DC III in Table 2.

Sims (2006) introduced the first systematic classification for genitive gaps in Greek. She focuses on gaps in the plural genitive and concludes that these occur when other forms in the paradigm do not predict a specific form. Nominals bearing stress in the same syllable in genitive singular and genitive plural do not exhibit gaps (13a). However, when the stress pattern in the plural deviates from that of the singular, as in (13b), gaps in the plural may emerge; see also Thomadaki (2012) and Tsompanidou (2023).

Alexiadou (2024), building on Mertyris (2019), focuses on derived nominals ending in -i, mainly diminutives ending in -aki, which lack genitive formations in both numbers, and argues that these gaps are created due to a conflict between two stress-related rules that apply in the derivation of these nominals.7

No matter which factors create such paradigm gaps, apo-PPs are traditionally assumed to step in for them:

Intriguingly, though, a paradigm gap is not sufficient to trigger the alternation. As confirmed by our informants, the substitution of a gap with an apo-PP in the context of ownership (15a) or kinship (15b) is degraded. Apo-PPs are, again, acceptable mainly in the context of part-whole relationships (14):

It seems, then, that from a synchronic point of view, paradigm gaps are not a sufficient condition for the alternation. The fact that the alternation can occur also in the absence of paradigm gaps, as already illustrated in (5) and (7), suggests that paradigm gaps are not a necessary condition for the genitive alternation either. By contrast, the type of relationship between the two nominals seems to be crucial in this respect; see more in Section 5. Nonetheless, paradigm gaps should be seen as a condition facilitating the genitive alternation, but not exclusively or to a greater degree than other conditions. Paradigm gaps, however, are important from a diachronic perspective, as they seem to indicate a significant change in the case system of a given language.

Mertyris (2014) holds that paradigm gaps constitute only one aspect or step of the gradual loss of the genitive. This is a phenomenon of language change that, according to the author, traces back to early Medieval Greek and is still ongoing in SMG, although it has reached a final stage in several dialects. Mertyris shows that the genitive in Greek has undergone significant changes due to its marked status in the language’s case system, particularly after the loss of the dative. Personal pronouns were the first to develop clitic forms exhibiting syncretism between the genitive and the accusative, thus manifesting the first step towards the loss of the genitive. Paradigm gaps in nominals requiring synizesis and a doubtful stress shift, as those mentioned above, constitute the second. The final stages of this change are detected in the majority of Northern Greek varieties in which a great degree of genitive singular loss and a complete loss of genitive plural are attested (Mertyris 2014: 210–263). The same path was followed in dialects spoken by minorities outside Greece (e.g., Italiot and Cappadocian Greek, see Melissaropoulou 2014 and references therein); these also lost genitive case marking, though in a more pervasive way due to the influence of the dominant languages (Italian and Turkish, respectively). Zombolou (2011) reports the same for Heritage Greek speakers in Argentina.

Despite the fact that the genitive has fallen out of use in a significant number of dialects, Standard Modern Greek shows resistance. We believe that the history of the language and, most importantly, its diglossic past could be responsible for it, see also Kavoukopoulos (1990). To this day, Standard Modern Greek constitutes a post-diglossic variety shaped by using Dhemotiki as a base and adding elements from Katharevousa. This process started at the dawn of the 19^th century and peaked at the beginning of the 20^th. Katharevousa (or, in fact, the several versions of it) constitutes the so-called higher variety that revived lexical items as well as phonological, morphological, and (to a smaller degree) syntactic rules of earlier eras of Greek and associated them with higher registers. It began as isolated to specific language communities but quickly rippled through new and larger communities as it became the language of Education. Dhemotiki represented the lower variety and was formed by filtering out regional dialectal idiosyncrasies. At the beginning of the 20^th century, it became the official language in Education and, in 1976, the official language of the country. The contact and interaction between the two varieties increased significantly over time, smoothed over their differences, and shaped what we call the Standard language, which is, indeed, still in the making. Nowadays, the characteristics of Dhemotiki preserved in SMG outnumber those of Katharevousa, remnants of which, however, are still attested.9 Inflectional genitives could be treated as one of those remnants since Dhemotiki, due to its dialectal background, would favor apo-PPs. Thus, the resistance to replacing the genitive with apo-PPs could be seen as a relic of this diglossic situation, which manifests itself today mainly as a competition between registers; see more in Browning (1983), Fragoudaki (1992), Papanastasiou (2024), and Tzitzilis & Papanastasiou (2024). This is also supported by the fact that SMG speakers who also speak, or are familiar with, dialects that do not exhibit genitive-related paradigm gaps, at least to the extent that SMG does (see ft. 4), are more hesitant to use and accept apo-PPs as alternatives to the genitive.

