*Name*
*Student number* rMA Linguistics
Universiteit van Amsterdam 24-06-2015
Pronouns in Nanosyntax
i Acknowledgements
This paper is the result of my Research Project Linguistics, the final part of my Research Master Linguistics at the University of Amsterdam. In following this rMA, I had the opportunity to study the fields that I am interested in the most: linguistic theory and typology. These two fields are fused in this research project, that was sometimes challenging but more often fun to work on.
I would like to thank all my teachers and fellow students, in particular my co-students at the rMA, for the educational and fun experience this rMA has been. In particular, I want to express my gratitude to two of my teachers: Eva van Lier and Jan Don.
I would like to thank Eva van Lier, for introducing me to the field of typology and for being an inspirational teacher. Besides that, I am grateful for her comments on earlier versions of this paper which made me evaluate my own work critically.
I am most grateful to Jan Don, for making me realise that linguistics is the most interesting field of study, for his valuable lessons about linguistics and academic research, and for his faith in my capacities. Above all, I want to thank him for being the most inspirational, encouraging and helpful supervisor of this thesis project that a student could wish for.
ii Abbreviations 1 1st person 2 2nd person 3 3rd person ABS absolutive ACC accusative
ALIEN alienable possession
C common gender COM comitative D dual DAT dative DEM demonstrative DTLZ determinalizer ERG ergative EXCL exclusive F feminine gender GEN genitive
INALIEN inalienable possession
INCL inclusive
INST instrumental
M masculine gender
MHG middle high grade polite form (used in Nepali)
N neuter gender NMLZ nominalizer NOM nominative OBJ object OBL oblique P polite PL plural POSS possessive SBJ subject SG singular
In the text, languages are referred to in the following template: 'Language (family; location; reference)'
1
Pronouns in Nanosyntax
A typological study of the containment structure of pronominals
*Name* Abstract
This paper describes a typological study of pronominal elements, including personal pronouns (I, me etc.), possessive pronouns (my etc.) and nominalized possessives (mine etc.). It aims to confirm a previous analysis of Dutch pronominals (Don et al. 2015) cross-linguistically. In this analysis, pronouns are complex forms that mirror an underlying containment structure. This structure is such that the accusative contains the nominative, the genitive contains the accusative, the possessive contains the genitive and the nominalized possessive contains the possessive. The results from the typological data argue indeed for an extension of this containment structure: it is not a structure underlying Dutch, but underlying all pronominals in human language. Besides this, it will be shown that this containment structure can be accounted for in a nanosyntactic framework (cf. Starke 2009; Caha 2009). Case studies of the implementation of the containment structure in nanosyntax are provided to strenghten this claim.
Keywords: nanosyntax; containment hierarchy; typology; pronouns; possessives
1 Introduction
Whilst studying the Dutch pronominal system, Don et al. (2015) observed patterns of formal (phonological) similarities between the different pronominal and possessive forms. In their analysis, they do not only study the standard possessive pronouns as in (1a), but also the 'nominalized possessives' in (1b-c).
(1) a. Dit is mijn boek
this is 1SG.POSS book
'This is my book.'
b. Dit is de mijne
this is the 1SG.POSS.NMLZ
'This is mine.' (Litt.: "This is the mine.") c. Dit is mijnes
this is 1SG.POSS.NMLZ
'This is mine.'
This formal correspondence is not acknowledged in grammatical descriptions of Dutch (cf. Broekhuis & Den Dikken 2012; Haeseryn et al. 1997), where personal and possessive pronouns
2
are treated as different paradigms. However, Don et al. (2015) claim that the personal and possessive pronouns (including the nominalized possessives) form in fact a single paradigm. They argue that the different pronominal and possessive forms can be analysed as complex forms that are built from each other. In their analysis there is an underlying 'containment structure' (cf. Bobaljik 2012), such that the accusative formally contains the nominative, the genitive contains the accusative, the possessive contains the genitive and the nominalized possessive contains the possessive, as in (2).
