3.1 Estonian nominals can have (at least) two genitive modifiers

Bošković (2008b; 2012) reveals a variety of properties that are shared by some languages without articles, and he proposes that these differences come from an NP/DP distinction. There are many generalizations from a variety of empirical domains; a sampling from the list of generalizations is presented below, but see Bošković & Hsieh (2015: 102–103) for an up-to-date version in full.

As is clear from the examples above, most of the generalizations are structured as implications. Rather than surveying all the properties enumerated by Bošković & Hsieh (2015), this section undertakes a narrow case study of property (18c) and shows that a thorough empirical investigation in a language without articles reveals that the crosslinguistic picture is more complex and less clear-cut than previous literature takes it to be.⁹ Further, the generalization in (18c) fits into a broader array of differences between genitives in Estonian and genitives in Serbo-Croatian, discussed momentarily. Taken together, the differences cast doubt on approaches seeking to analyze genitives (and more strongly, nominal syntax) in the two languages in the same way.¹⁰

The generalization in (18c) is based on work by Willim (2000) (see also Bošković 2008b: n7), who investigated the structure of nominal phrases in Polish. The logic of the implication predicts that only languages with articles—languages with DP under Bošković’s analysis—can permit transitive nominals with two genitives. Thus, if a language allows such constructions, then it must be a language with articles, i.e., a language with DP. If it does not, then we cannot conclude anything. The basic contrast can be exemplified with German, a DP language, and Polish, an NP language.

The German example (19a) is grammatical with each argument of Eroberung ‘conquest’ in genitive case, but the Polish version in (19b) is not (regardless of word order). As German is a language with articles and Polish is not, Bošković (2008b) suggests that the availability of multiple genitive arguments is tied to the presence of articles: if a language allows two genitive arguments, then it is a DP language.

This follows straightforwardly from the NP analysis on the assumption that the relationship between case-assigning heads and case-bearing nominals is one-to-one. If we assume that N⁰ has the ability to assign genitive case, but only once, any structure involving a nominal with two “genitive” arguments will be ruled out, because one of them will have an unvalued case feature. This is schematized in (20).

Because N⁰ has already assigned case to NP₁, it is not able to assign case to NP₂, thus ruling out the structure in (20).

Before turning to Estonian, some caveats are in order regarding what examples count for the generalization. First, Bošković (2008b, et seq) restricts this diagnostic to genitive arguments—in his words, he excludes “possessives.” By this, he means to rule out examples such as the grammatical Polish example in (21).

This example contains two genitive-marked nominals apparently modifying kolekcja ‘collection’. However, note that Piotra is not an argument of kolekcja ‘collection’, but a “semantically-restricted Possessor” in the terms of Willim (2000).

Second, Bošković (2012) also disregards examples involving genitive as an inherent Case, providing the Serbo-Croatian example below from Zlatić (1997).

In (22), both arguments of the nominal lišavanje ‘depriving’ are in genitive case. However, in this instance, the genitive case assigned to the second nominal phrase is preserved even when the case-marker is a verb (Bošković 2012: n9), meaning that this genitive case must be lexically-specified. Bošković disregards this kind of example as well.

Turning now to Estonian, the empirical observation is that nominals can have two genitive arguments. Examples are accepted by my consultants and are also attested in corpora.¹¹

In each of the examples above, we have nominalizations of transitive verbs (lööma ‘hit’ and tarbima ‘consume’) with two arguments. Neither of the arguments is a possessor in the sense referenced above. The genitive arguments are separated by adjectives that show concord with the nominalization. This is important—without an adjective, nominalizations with two prenominal genitives (like (25) below) are ambiguous.

In the first interpretation, each genitive is an argument of the nominalization. In the second interpretation, the first genitive is an adnominal genitive modifying the second, meaning the two genitives are a constituent to the exclusion of the nominalization. Because the intervening adjective in (23) shows concord with the nominalization itself, it is clear that the two genitive modifiers do not form a constituent to the exclusion of the nominalization. The genitive modifiers in (23) are both arguments of the nominalization löömine.

Furthermore, neither instance of genitive case in these examples can be said to be lexically-specified in the sense that Bošković discusses. While genitive case is one of the morphological cases borne by objects of verbs, it is likely structural.¹² Furthermore, it is only available for singular objects—plural objects in the same syntactic contexts must bear morphological nominative case.¹³

Plural objects of verbs in Estonian never bear genitive case, as exemplified by (27). However, plural arguments of nominalizations must bear genitive case, as in pannide ‘pans.PL.GEN’ in (23) or emiste ‘sows.PL.GEN’ in (24). Taking all of these facts together, it is clear the genitive case borne by the arguments of nominalizations have nothing to do with the verbs in the constructions. Rather, it is a result of the nominal structure they are connected to.