The following section offers a closer look at how SMG compares to Northern varieties.

In order to understand how restricted the distribution of apo-PPs in SMG is, we make a comparison with a Northern Greek dialect that has no inflectional genitive in common nouns and uses apo-PPs across the board, namely Grevena Greek (GG), as recently documented by Michelioudakis et al. (2024).

First, although GG PPs express the whole range of relations that SMG inflectional genitives can (compare 16a, 16b), their SMG counterparts do not (16c). In fact, as a reviewer also points out, examples like (16c) can hardly receive a possessive interpretation; they are most naturally interpreted as having a source-like reading, e.g., the book sent/donated by the mayor for (16c).

Unlike inflectional genitives (17b) and apo-PPs in SMG (17c), GG PPs can be iterated when the head (derived) nominal allows it (17a). In Section 5, we discuss more about the SMG pattern:

When iterated, GG apo- PPs are ambiguous with respect to theta assignment; the sentence in (17a) can be interpreted either as ‘the description of the mayor by the kids’ or as ‘the description of the kids by the mayor’.

An interesting similarity between SMG and GG is that the prepositional variant is resisted when the possessor/argument is a personal pronoun or a proper name. In this case, both varieties opt for realizing the personal pronoun as a clitic or via the possessive adjective ðikos mu (see footnote 4 for SMG, (18) for GG) and the proper name marked with an inflectional genitive in SMG (2, 3a) or an inflectional case marker syncretic between genitive and accusative in GG (19).

Michelioudakis et al. (2024) explicitly say that “the only inflectional genitives that are grammatical are those of pronominal clitics and proper names, including sometimes nouns with a proper name-like function, e.g., kinship terms, in some varieties (ts manas m ‘the.GEN mother.GEN my’, t patera m ‘the.GEN father.GEN my’)”. This is indeed reminiscent of other Indo-European languages, such as German, where relics of the genitive are attested with proper names.

Furthermore, Michelioudakis et al. observe that if an inflectional genitive is used, it must be post-nominal, and in the presence of a second argument, it must remain closest to the head nominal. In the latter case, ordering effects arise since the inflectional genitive is always assigned the highest (available) thematic role in Cinque’s (1980) thematic hierarchy.

The post-nominal position is also the unmarked order for the possessor and the nominal argument in SMG, regardless of its marking.

Therefore, there is empirical evidence suggesting that apo-PPs in SMG are not the same as GG apo-PPs. A closer look at their syntactic behavior indicates that SMG and GG apo-PPs should not receive a unified analysis either.

The comparison between (22) and (23) shows that GG PPs show the exact opposite behavior from their SMG counterparts. Specifically, the GG apo-PP (22a) can be fronted, unlike the respective SMG PPs (23a). GG apo-PPs can also be extracted from the DPs containing them (22b), whereas the same type of PPs in SMG cannot (23b).

In this respect, GG apo-PPs pattern with SMG inflectional genitives which can be fronted and allow sub-extraction:

In sum, GG apo-PPs show a distinct behavior from SMG apo-PPs and SMG inflectional genitives in crucial aspects. Table 3 summarizes how the three expressions compare to each other. The ✔ symbol suggests that the noun phrase has the property mentioned on the left, whereas the ✖ symbol suggests that it does not have the respective property.

Michelioudakis et al. (2024) conclude that the behavior of GG PPs is similar to that of de/di- genitives in Romance. Therefore, they adopt a parallel analysis for the GG PPs and hold that GG PPs are introduced in the structure as reduced relative clauses. Clearly, given their distinct behavior in several respects, this analysis cannot be maintained for SMG genitives or SMG apo-PPs. Although this section has demonstrated differences between inflectional genitives and apo-PPs in SMG, the following section capitalizes on that and sets the foundation for a syntactic analysis.

The previous section has shown that SMG apo-PPs differ from inflectional genitives regarding the following facts: apo-PPs (a) convey only a subset of the readings that inflectional genitives convey, (b) are highly dispreferred with proper names and pronouns, and (c) cannot be extracted or fronted. However, they pattern alike because they cannot be stacked or iterated and are preferred in the post-nominal position. In this section, we add evidence showing that apo-PPs cannot be treated as a mere alternative realization of inflectional genitives.