(2) [[[[[nominative]accusative] genitive] possessive] nominalized possessive]
Don et al. (2015) explain this containment structure by implementing their analysis of Dutch pronouns in the theory of nanosyntax (Starke 2009; Caha 2009). This leads to a nanosyntactic account of the Dutch pronominal system. However, previous work on both containment structures and nanosyntax has cross-linguistic implications, claiming explicitly that their generalizations hold across all human languages. This raises the question as to whether the analysis of Dutch pronouns by Don et al. (2015) can be confirmed cross-linguistically.
This paper aims at answering this question. Based on data from a genetically balanced sample of fifty languages, it is tested whether the containment structure of pronouns applies cross-linguistically. As in the work by Don et al. (2015), this paper studies pronouns, possessives and nominalized possessives. I will argue that the analysis of Dutch can indeed be extended cross-linguistically, and that there is a universal containment structure underlying pronominals. Furthermore, I will show how this containment structure can be accounted for in a nanosyntactic framework.
It is important to note that the claim of this paper is twofold. The first aim of this paper is to describe the containment structure of pronominal elements. Independent of a theoretical framework, the formal similarities between these elements can be established cross-linguistically. In itself this is an interesting and important typological conclusion. However, the second goal of this paper is to fit this statement in a theoretical framework. I will show that the cross-linguistic patterns follow naturally from a nanosyntactic account of the containment structure. This does not discard any other theoretical framework: that nanosyntax can account for the containment hierarchy does not mean that other frameworks cannot. In this paper, nanosyntax is shown to be a plausible framework to account for the data, but I explicitly do not want to claim that it is the compelling framework.
This paper is structured as follows. The next section will describe the theoretical background, including the nanosyntactic framework and previous work on Dutch pronouns, and present the research question and hypotheses. Section 3 describes the methodology used and
3
section 4 gives a first description of the data. In section 5, the data are analysed with respect to the hypotheses and several case-studies are presented. Section 6 provides a conclusion and some points for discussion.
2 Theoretical Background
Previous work on Dutch pronouns (Don et al. 2015) has shown that the patterns of formal similarity in Dutch can be accounted for in a nanosyntactic framework. This section will start with the observations on Dutch from Don et al. (2015), and will then show how these observations can be accounted for in a nanosyntactic analysis. This analysis leads to new research questions, hypothesis and predictions for a cross-linguistic study, that will be presented at the end of the section.
2.1 The empirical observation: Dutch pronouns
The paradigm of Dutch personal and possessive pronouns, based on Broekhuis & Den Dikken (2012) and Weerman & Evers-Vermeul (2002), is given in table 1. A first observation that was made by Don et al. (2015) is that the paradigm contains many formal similarities between the different personal and possessive forms. For the first person singular for example, the object form mij 'me' and the possessive mijn 'my' are phonologically similar. All formal similarities in the paradigm are indicated by grey cells in the table.
Nominative (Subject) Accusative (Object) Possessive Nominalized possessive 1 Nominalized possessive 2 1SG ik [ɪk] mij [mɛɪ] mijn [mɛɪn] de mijne [də mɛɪnə] mijnes [mɛɪnəs] 2SG jij [jɛɪ] jou [jaʊ] jouw [jaʊw] de jouwe [də jaʊwə] jouwes [jaʊwəs] 2SG.P u [y] u [y] uw [yw] de uwe [də ywə] *uwes 3SG.M hij [hɛɪ] hem [hɛm] zijn [zɛɪn] de zijne [də zɛɪnə] zijnes [zɛɪnəs] 3SG.F zij [zɛɪ] haar [ha:r] haar [ha:r] de hare [də ha:rə] ?hares [ha:rəs] 1PL wij [wɛɪ] ons [ɔns] ons [ɔns] de onze [də ɔnzə] onzes [ɔnzəs] 2PL jullie [jʏli] jullie [jʏli] jullie [jʏli]
*de jullie-e ?jullies
[jʏlis] 3PL zij [zɛɪ] hun [hʏn] hun [hʏn] de hunne [də hʏnə] ?hunnes [hʏnəs] Table 1. Overview of personal and possessive pronouns in Dutch
4
The nominative pronoun is used in subject function. The accusative pronoun is used as direct object, indirect object and in prepositional phrases. Examples of this are shown in (3), and for the sake of completion, the examples of possessive constructions are repeated in (4).