Returning to the main point, the generalization under consideration holds that only DP languages may allow nominals to have two genitive arguments. Thus, by the genitive argument diagnostic, Estonian must be a DP language.¹⁴ Under a DP analysis of Estonian, we could say that the extra genitive is made possible by the presence of D⁰ (with one genitive presumably made available by N⁰, as before). A structure for example (23) is given in (24) below.¹⁵

In (28), the inner DP is assigned case by the N⁰ head itself. The higher genitive is introduced by a functional projection below D⁰, which I assume is Ritter’s (1991) Number⁰ for concreteness. This genitive receives case from D⁰. If D⁰ is a case assigner, then it is immediately clear why the availability of a second genitive would correlate with the presence or absence of D⁰.¹⁶ In a world in which some languages without articles have DP and others do not, the distinction between languages like Estonian and languages like Polish or Serbo-Croatian with respect to this diagnostic receives a straightforward explanation.¹⁷

3.2 The binding possibilities of Estonian genitives

Another difference between NP languages (exemplified by Serbo-Croatian) and DP languages (exemplified by English) concerns the binding possibilities of genitives of nouns in subject position (Despić 2011; 2013). Before discussing the facts in Estonian, I briefly recapitulate Despić’s observations and analysis. To begin, observe that the Serbo-Croatian examples in (29) and (30) are ungrammatical, in contrast to their English translations. Note that we are interested only in the readings where the elements indexed with i co-refer.

Despić analyzes the distinction in grammaticality by appealing to binding. In a nutshell, the claim is that genitives in Serbo-Croatian (but not English) are capable of binding elements outside of the possessum phrase. Thus, if the genitive modifying the subject is coreferent with an object non-reflexive pronoun (as in the (a) examples) or name (as in the (b) examples), the construction is ungrammatical (due to Condition B for non-reflexive pronouns and Condition C for names).¹⁸

Despić adopts Kayne’s (1994) analysis of English genitives, wherein they are introduced as specifiers of PossP, a projection directly below DP. In contrast, Despić proposes that genitives in Serbo-Croatian are left adjuncts to NP, with no (obligatory) functional projections above them. These structures are given in (31) and (32) below.¹⁹

(31)	English:

(32)	Serbo-Croatian:20

He presents a number of the arguments from the literature in favor of such an analysis, some of which are discussed in section 3.3 of the present article.

The second piece of the analysis is Kayne’s (1994) definition of c-command which holds that adjuncts to a phrase XP (and specifiers of XP) c-command elements that XP itself c-commands.²¹ Under this definition, the adjoined genitive AP/NP in (32) c-commands everything that its adjunction site NP c-commands. Despić does not provide an example of a full clausal structure in Serbo-Croatian, but I sketch one simplified possibility in (33).

Under Despić’s assumptions, the AP/NP genitive Jovanov ‘John’s’ (boxed in (33)) c-commands everything the possessum NP (in the dotted box) c-commands. This means that Jovanov ‘John’s’ c-commands the object NP ga ‘him’ as well. Because the object pronoun is subject to Condition B, the binding represented in (33) is impossible.

In contrast, note that the structure Despić assumes for English genitives (and, I assume, genitives in DP languages more generally) buries the genitive below the DP layer.²² As a result, while a subject DP in a language like English c-commands the object DP, the genitive of a subject DP does not. It is the proposed differences in the syntax of possessive constructions in English and Serbo-Croatian that Despić identifies as the heart of the grammaticality distinction between the Serbo-Croatian examples in (29) and (30) and their English translations. Thus, if examples parallel to (29) and (30) are grammatical in a given language, then Despić would presumably analyze possessive structures in that language as something close to the English structure in (31). If they are not grammatical, then the language’s possessive structure must be like the Serbo-Croatian structure in (32).

Estonian behaves like English in this respect—sentences like those in (34) and (35) are grammatical, though some speakers prefer the (a) examples when compared side-by-side.²³

These are constructed examples, but it is also possible to find naturally-occurring examples.²⁴ Examples where the name is the genitive (as in the (a) variants) are more common than examples where the name is the object (as in the (b) variants), but I present an example of each below.²⁵

There are, in particular, abundant examples like (36) online—it is therefore possible for genitives modifying the subject to be coreferent with the object. As I mentioned, some speakers expressed a preference for the examples where the pronoun comes second (e.g., they prefer (34a) to (34b)). I believe this is true even in English; His mother loves John is certainly marked in out-of-the-blue contexts (though clearly not as bad as *He_i loves John_i). One context that supports such a construction is given in (38).

I suspect parallelism across conjuncts facilitates placing the proper name in object position rather than in genitive position in this case. Thus, though the configuration is rarer than the reverse (where the pronoun is the object), it is possible, at least for some speakers (see note 23). It seems to me, though, that Despić’s analysis predicts that both orders should be equally bad. That is not the case for Estonian, even for speakers who find the (b) examples unacceptable.

The most straightforward interpretation of the Estonian data under Despić’s analysis is that the Serbo-Croatian structure in (32) cannot be right for Estonian. Instead, we could say that Estonian genitives have a similar structure to English genitives as in (31), which would require that Estonian nominals are not simply NPs. This structure is represented for Estonian in (39).

If Serbo-Croatian is held up as a typical example of a language without articles, then this is another way in which Estonian does not behave like a language without articles.²⁶

A possible alternative, also suggested by Despić (2013: 268), is that genitives might be complements of N⁰ in some DP-less languages. If that were true, they would not c-command out of their nominal phrase, just as we have seen in Estonian. However, this alternative is not plausible for Estonian for the following reasons. First, genitives in Estonian always appear on the left of the head noun, but in all clear examples of complements to nouns in the language, the complement appears on the right rather than the left (M. Erelt 2009). Second, even if we stipulated that genitives are leftward complements in Estonian, we would then predict that all adjectives would surface on the outside of genitives, since adjuncts and specifiers are merged after complements. As we have seen, this is not true in Estonian—adjectives can easily occur between the genitive and the head noun, and in fact, this order is more common. If we adopt Despić’s analysis of the binding facts in Serbo-Croatian, the most reasonable conclusion about Estonian is that its nominal structure must be more similar to English than Serbo-Croatian.

