Keywords

1 Introduction

The prototypical case of grammaticalization is a process in the course of which a lexical word loses its descriptive content and becomes a grammatical marker. This paper discusses a more complex type of grammaticalization, in the course of which the phonologically null pro possessor of a possessive construction is lost, whereby the agreement suffix cross-referencing it on the possessum is reanalyzed as a derivational suffix marking partitivity.

The phenomenon in question is known from various Uralic languages, where possessive agreement appears to have assumed a determiner-like function. It has recently been a much discussed question how the possessive and non-possessive uses of the agreement suffixes relate to each other (Nikolaeva 2003); whether Uralic definiteness-marking possessive agreement has been grammaticalized into a definite determiner (Gerland 2014), or it has preserved its original possessive function, merely the possessor–possessum relation is looser in Uralic than in the Indo-Europen languages, encompassing all kinds of associative relations (Fraurud 2001). The hypothesis has also been raised that in the Uralic languages, possessive agreement plays a role in organizing discourse, i.e., in linking participants into a topic chain (Janda 2015).

This paper helps to clarify these issues by reconstructing the grammaticalization of possessive agreement into a partitivity marker in Hungarian, the language with the longest documented history in the Uralic family. Hungarian has two possessive morphemes functioning as a partitivity marker: -ik, an obsolete allomorph of the 3rd person plural possessive suffix, and -(j)A, the productive 3rd person singular possessive suffix.Footnote 1 As will be shown, they represent different stages of the pathway of grammaticalization that leads from a possessive morpheme denoting that its nominal base is the possessum of a pronominal possessor to the same morpheme denoting partitivity.

The paper is organized as follows: Sect. 2 is a brief survey of the literature discussing the non-possessive use of possessive agreement in the Uralic languages. Section 3 introduces the suffix -ik, a derivational suffix conveying partitivity and definiteness in Modern Hungarian. Section 4 argues that -ik functioned as a 3rd person plural possessive ending in Old Hungarian. Section 5 reconstructs the grammaticalization process that has resulted in the loss of its possessive function, while preserving its definite and partitive features. Section 6 shows that -(j)A, the 3rd person singular possessive agreement marker, is going through a similar grammaticalization process as its plural counterpart. Section 7 is a summary.

2 Previous Approaches

The grammars of many Uralic languages mention the fact that possessive suffixes, whose primary function is to mark the person and number of a (typically covert) pronominal possessor on the possessum, can also have a non-possessive, determining role. Nikolaeva’s (2003) survey distinguishes three types of non-possessive meanings: i. Identifying–deictic function, with the 3rd person singular possessive suffix marking that the referent of the possessum is uniquely identifiable, i.e., visible or otherwise salient, in a given situation. E.g.:

ii. Contrastive-partitive function, with the 3rd person plural possessive suffix marking that the referent of the possessum is a subset of a previously introduced set. (2a) also illustrates a collateral function of possessive agreement: it nominalizes the adjective it combines with.

iii. Associative function, expressing that the referent of the possessive morpheme (often the speaker or the addressee, referred to by a 1st or 2nd person singular suffix) is the reference point in the situation, e.g.:

The non-possessive use of possessive agreement is very frequent in the Uralic languages. In the Uralic Udmurt, for example, 30% of subjects and 40% of objects bear possessive agreement (Fraurud 2001). Fraurud sees a close connection between the possessive and the seemingly non-possessive functions of possessive agreement, arguing that possessive agreement in Uralic may also express anchoring to non-focussed or implicit referents, to contextual elements like time and place, to actions and states, and even merely to the linguistic or situational context. Fraurud sees no evidence suggesting that the determiner-like functions of possessive agreement might be the results of a grammaticalization process. According to her, possessive agreement is likely to have always had a wider range of functions in Uralic than in English.

Gerland (2014) formulated a similar view. As she put it, both possessive suffixes expressing agreement with a possessor and those expressing definiteness establish a relation; however, in the case of the non-possessive use, the suffix relates the possessum either to the discourse situation (with pragmatically unique referents) or to cultural knowledge (with semantically unique referents). She regards the non-possessively used possessive suffix as a definite article—despite the fact that its use as a definiteness marker is optional. Her main argument is that it can appear in all contexts that are typical of definite articles.