(3) a. Ik kus hem.
1SG.NOMkiss 3SG.ACC
'I kiss him.'
b. Ik dans met hem.
1SG.NOMdance with 3SG.ACC
'I dance with him.'
c. Ik geef hem een cadeau.
1SG.NOMgive 3SG.ACC a gift 'I give him a gift.'
(4) a. Dit is mijn boek
this is 1SG.POSS book
'This is my book.'
b. Dit is de mijne
this is the 1SG.POSS.NMLZ
'This is mine.' (Litt.: "This is the mine.") c. Dit is mijnes
this is 1SG.POSS.NMLZ
'This is mine.'
The examples in (4) show that a possessive pronoun in Dutch acts as determiner-element. It cannot be combined with other determiners (such as articles and demonstratives), precedes other prenominal elements and is always interpreted as definite (Weerman & Evers-Vermeul 2002: 313).
The possessive pronoun can be nominalized with two different strategies in Dutch. The first consists of the possessive suffixed with schwa (–e) and combined with a definite article (Broekhuis & Den Dikken 2012: 840). This –e suffix is a productive nominalizer, also found on adjectives (see (5)) and on demonstratives in some Flemish dialects (Devos 2010: 137).
(5) a. een leuk boek
a nice book
b. het leuk-e van het boek
5
'the nice aspect of the book' (litt.: "the nice of the book")
These nominalized possessives can be considered NPs, since they must co-occur with a (definite) determiner, they cannot combine with other NP-internal material (Neeleman & Szendrői 2007: 681), and they show gender agreement with their nominal referent (Broekhuis & Den Dikken 2012: 840). These characteristics are illustrated below.
(6) a. * De mijne mooie auto is gestolen the mine nice car is stolen
b. Dit is jouw auto en dat is de mijne
this is your car(C) and that is the.C mine
'This your car and that one is mine'
c. Dit is jouw paard en dat is het mijne
this is your horse.N and that is the.N mine
'This is your horse and that one is mine'
A possible alternative analysis of the de mijne-construction could be that it is an elliptic construction, in which the noun is omitted but the possessive modifier not. Don et al. (2015) discard this analysis, and I do the same in this paper. If de mijne would be an elliptic construction, the noun should have the option to be overt (since ellipsis is always optional). This is however not the case. In contrast to mijn which is of the category D, de mijne cannot be followed by an adjective or noun (Broekhuis & Den Dikken 2012: 841). This is illustrated in (7).
(7) Jouw rode auto is mooi, en de mijne (*auto)/ mijn blauwe (auto) ook.
your red car is nice, and the mine (*car)/ my blue (car) also ‘Your red car is nice, and mine (*car)/my blue one (car) is also nice.’
The second strategy to nominalize a possessive is more informal and less well-attested1. In this construction, there is no article and the pronoun is suffixed with schwa followed by –s (–es). It cannot combine with a determiner or DP internal material (Neeleman & Szendrői 2007: 681), and can therefore be considered a DP.
1 For their study, Don et al. (2015) consulted three corpora: CHILDES (MacWhinney 2000), the Corpus Gesproken Nederlands (CGN, 'Corpus Spoken Dutch', Oostijk 2000) and LASSY (Van Noord et al. 2013). The first type of nominalized possessives (de mijne etc.) was attested although not frequently, the second type of nominalized possessives (mijnes etc.) was only attested several times in CHILDES. However, on internetfora (a more informal medium), both types are attested.