3.3 The syntactic category and size of Estonian genitives

Finally, I turn to a discussion of the syntactic category and size of genitives. A consequence of the NP analysis proposed by Corver (1992), Bošković (2005) (et seq), and Despić (2013) is that demonstratives and genitives are analyzed as syntactically close to adjectives. This descriptive statement is formalized in various ways. For Bošković (2005), following Corver (1992), genitives and demonstratives actually are adjective heads (i.e., A⁰s). Otherwise, they are assumed to be NP adjuncts (40) or multiple NP specifiers (41).²⁷

I believe this conclusion about the syntax of genitives and demonstratives to be non-standard, and thus it could be seen as a weakness of the NP analysis. However, Bošković (2005: 6–7) argues that it is in fact a strength of the NP analysis, as genitives are adjectival in Serbo-Croatian based on a number of observations (see also Despić 2013: 250ff). I discuss three of them here.²⁸

First, genitives in Serbo-Croatian are syntactically small. They cannot be modified by other genitives or by adjectives.

In (42a), we see that a sequence of two genitives cannot be a constituent to the exclusion of the head noun in Serbo-Croatian. Thus, moj bratov ‘my brother’ is not a constituent in that example. More generally, genitives cannot even be modified by an adjective, as seen in (42b), which is only grammatical on the absurd reading where it is the neighbor’s horse who is rich rather than the neighbor. To account for this, Despić (2013: 250) proposes that these kinds of examples are impossible because they would require adjunction to an adjunct. Similarly, Bošković (2005) assumes Serbo-Croatian does not allow adjectives to modify other adjectives.

Second, genitives (and demonstratives) in Serbo-Croatian are “morphologically adjectives,” meaning that they show concord just like adjectives, and their declensions are identical (or at least very similar) to the declension of adjectives (Zlatić 1997), as we can see in (43).

In these examples, the adjective zelen ‘green’ and the genitive Milanov ‘Milan’s’ have similar endings (indicated by boldface). These endings are different from the endings on the nouns. If genitives are A⁰s themselves, we expect them to show adjectival morphology and behavior (e.g., we expect them to show concord), and these expectations are borne out in Serbo-Croatian.

Finally, genitives in Serbo-Croatian can occur in typical adjectival positions. Namely, the same element that is used for possession DP-internally is used in predicative position in Serbo-Croatian, as shown in (44)–(45).

This partially contrasts with English, where the attributive genitive determiners (except his) cannot be used predicatively—witness the ungrammaticality of *This book is my. However, non-pronominal genitives in English can occupy predicative position, as in (45). The claim is that the difference between English and Serbo-Croatian with respect to attributive genitives boils down to their syntactic status. For example, if genitives in Serbo-Croatian are A⁰s, it is no surprise that they can be used as adjectival predicates. In contrast, if the English possessive determiners are D⁰ heads, then according to Bošković (2005), it is unsurprising that they cannot occur as adjectival predicates.³⁰

To recap, the NP analysis of Serbo-Croatian leads its proponents to conclude that genitives (and demonstratives) are syntactically close to adjectives, because their usual location—specifier position of higher functional material—is not present. If we were to adopt the NP analysis of Estonian in line with the SNH, a specifier analysis of genitives would likewise not be possible for Estonian. As I show momentarily, Estonian genitives behave completely differently from Serbo-Croatian genitives with respect to the facts just discussed. When combined with the other facts considered in Section 3, the resulting picture is one in which the NP analysis of Estonian does not illuminate the behavior of its genitives in any obvious way.

First, recall that genitives in Serbo-Croatian cannot be modified in any way. In Estonian, modification of genitives (46) and recursion inside genitives (47) are possible and commonplace.

In (46), the genitive kõigi töötajate ‘all workers’ contains the quantifier kõik, which is one of the highest elements in the Estonian nominal extended projection (see section 2). Simple adjectival modification is also possible, as in rikka mehe hobune ‘rich.GEN man.GEN horse.NOM’, which is about a rich man and cannot be about a rich horse (unlike the Serbo-Croatian example (42b), which must be about a rich horse). In (47), we see a naturally-occurring example of genitive recursion in Estonian. The bracketing provided in (47)—one possibility though certainly not the only possibility—demonstrates three levels of genitive recursion. Of course, simpler examples such as mu venna sõber ‘[my brother]’s friend’ are also possible. To put it briefly, I am not aware of any differences between the modificational possibilities of adnominal genitive nouns and other nominals in Estonian, in stark contrast to Serbo-Croatian.

Next, as I noted in section 2, while most elements in the Estonian nominal phrase show concord in case and number, genitives are the one broad exception to that general rule. Genitives in Estonian do not show concord with the nouns they modify.

In (48), the quantifier kõigil ‘all’ and the adjective rikastel ‘rich’ are marked for plural number and adessive case, just as the head noun sõpradel ‘friends’ is. The genitives Kärdi ‘Kärt’s’ and mu ‘my’ are marked only for singular number and genitive case—no trace of plurality nor adessive case. Furthermore, marking the genitives with adessive case results in ungrammaticality. Thus, their failure to show concord sets genitives apart from adjectives.