Janda (2015) claims that both the possessive and the non-possessive uses of possessive agreement are manifestations of the same function, that of establishing a relation between two entities. The entity denoted by the possessive suffix is a uniquely identifiable reference point, usually the primary topic. Janda argues that the role of the possessive suffixes in a story is to link referents into a topic chain; the primary topic cross-referenced by the possessive suffix serves as an anchor for introducing new referents and re-introducing old ones.

3 A 3rd Person Plural Possessive Suffix Turned into a Partitivity Marker in Hungarian

Studies of the definiteness-marking function of the Uralic possessive suffix mention Hungarian as an exception, where the possessive suffix has no definiteness-marking role. In fact, the associative function of 1st and 2nd person agreement identified by Nikolaeva (2003) is attested in Hungarian, too. For instance, the expression ember-ünk man-1PL ‛our man’ is often used in the sense ‛the aforementioned man’. Here is a contemporary example from the Hungarian Historical CorpusFootnote 2:

More importantly, Hungarian also has two definiteness-marking possessive agreement suffixes, -ik and -jA. The suffix -ik appears (optionally) on universal, interrogative and existential pronouns, among them minden-ik/mindegyik ‛each’, mely-ik ‛which’, valamely-ik ‛some’, némely-ik ‛some’, bármely-ik ‛any’, akármely-ik ‛any’,Footnote 3 egy-ik ‛one’, más-ik ‛other’. Whereas the -ik-less versions of these pronouns are indefinite, the -ik variants are definite, which is indicated by the fact that, when used as objects, the -ik-less pronouns elicit the indefinite conjugation, and the -ik versions elicit the definite conjugation. (A verb in the definite conjugation is supplied with the sequence or the fusion of an object agreement suffix and a subject agreement suffix. Object agreement is only elicited by a definite object—see Bartos (1997)). CompareFootnote 4:

In these cases, -ik appears to fulfil a definiteness-marking role similar to that of the non-possessively used possessive suffixes of the sister languages. More precisely, the -ik suffix adds the features [+partitive] and [+definite] to the universal or existential quantifier it merges with; it expresses that the individual denoted by the quantified expression represents a proper subset of a familiar set. Observe two pairs of examples from the Hungarian Historical Corpus. Whereas a bare mely is a wh-pronoun mostly introducing an appositive relative clause or an exclamative (8), melyik is a partitive interrogative or relative pronoun, meaning ‛which one of those under discussion’ (9).Footnote 5,Footnote 6

Since an -ik-marked universal pronoun always denotes the members of a set present in the domain of discourse (9a), it is not suitable for generic statements (9b):

A further function of the suffix -ik is to derive ordinals from fractionals (másod-ik ‛second’, harmad-ik ‛third’, negyed-ik ‛fourth’).

The suffix -ik can also appear on comparative and superlative adjectives. The -ik-marked adjective can function as a nominal, i.e., the AP projection can also be assigned a nominal shell, the empty head of which is the equivalent of the English one. The suffix -ik supplies the expression with the features [+partitive] and [+definite]. As a definite nominal, the -ik-marked adjective takes a definite article:

The -ik suffix in (10) has the same function that is identified by Nikolaeva (2003) as the contrastive-partitive function of possessive agreement.

4 -ik in Old Hungarian

In Hungarian possessive constructions, it is the possessum that must be marked; in the presence of a pronominal possessor it bears a suffix agreeing with the possessor in person and number. A pronominal possessor is silent (unless it is contrasted); it can be reconstructed from the agreement suffix of the possessum, i.e.Footnote 7:

In present-day Hungarian, the productive 3PL possessive suffix is -Uk (i.e., -uk/ük), and the assumption that -ik was also a 3PL possessive allomorph, first raised by Simonyi (1895: 716), is not generally accepted (Korompay 1992: 353).Footnote 8 We have the following reason to assume that the -ik suffix appearing on Old Hungarian pronouns and numerals is a 3PL plural suffix:

A comprehensive search of the Old Hungarian database (http://omagyarkorpusz.nytud.hu/) shows that in Old Hungarian documents, only the -ik-less versions of existential and universal pronouns occur as determiners or modifiers of lexical nouns; all -ik-marked pronouns behave like nominalizations. They represent the possessum in possessive constructions, where the possessor is a 3rd person plural pro coreferent with a plural lexical antecedent in a preceding sentence (marked by underlining in the examples below). In the underlying syntactic structure, the -ik-marked existential or universal pronoun is the modifier of an ellipted nominal (the equivalent of the English one), and it functions as the phonological host of the suffix -ik assigned to the possessum:

Compare some examples from the Old Hungarian Corpus (http://oldhungariancorpus.nytud.hu/) illustrating the syntactic contexts in which the -ik-less and ik-marked versions of universal, interrogative and existential pronouns occur:Footnote 9,Footnote 10

In Old Hungarian, ordinal numerals are still non-distinct from fractionals. Ordinals occurring in modifier position are -ik-less; the -ik-marked variants are understood to be nominalizations representing the possessum in possessive constructions; more precisely, they are understood as modifiers of an ellipted nominal. That is, the -ik-marked and -ik-less variants of ordinals show the same distribution as the -ik-less and -ik-marked variants of pronouns:

The assumption that -ik was an allomorph of the 3rd person plural possessive suffix in Old Hungarian has been questioned because the common Old Hungarian 3rd person plural possessive allomorphs appearing on lexical nouns were -ok and -ek (see Korompay 1991, Hegedűs 2014). Actually, the very first documented occurrence of the 3rd person plural possessive agreement suffix from 1192 is -ik:

The example below suggests that menden-ik and menden-ek were free variants, both meaning ‘each of them’:

Whereas the facts surveyed above show that in Old Hungarian possessive constructions, the overwhelming majority of pronominal heads bear an -ik suffix, it is also a fact that the great majority of lexical heads bear -ok/ek. What this apparent contradiction indicates is that a fission took place among these allomorphs before or around the beginning of the documented phase of the Old Hungarian period. The -ok/ek versions (which have developed into the present -uk/ük) came to be restricted to the context of overt nominal stems, whereas the -ik version was used elsewhere.

5 The Possessive Agreement → Derivational Suffix Reanalysis

Whereas in the Old Hungarian period, -ik is undoubtedly an allomorph of 3rd person plural possessive agreement, in the Middle Hungarian period we have more and more evidence of its being reanalyzed as a suffix marking partitivity. As will be argued below, the reanalysis involved a category type shift; the original inflectional morpheme became a derivational suffix, and this category change displays properties of grammaticalization.

A symptom of the reanalysis of -ik as a partitivity suffix is the appearance of -ik-marked elements in modifier and determiner positions, where they cannot be interpreted as heads of possessive constructions any more.

The reanalysis must first have taken place in the case of numerals and comparative adjectives. The first documented occurrences of -ik marked ordinals and -ik-marked comparative adjectives in modifier position are from around 1500:

We attest the first -ik-marked pronouns (mindenik ‛each’, melyik ‛which’, némelyik ‛some’, valamelyik ‛some’ etc.) in determiner position in 17th–18th century texts (Középmagyar magánéleti korpusz [Middle Hungarian vernacular corpus] http://tmk.nytud.hu/):

A further symptom of the reanalysis of -ik as a partitive suffix is the appearance of -ik-marked pronouns bearing an additional productive possessive suffix. Again, we attest the first sporadic occurrences in 17th -18th century texts. These involve a (singular or plural) lexical possessor, in which case the possessum bears an -a/e possessive suffixFootnote 11:

In the historical databases, the first documented occurrence of mindenikük/mindegyikük with the productive 3rd person plural possessive agreement suffix following its obsolete allomorph is from 1840Footnote 12:

The reanalysis of -ik as a partitivity marking suffix must have proceeded through the following stages:

The claim that the reanalysis of -ik represents the recategorization of an inflectional morpheme as a derivational suffix is based on the following considerations.Footnote 13

  1. (i)

    If the partitivity-marking -ik were an inflectional morpheme, the relation between the stem and the stem+ik complex ought to be transparent and predictable both morphologically and semantically. However, the transparent minden-ik ‘every-ik’ has become a dialectal version used in Transylvania, and has been replaced by the morphologically non-transparent mind-egy-ik ‛all-one-ik’ in Standard Hungarian. Mely ‛what’, the -ik-less variant of mely-ik, is also becoming outdated; the common alternative wh-pronoun is milyen, and the common alternative relative pronoun is ami.

  2. (ii)

    Hungarian inflectional morphemes tend to participate in vowel harmony. Almost all of the non-harmonizing suffixes are derivational. (The relation is not bidirectional though; whereas almost all inflectional suffixes are harmonizing, derivational suffixes include both harmonizing and non-harmonizing suffixes.) The fact that -ik was a non-harmonizing allomorph may have facilitated its fall from the possessive agreement paradigm.