In Section 4, we demonstrated that stacking inflectional genitives and apo-PPs with derived nominals is prohibited in SMG. However, a context where stacking is allowed can be found with non-derived nominals. In the following example of a recursive possessive phrase, we aim to convey the meaning of ‘the support of the table’s leg’. The grammatical phrases in this case can include two genitives (25a), but not two apo-PPs (25b). Moreover, although both inflectional genitives can be replaced by an apo-PP individually (26), only the innermost one alternates with an apo-PP if both are present (27). This would come as a surprise if apo-PPs were replacing inflectional genitives.

The examples above illustrate what was already mentioned in Section 2: the alternation is natural in part-whole relationships. The fact that we observed the least variation among our informants and the highest acceptability in this context further supports this claim. Moreover, in view of examples like (16c), we can also conclude that apo-PPs are preferred for “possessors” that can be conceived of as sources. By contrast, in Section 2, we also showed that ownership and kinship resist the genitive alternation. Indeed, our informants agree on the unnaturalness of ownership and kinship examples containing apo-PPs.

Therefore, given the significant differences between inflectional genitives and apo-PPs, we conclude that the latter do not constitute a mere alternative realization of the former. As described so far, the distribution suggests that, in SMG, when the possession relation cannot be expressed via an inflectional genitive, it is coerced through a part-whole or source reading. In other words, we would like to argue that the possessive reading of apo-PPs should be best captured as a re-interpretation of partitivity.11 This also means that the so-called genitive alternation constitutes a case of substitution rather than suppletion.

Further support to this conclusion is provided by the facts already presented in Sections 2 and 3, according to which (a) proper names and pronouns, as well as animate possessors, are not preferred in apo-PPs, and (b) plural marking and modification of the possessor contained in the apo-PP increases, for several speakers, the acceptability of the phrase. As an anonymous reviewer points out, these facts indicate that apo-PPs are mainly accepted when the nominals they contain can denote (at least more easily) sets.

On the one hand, humans, particularly when presented as individuals via a proper name or a pronoun, can hardly be conceived of as introducing a set. In contrast, when presented via common nouns, particularly modified ones, they can do so more easily since they can imply a contrast with pragmatic alternatives. For instance, consider the example from footnote 8 repeated below:

Note that even if the second sentence in (28) were missing, the modified phrase isixo peðaki ‘quiet kid’ would pragmatically imply the existence of at least one other non-quiet kid. In this sense, we could say that for some speakers, including one of our reviewers, modified nominals can more easily be understood as sets that are contrasted to sets of alternatives and, hence, take part in an apo-PP. Plural marking facilitates apo-PPs because it creates sets of individuals semantically, i.e., in a more straightforward and prominent way. Therefore, in light of these facts, we believe that, unlike inflectional genitives, apo-PPs do not encode possession, but rather partitivity, which, under specific circumstances, can be coerced into possession.

Before arguing that the difference between inflectional genitives and apo-PPs is not only semantic but also syntactic, we need to consider one final possibility. Could the lexical semantics (if any) provided by apo be the reason why PPs cannot alternate with inflectional genitives or, at least, cannot express the whole range of interpretations available to the latter? The answer to this question is that if lexical semantics are indeed provided by apo, they do not suffice to explain the differences between inflectional genitives and apo-PPs.

First, if one looks at the possible readings an apo-PP can have in Dictionaries and Grammars of SMG, they will observe a very extensive range. This range is, in fact, so broad that it questions the characterization of apo as a lexical preposition. Theophanopoulou-Kontou (2000), Lechner & Anagnostopoulou (2006), Botwinik-Rotem & Terzi (2008), and Ramadanidis (2022) consider it a functional or ‘light’ preposition. In contrast, Alexiadou & Stavrou (2020) argue that apo is a meaningful, i.e., a lexical preposition, at least in the context of partitive constructions.

Second, proper names and pronouns are not, in general, unacceptable as complements of apo (29). Then, the restriction on proper names and pronouns we observe in the cases under discussion suggests that something else must be blocking their appearance.

Finally, if it were only for lexical semantics, we couldn’t explain the situation presented in examples (26–27). In principle, since substituting each inflectional genitive by an apo-PP individually is licit (26), replacing both should be possible, contrary to fact (25b, 27b).