Finally, there is the issue of genitives in predicative position. Recall that the genitives in predicative position in Serbo-Croatian are the same form (i.e., the same words) as those in attributive position. When it comes to Estonian, the facts are not as clear-cut, although I contend Estonian is more dissimilar to Serbo-Croatian in this respect than similar. To start with the clearest cases, non-pronominal genitives by and large cannot be used alone as predicates. In order to express a possessive meaning in a copular clause with non-pronominal genitives, the adjective oma ‘own’ is required (or strongly preferred) in addition to the genitive, as we can see in (49).

If oma is missing from examples such as these, the examples are seriously degraded. I call oma an adjective because it agrees like an adjective—for example, if the possessum is plural, oma is plural as well.³¹

Of course, oma could be a noun in these examples as well, but there are no issues with treating it as an adjective, and it can in fact be used with overt nouns as well, making an adjective analysis simpler overall (see (52)–(53)).

In truth, I do not think the choice of category is critical for my analysis, although a fuller investigation of predicative possession and oma’s connection to it may reveal more evidence that can be brought to bear on this issue.³²

Returning to the main point, it seems that proper names may be slightly more acceptable than common nouns without oma in predicative possession, but neither are preferable without oma. When presented with English sentences and asked to translate, my consultants invariably produced versions with oma. I take these facts together to indicate that predicative possession without oma is at best degraded with non-pronominal genitives.

The situation is slightly different with pronominal genitives. First, examples of pronominal genitives in predicative position are attested in corpora.

Most of the corpus examples I have found involve the first- or second-person singular pronouns minu and sinu, and my consultants all accepted similar examples, such as the one in (54). However, even in these cases, oma was still possible for my consultants, and in some cases, preferable. I will not attempt an analysis of predicative possession in Estonian, as it would take us too far afield, and the point I wish to make is simpler.³³ Namely, genitives in Estonian—unlike genitives in Serbo-Croatian—are not adjectives that can either modify nouns (in attributive position) or stand alone as predicates. To be sure, normal adjectives in Estonian can occur in the predicative position of a copular clause.

While it is possible to find examples of genitives in predicative position, it is not possible to the same degree as in Serbo-Croatian. Thus, this is another way in which Estonian genitives fail to show adjectival behavior in the sense discussed by Boškoć.

In sum, Estonian genitives behave differently from Serbo-Croatian genitives with respect to all three of the adjectival properties discussed here, as summarized in Table 2.³⁴ If those properties are to be tied to the syntax of genitives in Serbo-Croatian, as Bošković’s argumentation suggests, then it would be difficult to argue that the syntax of genitives in Estonian is the same as the syntax of genitives in Serbo-Croatian, as would arguably be required for an NP analysis of Estonian nominals.³⁵ Instead, I propose that genitives in Estonian are ordinary nominal extended projections, that is, they are no different from the extended projection that they modify. This immediately predicts that they should have the same range of modification possibilities as normal (i.e., non-genitive) nouns, and it seems that they do (see (46)–(47)). Per the analysis presented above, genitives are introduced by a functional projection above NP, though some nouns (e.g., nominalizations) can also license a genitive in their own specifier. As for concord, it would be too terse to say that genitives in Estonian “do not show concord.” They do show concord, but it is concord internal to the genitive DP rather than concord with the head noun, exactly as expected if they are full nominal extended projections.

3.4 Interim summary: Estonian genitives

Thus far, I have considered the prospects for an NP analysis of Estonian nominals by way of an investigation of Estonian genitives. I began by discussing the Bošković (2008b; 2012) generalization about nominals with multiple genitives. The generalization holds that only languages with DP may allow nominals with multiple genitive arguments. As Estonian allows this possibility, this must mean that Estonian is a DP language. I then turned to an investigation of the syntax of genitives in Estonian, with a particular focus on the conclusions that would be expected—more strongly, required—under the NP analysis. What we saw is that genitives in Estonian are quite different from genitives in Serbo-Croatian, the language that is held up as a clear case in favor of the NP analysis. It would be puzzling if genitives in the two languages had the same syntax yet behaved so differently, especially if that behavior is to be explained syntactically.

I conclude that the DP hypothesis is a more promising analysis of Estonian nominals. At a minimum, it allows for a formal understanding of the phenomena considered in this section. In addition, adopting an analysis with functional structure for Estonian nominals allows for a rather straightforward analysis of other aspects of Estonian nominals. In the next section, I consider analyses of indefinite pronouns, wh-determiners, and demonstratives, all of which have previously been argued to occupy the DP layer in other languages (see, e.g., Abney 1987). Adopting DP for Estonian allows a rather straightforward analysis of the behavior of these elements.

4.1 Estonian indefinite pronouns

The examples in (57) and (58) exemplify Estonian’s versions of what are typically called indefinite pronouns.

The examples involve the indefinite pronouns miski ‘something’ and keegi ‘someone’ modified by an adjective. Note that the adjective must follow the indefinite pronoun—the same is true for their English translations.³⁷ Morphologically, indefinite pronouns in Estonian are built from the wh-pronouns mis ‘what’ and kes ‘who’, followed by the suffix -gi/ki, which exists in the language as a clitic serving some information structural purpose.³⁸ The indefinite pronouns inflect for case, with -gi/ki surfacing on the outside of case-marking, as seen in the following examples.