  3. (iii)

    The -ik deriving ordinals from fractions turns nouns into adjectives, i.e., it changes the grammatical category of the relative stem, which only derivational suffixes are capable of. E.g., öt-öd ‛fifth’, a fraction, is a noun, whereas öt-öd-ik ‛fifth’, an ordinal, is an adjective. Ordinals can be subject to further derivation, e.g.: öt-öd-ik-es ‛fifth-grader’.

  4. (iv)

    The suffix -ik follows the comparative suffix of adjectives, so its derivational suffix status can only be maintained if the comparative suffix is also derivational. Evidence of its derivational status is provided by the fact that comparative adjectives are input to further derivation, e.g.: jo-bb-ít good-COMP-V ‛improve’, kis-ebb-edik small-COMP-REFL ‛lessen’, ritká-bb-an rare-COMP-ADV ‛more rarely’.

The question also arises whether the inflectional suffix → derivational suffix reanalysis can be regarded as a grammaticalization process, the prototypical cases of which involve the reanalysis of lexical items as function words. The change from possessive agreement to partitivity marking displays defining features of grammaticalization: morphological decategorization, simplification (paradigm loss), and semantic bleaching (the loss of person and number features). The loss of pro is reminiscent of the loss of movement traces attested in prototypical cases of grammaticalization.

6 The 3rd Person Singular Possessive Suffix Turned into a Partitivity Marker

The 3rd person singular -ja/-je suffix of the possessive paradigm cited under (11), too, can function as a marker of partitivity; it combines with adjectives, and turns them into partitive nominalizations. The resulting noun phrase always involves a definite article:

These possessive-marked adjectives represent the possessum of possessive constructions containing an implicit possessor. The possessor can be reconstructed from the situation or from the context. (26a–c) are likely to be assigned interpretations similar to those in (27a–c):

In the case of (27a), the possessor is likely to be physically present in the situation; in the case of (27b), it is just vaguely identifiable, whereas in the case of (27c), the implicit possessor belonging to the -je-marked adjective is conventionally fixed; it is practically part of its lexical meaning. Fehér-je ‛white-JE’, i.e., ‛egg-white’, and sárgá-ja ‛yellow-JA’, i.e., egg-yolk’ are also nominalized adjectives of this type.Footnote 14

The nominalizing role of the suffix is a consequence—or, after its reanalysis as a derivational suffix, a relic—of its original possessive agreement function. A possessive agreement suffix can only merge with a N head, hence its presence on an adjectival stem presupposes a nominal projection above the adjective.

Naturally, the question arises whether the implicit possessor of -jA-marked adjectives is present in syntax. If the -jA-marked nominalized adjectives in (28a, b) contained a pro possessor, we would expect a singular agreement suffix on the adjective in (28a), and a plural agreement suffix in (28b); however, the possessive ending appearing on the adjective in this construction is always singular:

According to the standard generative view (Bartos 2000: 684; Rebrus 2000: 773), the Hungarian possessive agreement suffixes are, in fact, morpheme complexes involving a general possessedness suffix and an agreement suffix. This is clearest in the case of a plural possessum, where the plural suffix intervenes between the general possessedness suffix and the agreement morpheme:

An agreement morpheme is only elicited by pronominal possessors; lexical possessors, whether singular or plural, only elicit the general -jA possessedness suffix on their possessum. Since the 3SG agreement suffix is zero, the -jA + Ø morpheme complex elicited by a 3SG pronominal possessor is indistinguishable from the -jA possessedness suffix elicited by a singular or plural lexical possessor. Consequently, the phonologically null possessor of a -jA-marked adjective could, in principle, be either a 3SG pro or a singular or plural ellipted lexical noun phrase. However, ellipted objects are only licensed by an antecedent in a parallel construction; the referent of an ellipted nominal cannot be identified situationally. For example, in a situation where the speaker is pointing at three boys approaching, and is wondering if his partner can recognize them, he cannot ask (30a); the plural pronoun must be spelled out as in (30b).

Hence the phonologically null possessor of -jA-marked adjectives cannot be an ellipted lexical noun phrase (except for parallel coordinate or question–answer constructions). A possibility is to identify it as a pro possessor, eliciting an invariant (default 3SG) agreement suffix on the possessum.