Therefore, if lexical semantics of apo are present, they do not suffice to explain why the alternation between inflectional genitives and apo-PPs is restricted in SMG. Thus, apo-PPs cannot be analyzed as an alternative morphological realization of inflectional genitives. Therefore, we must assume that a distinct syntactic mechanism is employed in each case. We consider derived nominals and non-derived ones in turn.

In this section, we argue that, in general, apo-PPs accompanying derived nominals are of two sorts: they are either counterparts to English by-phrases (when they do not alternate with genitives) or similar to SMG apo-PPs with non-derived nominals (when they alternate with genitives). We will briefly consider the counterparts to by-phrases first and then focus on the alternating apo-PPs.

As a background, keep in mind that derived nominals in SMG are, in principle, ambiguous between a complex event reading, a simple event reading, and a result reading, in the sense of Grimshaw (1990). In their complex event reading, derived nominals support an argument structure comparable to their corresponding verb, whereas in their other readings, they do not; in these uses, they behave similarly to non-derived nominals in terms of argument structure, see Alexiadou (2001; 2009).

In Section 2, we demonstrated that external arguments of derived nominals always surface as apo-PPs, i.e., they do not alternate with inflectional genitives. Some relevant examples are repeated below:

Grimshaw (1990), Alexiadou et al. (2009; 2015), Bruening (2012), and Alexiadou (2017), among others, show that the licensing of PPs introducing external arguments depends on the transitivity of the head (derived) nominal. As these authors put it, since, for independent reasons, there is no position for the external argument within the nP, the latter should be introduced as an adjunct, i.e., via a by-phrase in English or an apo-PP in Greek. Typically, these PPs are considered adjuncts of vP/VoiceP in a structure similar to (31). Even so, their exact derivation will not concern us here. What is important is that these constitute a distinct type of PPs; as they are not part of an alternation and they introduce external arguments, these apo-PPs are the equivalents of by-phrases and they must be analyzed independently.12

Bringing our focus back to the genitive alternation, we have already demonstrated that there is a limited set of derived nominals allowing this. By looking for examples for the nominals deriving from the verb classes listed in Levin (1993), we concluded that nominals denoting senses (32a) or object alternation (32b) (particularly in the culinary context) license apo-PPs most frequently. See also (7) in Section 2.

Nominals derived from verbs of motion also allow apo-PPs, although not in such high frequency:

A closer look at the behavior of the derived nominals that license alternating apo-PPs suggests that, in this context, they do not pattern with complex event nominals but with result nominals. In other words, derived nominals licensing apo-PPs are not ambiguous between a complex event reading/structure and a result reading/structure as they only have the latter. This means that they do not have an argument structure supported by a verbal component. Hence, the apo-PP they license should be analyzed similarly to apo-PPs found with result nominals, i.e., similarly to possessive apo-PPs, see Section 7.

Moreover, this view correctly predicts the distribution already presented in (8d) and (17c) according to which we should not find derived nominals licensing two types of apo-PPs; since these nominals do not have a complex event structure when their internal argument surfaces as an apo-PP, they are not expected to license apo-PPs introducing external arguments. In this case, they behave like any result or non-derived nominal, which licenses only one possessor-like argument, see also Section 8.

Grimshaw’s (1990) criteria for distinguishing complex event nominals from result ones provide the necessary evidence for this hypothesis. On the one hand, the constructions under discussion do not tolerate aspectual (for-x-time) modifiers (34), a property characteristic of result nominals.

Their interpretation as complex event nominals is available only in the presence of inflectional genitives. This is supported by the fact that aspectual modifiers are licensed in this case:

On the other hand, when accompanied by an apo-PP, derived nominals can be pluralized, another indication that they constitute result nominals:

Therefore, it is the case that apo-PPs found with derived nominals should be analyzed parallel to possessors of non-derived nominals.13 We turn to an analysis for non-derived nominals next.

Inflectional genitives instantiating possessors do not have a uniform structure since the (in)alienable distinction creates variation. Roughly speaking, the term alienable refers to possessees that can be separated from their possessor. In contrast, the term inalienable refers to possessees that are inseparable from their possessor (Bally 1996), see also Nichols (1988), Chappell & McGregor (1989), Aikhenvald (2012; 2019), and Haspelmath (2017). According to Alexiadou (2003), inalienable possessors are introduced as complements of the possessee nominal (37a), while alienable ones are external arguments introduced by a dedicated Poss(essor)P (37b); see also Guéron (1985), Vergnaud & Zubizarreta (1992), Español-Echevarría (1997), den Dikken (2015), among others. For Michelioudakis et al. (2024), both types of possessors are introduced in [Spec, nP].