Just like in a normal nominal phrase, the adjective and the indefinite pronoun must show concord, although it is only possible to see this for morphological case as the indefinite pronouns are always singular. Thus, in (57), both are in partitive case, and in (58), both are in nominative case. This is also true for most other cases, e.g., allative case in (61)–(62).

However, the terminative, essive, abessive, and comitative show a different pattern of case-marking: only the adjective bears the case-marker, and the indefinite pronoun appears in genitive case. This is exemplified with the comitative in (63)–(64).

This is normal behavior for these four cases: even in normal nominal phrases (i.e., those headed by a lexical noun) comitative case-marking appears only once on the rightmost element within the nominal phrase, and everything else surfaces in morphological genitive case.³⁹

It is not possible to mark only the adjective or both the adjective and noun with comitative case in examples like (65) and (66). It is only the noun, which is rightmost, that bears the comitative case-marking.

In the literature on indefinite pronouns, one of the most-discussed facts is the position of the adjective. In every grammatical example seen so far, the adjective follows the indefinite pronoun. Recall as well from (57)–(58) that it is not possible for the adjective to precede the indefinite pronoun. This behavior is different from what we saw in section 2, where I showed that adjectives in Estonian typically precede the nouns they modify. This is repeated below.

The relative position of indefinite pronouns with respect to adjectives has been used as an argument for treating indefinite pronouns differently from ordinary nouns. I turn now to a discussion of the kinds of analyses that have been proposed to account for that difference.

4.1.1 Analyses of English indefinite pronouns

The most well-known analyses of indefinite pronouns are based on English. In his seminal work, Abney (1987) analyzed English indefinite pronouns like everything, something, nothing, anything, and their kin, as wearing their syntactic structure on their morphological sleeve, so to speak. Morphologically, they appear to be composed of a determiner (e.g., every, some) and a noun (e.g., thing, body). Abney thus proposes that the determiner piece is a D⁰, the noun piece is an N⁰, and indefinite pronouns are derived via N⁰-to-D⁰ movement, as depicted in (68).

(68)	Abney’s (1987) analysis of indefinite pronouns

This analysis was later updated by Kishimoto (2000) as movement of N⁰-to-Num⁰. The core claims remained the same: the determiner-like element is actually a D⁰, the noun-like thing is actually an N⁰, and the adjective-final order is derived from an underlying basic order through movement.

This analysis offers a rather elegant explanation for the syntax and morphology of English indefinite pronouns, but it ends up being rather English-centric, as the cognate forms in other languages are not always morphologically decomposable in the same way, as Leu (2005) notes. This is certainly true for Estonian keegi ‘someone/somebody’ and miski ‘something’, which show no trace of a lexical noun. It is perhaps not surprising, then, that more recent work on these constructions (especially Larson & Marušič 2004; Leu 2005) has shown quite convincingly that the N⁰-raising analysis of indefinite pronouns is implausible even in English. Instead, they propose that indefinite pronouns are base-generated above the adjective. Such an analysis is sketched in (69) for English.

(69)	Base-generation:

This structure is closest to one given by Larson & Marušič (2004) but captures what it has in common with Leu’s (2005) analysis: (i) that the indefinite pronoun is base-generated higher than adjectives and is not morphologically derived by movement, and (ii) that there is a null or empty noun heading the entire projection.

4.1.2 Estonian indefinite pronouns are not generated in N⁰

Turning back to the Estonian constructions, I have represented both analyses—the base-generation analysis and the movement analysis—in (70a) and (70b).

(70)	a.	Base-generation:
	b.	N0-to-D0 Movement:

In either case, the surface position of the indefinite pronoun is higher than N⁰, and we have a clear explanation of the adjective ordering asymmetry: in Estonian, adjectives are ordered before nouns and after functional elements higher in the nominal extended projection. However, the analyses make different predictions concerning what the indefinite pronoun can co-occur with. The base-generation analysis leaves open the possibility for indefinite pronouns to co-occur with overt nouns. However, the N⁰-to-D⁰ movement analysis does not: because keegi is an N⁰ itself under that analysis, there is no place for another N⁰ to go.

In fact, indefinite pronouns in Estonian can co-occur with an overt noun, which provides a strong argument against the N⁰-to-D⁰ movement depicted in (70b). Some examples are provided in (71) and (72).

1. (71)

1. Nemvalts (1996: 59)
1. Laua
2. table.GEN
1. all
2. under
1. on
2. be
1. mis-ki
2. something-GI
1. kleepuv
2. sticky
1. asi.
2. thing
1. ‘There is a sticky thing under the table.’

1. (72)

1. Kee-gi
2. someone-GI
1. mees
2. man
1. astu-s
2. step-PST.3SG
1. sisse.
2. in
1. ‘Some man entered.’        (EKSS, entry for keegi)

In each of these examples, we have the same miski ‘something’ and keegi ‘someone/somebody’ as before, but this time, they occur inside nominal phrases with an overt N⁰. This is an argument against the N⁰-to-D⁰ movement analysis, and thus, the base-generation approach seems the most promising of the two options. This option necessitates the presence of a higher functional position. I have been labeling this projection as D⁰, although the analysis would not change if we called it something else (e.g., Num⁰).