Default agreement also appeared elsewhere in Hungarian grammar. Early Old Hungarian abounded in non-finite subordinate constructions, which tended to have subjects of their own eliciting agreement on the non-finite verb—see (31a). These constructions have evolved either into finite subordination, or into canonical non-finite subordination involving a non-finite verb with a PRO subject and no agreement (31c). As shown by Dékány (2012), an intermediate stage in this process was the appearance of default, i.e., 3rd person singular, agreement on the non-finite verb with no regard to the person and number of its subject (31b).

Another possible analysis of the -jA appearing on adjectives is to assume that it has fully grammaticalized into a derivational suffix; it has developed into a nominalizer conveying partitivity, evoking the presence of a superset only on the notional level. These two possibilities may very well represent two subsequent stages of a grammaticalization path, which some adjectives, e.g. kövérje ‛fat-of-meat’, fehérje ‛egg-white’, sárgája ‛egg-yolk’, have passed all along, whereas others are in the process of completing. This grammaticalization path includes the following stages:

By the end of the grammaticalization path, the nominalized adjective loses its grammatically represented pro possessor with a specific number and person feature, but maintains the partitivity and—owing to the definite article, also the definiteness—of the original possessive construction. At stage (iii), the -jA morpheme behaves as a derivational suffix. The [+partitive] feature of -jA marks the presence of a notionally given superset, which is enough to block the addition of a syntactic possessor denoted by (another) possessive agreement suffixFootnote 15:

In some cases, the output of the grammaticalization process in (32) has also undergone idiomatization. Thus ‛in groups of two/three …’ is expressed by a construction involving a numeral supplied with an adjectivalizing suffix, a nominalizing -jA, and instrumental case:

An -ik-less ordinal supplied with -jA and sublative case means ‛for the 2nd, 3rd, etc. time’:

7 Conclusion

This paper has described a non-canonical type of grammaticalization path: it has argued that the Hungarian -ik partitive suffix has grammaticalized from a 3PL possessive agreement morpheme, undergoing semantic bleaching (the loss of person and number features, i.e., the loss of referential identifiability), decategorization, and morphologial simplification.

The use of possessive agreement for the encoding of definiteness/specificity is typical of most Uralic languages, however, only the history of Hungarian is documented long enough to allow the tracking of the evolution of the non-possessive function of the agreement suffix. It has been demonstrated that the suffix -ik, attached to pronouns, numerals and comparative adjectives in Modern Hungarian, expressing partitivity (e.g., minden-ik lány ʻevery one of the girls’ versus minden lány ‘every girl’) was in Early Old Hungarian an allomorph of the 3PL possessive agreement suffix; it cross-referenced a pro-dropped 3PL possessor on a possessum consisting of a determiner or modifier and an ellipted nominal (proi minden-Ø-iki ʻevery one of them’). Owing to its covertness, the pro possessor came to be ignored, and the -ik-marked expression, originally encoding a subset–set relation between the pronominal possessor and the possessum, assumed a general partitive interpretation, with -ik reinterpreted as a derivational suffix. The Hungarian 3SG possessive agreement suffix, -jA, is going through a similar grammaticalization process. -jA-marked adjectives, preceded by a definite article, can still be analyzed in most cases as possessive constructions with an ellipted nominal head; merely their pro possessor elicits default agreement. However, the possessor can also be absent altogether, in which case -jA behaves as a derivational suffix assigning the category ʻnoun’ and the feature [+partitive] to the adjective. The -jA-marked element cannot take a further possessive suffix even in the latter case, which suggests that the suffix still evokes a possessor on the notional level, which blocks the appearance of a further possessor in syntax.

It has been debated whether the non-possessive use of possessive agreement in the Uralic languages is a relic of an atypical possessive relation with a wide range of functions in Proto-Uralic, or it is the result of an evolution from marking possession and whole-part relation to marking associability, and contextual identifiability, i.e., specificity. The Hungarian data analyzed above support the latter view: the reanalysis of possessive agreement as a marker of partitivity/specificity is the result of a grammaticalization process triggered by the possibility of a silent, hence ignorable, pro possessor. It is the silent pro that opens up the way to reanalyzing possessive agreement as a derivational suffix which conveys partitivity without denoting a possessor in syntax, expressing merely that the referent is a proper subset of a situationally or contextually given set. The different Uralic languages may differ in how “strictly” they interpret this subset relation; whether they require a contextually or deictically identifiable superset, or they can also assume a subset relation between a referent and the larger situation that it is part of.