6
Considering the paradigm of Dutch pronouns (table 1), two observations can be made. Firstly, as already mentioned, phonological overlap is observed, as indicated by the grey cells in the paradigm. Secondly, it can be observed that this phonological similarity is not random but follows a strict pattern. The table shows that once there is suppletion, this suppletive form is also used in the forms more to the right in the paradigm. The first person singular illustrates this: the object is suppletive with respect to the subject (ik 1SG.SBJ versus mij 1SG.OBJ), and this
suppletive form is found in the rest of the paradigm (mijn, de mijne, mijnes 'my, the mine, mine'). Symbolizing the pronominal forms as letters (A, B, C, etc.) we could formulate this observation as follows: we do not find a pattern in which a stem A changes via suppletion to a stem B, and reappears as A more to the right. In other words: we do not find ABA-patterns, hence *ABA (cf. Bobaljik 2012; Caha 2009: 7, 36). This *ABA-observation is found between subject, object, possessive and the two types of nominalized possessives. So, if a possessive is suppletive with respect to the accusative (e.g. third person singular masculine hem – zijn), this suppletive form is found in the cells more to the right of the possessive (i.e. the nominalized possessives de zijne and zijnes).
2.2 Explaining the observations: a containment structure
Bobaljik (2012) discovered a *ABA-pattern for comparative and superlative adjectives, similar to the pattern described above. He observes that adjectives with a suppletive comparative have a superlative suppletive as well (and vice versa). This generalization holds for over 150 languages (Bobaljik 2012: 16). Two examples are given in (8).
(8) a. English: good - better - best
b. Basque: asko - gehiago - gehien ('many' – 'more' – 'most') (Bobaljik 2012: 28)
In his analysis, this *ABA-pattern results from the structural relation between the forms, such that the superlative contains the comparative, that in turn contains the adjective (Bobaljik 2012: 4), as in (9).
(9) [[[adjective] comparative] superlative]
As shown in table 1 and described in section 2.1, the paradigm of Dutch personal and possessive pronouns shows similar *ABA-patterns. The question Don et al. (2015) ask is whether we can apply a containment hypothesis to the Dutch pronominal paradigm following the Bobaljik’s logic. This hypothesis comes down to the structure in (10).
7
(10) [[[[[[NOM] ACC] POSS] nominalized POSS NP] nominalized POSS DP]
This hypothesis correctly predicts that if an accusative pronoun is suppletive with respect to the nominative, the possessive and nominalized possessive will be suppletive as well (e.g. the first person singular); and that if a possessive is suppletive with respect to the accusative pronoun, the nominalized possessive is suppletive as well (e.g. the third person singular masculine).
The structurally deepest part of (10) reminds us of the structure Caha (2009) proposes for case in his nanosyntactic approach. Don et al. (2015) argue that this nanosyntactic approach can also be applied to Dutch pronouns, and argue that these are complex forms containing each other from left to right in the paradigm. In the next section, nanosyntax will be introduced, followed by the nanosyntactic account of Dutch pronouns proposed by Don et al. (2015).
2.3 Nanosyntax
In a large-scale cross-linguistic study on case features, Caha (2009) finds several patterns that lead to the formulation of a universal case-hierarchy. This universal hierarchy is supported by three empirical facts. Firstly, Caha (2009: 7, 36) describes phonological relations between forms in a language such that a *ABA-pattern can be found (cf. section 2.2). Secondly, it is shown that only case features that are adjacent in the hierarchy can form syncretic cases (Caha 2009: 22). And finally, Caha (2009: 49) shows an implicational hierarchy of case, such that if a language displays a certain case feature, it will also display all case features to its left in the structure in (11) below.
These empirical findings and the formal similarities lead Caha (2009) to the formulation of a containment hypothesis of case. This is shown in (11). This containment hypothesis is to be read as follows: the deepest case is the nominative. Built from this is the accusative, that thus is more complex and contains the nominative. Built from the accusative is the genitive, that thus contains the accusative (which contains the nominative), etc.