4.2 Estonian wh-determiners mis and milline

The next category I discuss is wh-determiners. In English, this class consists of only what and which, as exemplified in (73) and (74).⁴⁰

Abney (1987) proposes that these words occupy the D⁰ position, and this assumption is adopted in later work by at least Kayne (1994); Bianchi (2000) for English and Szabolcsi (1994) for Hungarian.⁴¹ Under this kind of analysis, the bracketed DP in (73) has the structure in (75).

Analyzing these elements as determiners is reasonable given that their presence is enough to license the use of a singular count noun, one of the classic syntactico-semantic functions of determiners in English. They differ from articles in English (and indeed, from other elements which may occupy the D⁰ position), in that they are [+WH]—that is, the DPs they head participate in wh-movement.

Estonian also has wh-determiners of this sort: the words mis ‘what’ and milline ‘which’. Some examples are given in (76) and (77). Just as in English, the wh-determiners appear before their NP complements.

There is one morphological difference between mis and milline that I wish to comment on before continuing. Milline is like most nominal elements in Estonian in that it shows concord in case and number. In (77a), it expresses plural number and nominative case, and in (77b), it expresses singular number and inessive case. In contrast, mis does not show concord in these examples. In both examples in (76), mis is invariant despite the fact that the noun expresses different number and case values in each example. I do not believe there is a deep reason for this—it is simply a fact that in languages with concord systems, some individual lexical items do not show concord.⁴²

While milline and mis are different morphologically, I propose that they are of the same syntactic category, namely D⁰, following the proposals of Abney (1987) and Szabolcsi (1994). Under this analysis, the DPs in (76a) and (77a) have at least the syntactic structure in (78a) and (78b).⁴³

(78)

a.

b.

These structures are the same as the structures just proposed for indefinite pronouns: D⁰s with (potentially) overt NP complements. The point of these examples is that adopting a DP analysis of languages without articles does not require that all D⁰’s are null, because the category D⁰ contains more than just those words that are traditionally called articles. This is certainly not a radical claim given that the term determiner originally applied to words of various descriptive categories. However, the claim is still worth emphasizing, especially in light of research since Abney (1987) arguing that many of those elements are not D⁰s after all. If this analysis is on the right track, then equating determiners with articles only would go too far. Put another way, asking whether a language has overt determiners is not the same as asking whether a language has articles. Estonian has no articles, but that does not mean it lacks overt elements of category D⁰ entirely.⁴⁴

Finally, I turn my attention to another descriptive category that was once assumed to be a D⁰: demonstratives. As I have already mentioned, it is often argued that demonstratives occupy a specifier position rather than a head position. However, the Estonian demonstrative see has been studied as a potential case of an emerging article, and thus it is worth considering whether it, too, could occupy the D⁰ position. In the next section, I ultimately argue that demonstratives behave like phrasal elements in Estonian, and thus, they are not of category D⁰. However, in so doing, I discuss some interesting evidence for the existence of an edge position (which I propose is Spec,QP, higher than DP) in Estonian nominal phrases. The evidence comes from a DP-internal movement of genitives and demonstratives.

4.3 Evidence for D⁰ from demonstratives

Research that addresses Finnic nominal syntax often locates demonstratives in the D⁰ position. For example, in an article on case-marking in Finnish nominals, Brattico (2010: 52) says, “Finnish is often said to lack the category of articles, but instead of an article the nominal projection may be headed by a demonstrative (glossed as D in this article).” This may be done simply in the interest of simplicity, but it is noteworthy in light of work which claims that there are definite and indefinite articles “developing” in Estonian (Pajusalu 1997; 2000; Hiietam & Börjars 2003; Pajusalu 2009). These claims are based on examples like the following.

In (79), a demonstrative can optionally be used when referring back to a previously established referent.⁴⁵ This is undoubtedly a property shared with the definite article in English. Tracking all of the reference properties of English demonstratives and the English definite article, Hiietam & Börjars (2003) conclude that the Estonian demonstrative see is somewhere in between. The diagnostics they present are semantic or pragmatic in nature, and I submit that having similar semantic or pragmatic properties to definite articles does not mean that a particular word is a syntactic D⁰. Indeed, given the number of functional heads—and their specifiers—argued to comprise the nominal extended projection, it is rather challenging to argue convincingly that such-and-such element is specifically a D⁰ as opposed to some other category.

One kind of argument that can be made is that a particular lexical item shows behavior similar to phrasal elements rather than other heads. In the remainder of this section, I present an argument of this kind, arguing that demonstratives are more likely phrases than heads, a conclusion that is in line with most research on demonstratives. The argument comes from a word-order permutation in Estonian nominals which appears to target both genitives and demonstratives.

4.3.1 DP-internal phrasal movement in Estonian

Recall from section 2 the basic order of elements in Estonian DPs, repeated in (17), and exemplified in (15), repeated from above.

(17)	Neutral order of elements in the Estonian noun phrase
	Q > Dem > Poss > Adj > N

1. (15)

1. kõik
2. all.PL.NOM
1. nee-d
2. these-PL-NOM
1. Kärdi
2. Kärt.GEN
1. punase-d
2. red-PL-NOM
1. auto-d
2. car-PL-NOM
1. ‘all these red cars of Kärt’s’

Word order within nominals in Estonian is impressionistically more rigid than word order within clauses. However, in certain circumstances, demonstratives and genitives can come before strong quantifiers.⁴⁶

1. (80)

1. Demonstratives before Q⁰:

1.  
1. a.