Besides evidence provided by Alexiadou (2003) and the authors listed in the paragraph above, the structural difference between (in)alienable possessors is depicted in the available quantifier-scope readings. Although inalienable possessors are ambiguous between a surface and an inverse scope reading (38a), alienable ones are not (38b); the latter allow only for the inverse scope reading. Specifically, (38a) means either that a different leg of each table broke (i.e., the front-right of one table and the back-left of another – inverse scope reading) or that the same leg of each table broke (i.e., the front-right of both tables – surface scope reading). The sentence in (38b) means that there was a different toy that each kid owned, e.g., one kid owned a toy car and a second kid owned an action figure, and these toys broke (inverse scope reading), not that there was the same toy that each kid owned, e.g., each kid owned an action figure of the same kind, and all action figures broke (*surface scope reading). This provides additional support to the analysis that inalienable possessors are introduced lower than the possessee nominals, i.e., as complements, whereas alienable ones are introduced higher, i.e., in a specifier position.

In this respect, apo-PPs behave like inalienable genitive possessors, as they license both scope readings independent of the type of relationship they express. The sentence in (39a), which contains an inalienable possessor, is ambiguous between the two readings available also for (38a). The sentence in (39b) is also ambiguous, in contrast with (38b), as it means that a different toy belonging to each kid broke (inverse scope reading) or that a toy owned by every kid broke (surface scope reading).

This points towards the hypothesis that apo-PPs are complements in all their uses under discussion. The findings of the previous sections also suggest that SMG apo-PPs with non-derived nominals should be analyzed in terms of complement selection. The fact that apo-PPs are not allowed to stack provides further support since, according to the traditional view, a complement of a head is unique. This holds because complements are selected constituents, and each head can select only one constituent. This assumption also captures the fact that the post-nominal is the most natural position for apo-PPs.

Meanwhile, the comparison between SMG and GG leads to the conclusion that a parallel analysis cannot be maintained because the freer distribution and interpretation of GG apo-PPs contrast significantly with the restricted SMG apo-PPs. (40) presents a simplified version of the structure Michelioudakis et al. (2024) propose. In this representation, the PP ap tun ðimarxu ‘of the mayor’ in GG is introduced as a reduced relative clause (RR) that adjoins within the DP, above a functional projection responsible for hosting adjectives (FP). This hypothesis cannot be maintained for their SMG counterparts.

Selection is, in principle, compatible with several analyses, including mere selection of the PP by the head nominal, introduction of the PP in a small clause, or via a relator. In other words, the complexity of the inner DP structure may vary.

An analysis for possessor apo-PPs in terms of mere selection would look like (41): the apo-PP would be directly selected by the possessee nominal.

An advantage of this is that it creates a parallel with PPs interpreted as sources. To our knowledge, there is no in-depth investigation of the syntax behind sources within the DP. As already shown in (5b), sources in SMG are introduced as inflectional genitives or as apo-PPs. In fact, their realization as inflectional genitives is not productive in SMG where the prepositional alternative is preferred. Introducing sources as inflectional genitives was more common in earlier stages of Greek, see Mertyris (2014). Traditionally, sources are analyzed as complements of the head nominal, i.e., as having the syntax in (41), but see more below.

Under this hypothesis, apo-PPs are not expected to be extracted or fronted, see Safir (1987); since they are DP-internal, they should be able to do so only along with the head nominal, see also Angelopoulos & Michelioudakis (2023). This is indeed born out as already presented in Section 4.

The fact that NPs and PPs can be relativized together (42) provides further support for this analysis. Keenan (1987) shows that head nominals should be relativized together with their modifiers, not without them:

According to an alternative hypothesis, the constituent selected by D is larger as it contains a predicative structure, i.e., a small clause (SC) headed by Pred (43). In this SC, the possessor apo-PP is in the complement position, whereas the head ‘possessee’ nominal constitutes the subject of predication. This analysis still accounts for the facts that apo-PPs occur naturally in the post-nominal position and cannot stack (because they remain in a complement position).14

This hypothesis is motivated by the fact that the apo-PP can be found in a post-copular position within a be-sentence, i.e., in a position typical for predicates (44a). Interestingly, this is also true for source-like apo-PPs (44b). This means that sources within the DP could possibly have a structure parallel to (43). Investigating the syntax of sources, however, is left for future research.