1. Tea-des
2. know-DES
1. ne-id
2. this-PL.PAR
1. kõiki
2. all.PL.PAR
1. keerukus-i,
2. complication-PL.PAR
1. …
2.  
1. ‘Knowing all these complications, …’        (PARLIAMENT)

1.  
1. b.

1. nee-d
2. this-PL.NOM
1. kõik
2. all.NOM
1. ettevõtte-d
2. company-PL.NOM
1. ‘all those companies’        (BALANCED)

1. (81)

1. Genitives before Q⁰:

1.  
1. a.

1. Kärdi
2. Kärt.GEN
1. kõik
2. all.NOM
1. poja-d
2. son-PL.NOM
1. käi-vad
2. go-3PL
1. kooli-s.
2. school-INE
1. ‘All of Kärt’s sons go to school.’

1.  
1. b.

1. selline
2. the.kind
1. akt,
2. act
1. [mille
2. whose.GEN
1. iga
2. each.GEN
1. paragrahvi]
2. paragraph.GEN
1. kohta
2. about
1. on
2. be
1. palju
2. many
1. eriarvamusi.
2. dissent.PL.PAR
1. ‘the kind of act for which there are many differing opinions about every paragraph.’        (PARLIAMENT)

Whereas normally kõik ‘all’ and iga ‘each’ are first within the DP, in (80) we find the demonstrative see coming first (in its plural form need), and in (81) we find genitives in the same position. In (81a), a normal DP genitive precedes kõik ‘all’, and in (81b), the wh-pronominal mille ‘whose’ precedes iga ‘each’.

This movement is generally optional, but there is one exception: wh-pronoun genitives (like mille ‘whose’) cannot remain in situ. They must appear on the left edge of the nominal. This is shown in (82) and (83).

1. (82)
1. a.

1. *Kõik
2.   all.NOM
1. kelle
2. who.GEN
1. poja-d
2. son-PL.NOM
1. käi-vad
2. go-3PL
1. kooli-s?
2. school-INE
1.   Intended: ‘All of whose children go to school?’

1.  
1. b.

1.   Kelle
2.   who.GEN
1. kõik
2. all.NOM
1. poja-d
2. son-PL.NOM
1. käi-vad
2. go-3PL
1. kooli-s?
2. school-INE

1. (83)
1. a.

1. *Kõik
2.   all.NOM
1. mille
2. what.GEN
1. jala-d
2. leg-PL.NOM
1. oli-d
2. be.PST-3PL
1. sinise-ks
2. blue-TRL
1. värvi-tud?
2. paint-PASS.PST.PCPL
1.   Intended: ‘All of what thing’s legs were painted blue?’

1.  
1. b.

1.   Mille
2.   what.GEN
1. kõik
2. all.NOM
1. jala-d
2. leg-PL.NOM
1. oli-d
2. be.PST-3PL
1. sinise-ks
2. blue-TRL
1. värvi-tud?
2. paint-PASS.PST.PCPL

In the (a) examples above, the wh-genitive remains in its normal position lower than the quantifier, and the result is ungrammatical. The only way to ask such questions is given in the (b) version, where the wh-genitive has moved to a position higher than the quantifier.

An anonymous reviewer observes that the wh-genitive is in the subject in both examples, and thus it could be that the wh-word is undergoing clause-level movement rather than DP-internal movement. To be sure, it is possible to have a wh-genitive in a non-subject, although that constituent will still appear clause-initially (via a successive cyclic movement process that Huhmarniemi (2012) dubs snowball wh-movement). Some representative corpus examples are below.

1. (84)

1. …
2.  
1. keele-s,
2. language-INE
1. [mille
2.  what.GEN
1. kõik-i
2. all-PL.PAR
1. detail-e          ]
2. detail-PL.PAR
1. me
2. we
1. ei
2. NEG
1. mõist-a.
2. understand-INF
1. ‘…in a language whose every detail we do not understand.’        (ETTENTEN)

1. (85)

1. …
2.  
1. korralduse-d,
2. rule-PL.NOM
1. [mille
2.  what.GEN
1. kõik-i
2. all-PL.PAR
1. nüanss-e          ]
2. nuance-PL.PAR
1. ma
2. I
1. praegu
2. now
1. su-lle
2. you-ALL
1. seleta-ma
2. explain-SUP
1. ei
2. NEG
1. hakka,
2. begin
1. …
2.  
1. ‘(I have brought on board my own) rules, whose every nuance I will not try to explain to you at the moment, …’        (ETTENTEN)

These examples are both relative clauses, which also involve wh-movement. In both examples, the wh-pronoun is in initial position (along with the rest of the object DP, which is pied-piped). Thus, in actuality, these constructions involve both DP-internal movement and clausal movement: the wh-word moves to the left edge of DP, and the entire DP constituent is then pied-piped with the wh-word to the left edge.

All of the examples seen so far involve single words: the wh-pronouns kes (genitive kelle) and mis (genitive mille). Following standard bare phrase structure assumptions, one could argue that these wh-pronominal genitives are simultaneously minimal and maximal, that is to say, both heads and phrases. To show that this is really phrasal movement, consider that syntactically complex genitives can raise as well. In the case of wh-genitives, they must raise. This is shown in (86) and (87) below.