The structure in (43) presents the PP as a DP-internal constituent. Therefore, it still explains its inability to be extracted and fronted on its own, as well as its ability to relativize with the head ‘possessee’ nominal.15

A common characteristic between the structures in (41) and (43) is that they treat apo as a lexical preposition. Thus, they can be maintained if one assumes that the lexical semantics of apo exist and play a role in the distribution. Recall that Alexiadou & Stavrou (2020) argue that apo is a lexical, contentful preposition, at least in partitives. However, as already mentioned in Section 5, a considerable amount of research does not agree on this. The fact that apo functions as the complement of another preposition/adverb (45a) and does not license clitics (45b) constitutes sufficient evidence to characterize it as a non-lexical preposition.

In view of this fact, we should seek a third alternative that could treat apo as a functional preposition. Kampanarou (2021; 2023) offers such an alternative. In her discussion on sentence-level possession in SMG (46), she argues that the possession relationship is indeed construed within a small clause.

Independent evidence leads to the conclusion that this small clause is not headed by a Pred head but rather by a light preposition, a small p, that, roughly speaking, functions as a relator between the possessor and the possessee in the sense of den Dikken (2006). More specifically, this small p is a functional layer projected above lexical P that makes it transitive by licensing a position for an external argument. According to this view, which is inspired by Svenonius (2010), p constitutes the counterpart to Kratzer’s (1996) Voice. The hierarchy of the constituents in this case remains the same as the apo-PP is in the complement position and the head nominal is in the subject position. Adopting her view in the context of our discussion leads to the hypothesis that possessor apo-PPs are introduced via small clauses headed by p as in (47):

The benefit of this analysis is that it predicts that non-lexical prepositions can define possessive relationships. English of could be such an example. Assuming the purely functional layer of p, we could posit that functional prepositions like English of are merged in this layer. As there would be no lexical semantics provided by the preposition, there would be no restrictions on the relationships established between the possessor and the possessee.

Moreover, this structure can account for prepositions that are being grammaticalized via reanalysis in the sense of Roberts & Roussou (2009), see also Kayne (2000: 314). Kampanarou (2023: 257–260) discusses several cross-linguistic examples of a P-to-p reanalysis. The Greek preposition me ‘with’ found in sentence-level possessives (46b) arguably receives the same analysis. In brief, the structure in (47) can capture the grammaticalization of a preposition as follows: a lexical preposition is inserted in P. Once it becomes “transitive”, in the sense that it takes an external argument, it projects p and a specifier position to host it. The preposition moves from P to p, like a transitive verb that moves from v to Voice. When the prepositions lose their lexical status and become fully functional elements, they no longer move from P to p, but they are directly inserted into p. In this latter stage, prepositions are fully grammaticalized.16

Note that the proposal according to which prepositions can merge in P and move to p also captures the notion of semi-lexical (or semi-functional) categories as proposed by van Riemsdijk (1998). This is a welcome result as Alexiadou & Stavrou (2020) point out. According to van Riemsdijk, each category has a number of lexical and functional characteristics. Each representative of a category gathers a different set of these characteristics. Thus, it ends up closer to one of the two ends. In this sense, each category ranges from lexical to functional while defining a semi-lexical or semi-functional area in the middle. In our case, we could say that apo is a representative of the prepositional category, exhibiting properties that bring it closer to the functional end.

Finally, it is important that this analysis does not disregard the parallel with partitives but allows us to reconcile it in a slightly different way. If the partitive small-clause structure proposed by Alexiadou & Stavrou (2020) (see footnote 12) is combined with Kampanarou’s p-based analysis, we can assume that partitive apo-PPs are also introduced via a p-headed small clause; given that Kampanarou (2023) argues that this p-head exists in sentential possessives as well, this head can definitely serve as the abstract ‘HAVE’ predicative head required for partitives according to Alexiadou & Stavrou.

In any case, maintaining an affinity between partitivity and possession is necessary to explain the distribution presented in the previous sections. Our observations that apo-PPs are preferred in part-whole relations, unlike kinships and ownerships, and lack readings available to inflectional genitives can be explained if we assume that an apo-PP does not constitute a true expression of possession but rather a ‘roundabout’ for a possessive interpretation which is coerced through a partitive one. This view also predicts that proper names and pronouns, or even animate entities, should resist apo-PPs because they denote entities that can hardly be conceived of as sets unless marked with plural or when modified, as is the case (see Section 5).

Although the latter analysis seems to account better for the facts, it is essential that all the aforementioned analyses are viable and suitable to our main point of focus here, as they distinguish SMG inflectional genitives from the apo-PPs that seem to replace them in terms of syntax.17 Each one is compatible with our claim that the syntax behind case-marked possessors is not similar to the syntax underneath possessive apo-PPs. Both of them are also distinct from the syntax of apo-PPs in varieties where inflectional genitives are entirely lost.

The findings of this study strongly suggest that the so-called genitive alternation in Standard Modern Greek is not a case of morphological substitution but rather an instance of structural variation. Our comparative analysis of inflectional genitives and apo-PPs in the contexts of possessives and derived nominals reveals that apo-PPs do not simply differ in the realization component; instead, they enter the syntactic structure through a distinct mechanism. We maintain that inflectional genitives are introduced as arguments of the nominal head (or its functional layers), whereas apo-PPs in non-complex event nominals are introduced as complements, through predication or not.

The current situation in SMG is that inflectional genitives are being drastically delimited while apo-PPs are gaining ground. The presence of paradigm gaps in the declension system of nominals in SMG has facilitated the increased use of apo-PPs. However, the fact that the alternation is not restricted in these cases but, in general, extends to contexts characterized by the presence of a concomitant part-whole reading suggests that apo-PPs emerge independently.

Our comparative examination of dialectal evidence, particularly from Northern Greek dialects, where inflectional genitives of common nouns have been entirely lost, is fundamental for this hypothesis. Section 4 highlights that SMG apo-PPs exhibit distinct semantic and syntactic properties compared to their dialectal counterparts.

The contrast between SMG and these dialects suggests that language change is still in progress for SMG. This change, though, is not driven by simple morphological attrition but by broader syntactic restructuring. The pathway is indeed familiar if one looks at the diachrony of the language or Indo-European languages in general. However, it remains to be seen whether the inflectional case system will be entirely substituted by a preposition-based one or whether the distribution of the prepositional substitutes will remain constrained by interpretative and register-based factors.

Finally, this study has implications for syntactic theory, as it considers the evolution of morphological case systems and the development of adpositional alternatives in the context of a language with a very distinctive characteristic: the so-called single genitive restriction. This relates to the fact that SMG allows only one genitive-case-marked argument (or one of its alleged alternates) within the DP. To our knowledge, this has also been reported for Icelandic as well (see Wood 2023 for a recent discussion and references). English, on the other hand, constitutes one of the languages that do not show such restriction since ‘s-genitives can cooccur with of-genitives. Romance languages and Hebrew also allow multiple genitives within the DP.

An explanation that has been entertained in the literature to derive this contrast bears on the A/A’-status of [Spec, DP]. Horrocks & Stavrou (1987), Alexiadou et al. (2007), and Alexiadou (2001; 2025), among others, extensively argue that the single genitive restriction in Greek stems from the fact that [Spec, DP] in Greek is an A’ position, i.e., it does not allow the introduction of arguments. Thus, Greek DP has only one position for introducing arguments, namely, the complement position. Abney (1987), Corver (1990), Haegeman (2004), and Georgi & Salzmann (2011), among others, show that this does not hold for English. In this language, [Spec, DP] is argued to be an A-position, i.e., a locus where arguments can be introduced. Therefore, besides the complement position (where of-genitives are most likely to be inserted), there is also [Spec, DP] where the nominal preceding the ‘s-genitive is introduced. An explicit such explanation is offered in Alexiadou (2025). Romance and Hebrew allow multiple genitives of the same type (PPs) within the DP. For this reason, these genitives are often analyzed as adjuncts, see Borer (2013).

Our study shows that the single genitive restriction applies to PPs as well. Thus, Greek DP is essentially characterized by a single argument restriction. This might relate to the fact that, since both arguments are in the same domain, not more than one of the same type is allowed, as the presence of two genitives or two PPs would violate Distinctness (Richards 2010); see Alexiadou (2025). In view of this, further research in this direction is required to investigate whether we should rethink the cross-linguistic picture. This would improve our understanding of the genitive alternation and would shed light on how case systems evolve and how adpositional ones substitute for them.