1. (86)

1. Millise
2. which.GEN
1. võistleja
2. competitor.GEN
1. kõik
2. all.PL.NOM
1. medali-d
2. medal-PL.NOM
1. ripu-vad
2. hang-PRS.3PL
1. spordimuuseumi-s?
2. sports.museum-INE
1. ‘All of which competitor’s medals are hanging in the sports museum?’

1. (87)

1. Millise
2. which.GEN
1. helilooja
2. composer.GEN
1. kõiki
2. all.PL.PAR
1. sümfoonia-id
2. symphony-PL.PAR
1. hoit-akse
2. keep-PASS.PRS
1. ülikooli
2. university.GEN
1. raamatukogu-s?
2. library-INE
1. ‘All of which composer’s symphonies are kept in the university library?’

In (86) and (87), genitives headed by the wh-determiner milline ‘which’ must raise to the edge of the nominal phrase in the same way as wh-pronominal genitives. This is significant, because it demonstrates that this movement must target phrases. Thus, we have two kinds of elements—demonstratives and genitives—that can be displaced to the initial position in Estonian nominals, and this must be a position to which phrases may move.

The fact that phrases can move to this position suggests that this is not a higher head position, as full phrases do not typically raise to head positions. The fact that this is the position of wh-phrases suggests to me that the position to the left of kõik and iga is an A′-position (i.e., the “edge” of the nominal extended projection). I propose that the movement of demonstratives and genitives is the same: movement to the highest specifier in the nominal extended projection, which I propose is Spec,QP.⁴⁷ This is depicted in (88) and (89) below.

1. (88)

1. 

1. (89)

1. 

This analysis treats demonstrative fronting and genitive fronting as the same movement, following proposals about the same kinds of processes in Modern Greek (Horrocks & Stavrou 1987; Alexiadou, Haegeman & Stavrou 2007). Following these authors, I assume the movement is focus-based for non-wh-elements, although I cannot provide a precise breakdown of its proper context of use here. As I have shown, this movement is a phrasal movement, and thus I conclude that demonstratives must be phrasal in Estonian, as they participate in this movement.

In order to make this point a bit more forcefully, I briefly consider and reject an alternative analysis where demonstratives are D⁰s and demonstrative fronting is head movement. Under this view, demonstrative fronting would have to be analyzed as head movement from D⁰-to-Q⁰, followed by particular statements about how complex heads are linearized in Estonian. This alternative is presented in (90) below.

(90)	Alternative: demonstrative fronting as head movement

Note that treating demonstrative fronting as head movement precludes the possibility of a unified analysis of genitive fronting and demonstrative fronting, assuming that heads cannot move to phrasal positions and vice versa.⁴⁸

More importantly, this head movement seems rather ad hoc. First, head movement is typically used to build complex words from syntactically separate material. This is certainly true in Estonian; all potential cases of head movement in Estonian involve one word (or root) and one or more affixes. Yet in this case, head movement would not create a word in any obvious sense: the two elements need ‘these’ and kõik ‘all’ are written as separate words, speakers view them as separate words, and they do not seem to form a phonological word.⁴⁹ Second, head movement is typically motivated by morphosyntactic considerations (Matushansky 2006), and in the case of this movement, it seems unlikely that morphosyntax is at play. In short, it is not clear to me how a head movement analysis illuminates the phenomenon in question.⁵⁰ For these reasons, a phrasal movement analysis of demonstrative and genitive fronting is more promising.

Thus, although demonstratives may fulfill some of the semantic functions common to definite articles in other languages, it would be premature to analyze Estonian demonstratives as exponents of D⁰. In other words, while the demonstratives may be developing into articles, the change is not yet complete.

4.4 Interim summary: Evidence for D⁰ in Estonian

In section 3, I situated Estonian within the NP/DP landscape, arguing that its genitive system is best understood if Estonian nominals contain DP, and in fact, an intermediate projection as well. While this requires there to be a phonologically null determiner in Estonian (in at least some cases), I argued in this section that other (i.e., non-article) elements may occupy D⁰ in Estonian, as well. I investigated the indefinite pronoun construction in Estonian, ultimately arguing that they are D⁰s (Abney 1987; Larson & Marušič 2004). I also proposed that wh-determiners mis ‘what’ and milline ‘which’ are clear candidates for exponents of the D⁰ position.

In contrast, I argued that demonstratives are not D⁰s, despite their potential to contribute determiner-like semantics (Hiietam & Börjars 2003). The evidence came from a process of fronting inside Estonian nominals targeting genitives and demonstratives. Because this fronting is obligatory for wh-genitives, I proposed that this was movement to the edge of DP. The fronting of genitives and demonstratives can be analyzed in the same way if both elements are phrasal, but not if demonstratives are D⁰s. My conclusions about demonstratives are thus in line with many previous analyses of demonstratives (Giusti 1997; Brugè 2002; Alexiadou, Haegeman & Stavrou 2007; Kramer 2009; Deal 2010; Harizanov 2011).

Taking all of this together, my proposal for the syntactic structure of Estonian nominals is presented below.

I proposed that Estonian nominals contain three functional projections, which I label Num⁰, D⁰, and Q⁰, in line with conclusions about nominal structure in a variety of languages (see references at the beginning of this section). Estonian has (at least) two genitive positions: one in Spec,NP and one in Spec,NumP. Demonstratives are typically in Spec,DP, although they may optionally raise to Spec,QP in some circumstances.