(11) Containment Hypothesis of Case (Caha 2009: 49)
[[[[[[nominative] accusative] genitive] dative] instrumental] comitative]
Basing himself on Starke (2009; 2010), Caha (2009) discovers and implements this containment hierarchy in a nanosyntactic framework in his influential dissertation. The proposed case-hierarchy is amply supported by case syncretisms which are predicted to involve only adjacent cases in the hierarchy. In nanosyntax (Starke 2009, 2010; Caha 2009) syntactic structures are built by an operation Merge from atomic features. Each atomic feature is a terminal node in the
8
structure. In nanostyntax, there is only one mode of grammatical organisation, and no distinction between morphology and syntax is made (cf. Caha 2009: 17). It is proposed that the principles for movement applied to functional projections in standard generative grammar (Cinque 2005) are applicable to individual features, hence the word 'nano'. So, instead of labels such as 'CasePhrase' or 'TensePhrase' used in standard generative grammar (e.g. Chomsky 1981; Chomsky 1995; Cinque 2005), nanosyntax uses labels such as 'Nominative' and 'Accusative' that can move in the structure in the same way functional projections may move in standard approaches.
Each case feature in the containment hierarchy in (11) thus corresponds with a single atomic feature in a nanosyntactic tree structure. This tree structure is shown in (12).
(12) Nanosyntactic hierarchy of case (Caha 2009: 24)
As can be seen in this structure, the head of the NP (or DP) is the lowest, rightmost element in the structure and the case features are base-generated above (or to the left) of the NP. This means that if the structure would be spelled out as such, case will be spelled out before (to the left of) the noun. As it is cross-linguistically the (far) most common strategy to spell out case as a suffix (Dryer 2013), the theory thus allows for a Movement operation. Movement obeys to Cinque's (2005: 327) rules, presented in (13).
(13) Rules of Movement
a. movement is only to the left
9
If there is no movement, case will thus be spelled out before the head noun (Caha 2009: 26) and in order to spell out the case features as a suffix the 'constituent containing the noun must move to the left of all features that a particular case is composed of' (Caha 2009: 27). In order to spell out the dative as a suffix for example, the constituent containing the noun must move to the left of the dative, and consequently also to the left of all the features that are contained in the dative (i.e. genitive, accusative, nominative).
The constituent containing the head has two ways of movement, resulting in two orders of suffixes, e.g. NP-GEN-ACC and NP-ACC-GEN. The first order results when the NP moves directly to
the left of the genitive, as in the structure in (14). The second order is derived via pied-piping: the NP moves first to the left of the accusative and then this complex (NP + ACC) moves to the left
of the genitive. Via pied-piping, the universal order of the case hierarchy is thus mirrored in the spelled out form. It is language specific whether a language uses pied-piping or not.
10 (15) Movement with pied-piping
An example of a language that uses pied-piping is West Tocharian. In this language, the order of cases in NP-ACC-GEN, as can be seen in the example below.
(16) West Tocharian (Caha 2009: 69)
yakwe-m-ts
horse-ACC.PL2-GEN 'of the horses'
Cinque (2005) shows that it is language specific how high a noun can move in its extended projection that contains different nominal modifiers (adjective, numeral and demonstrative; Cinque 2005). Similarly, Caha assumes that it is language specific how high a noun can move in the case features in this extended projection (Caha 2009: 27, 30). The highest ‘landing site’ of the noun in the tree in (12) is language specific. Positions lower than the maximal landing site are expressed with suffixes, and positions higher than the maximal landing site are expressed with prepositions3. These prepositions will be combined with a noun in the 'highest' suffixal case
2 This example contains a fusional morpheme that spells out both case (accusative) and number (plural). For the sake of simplicity, I do not go into the details of this.
3 This generalization comes from Caha (2009: 43). It can be assumed that there is no difference between suffixes and postpositions here: both follow the head as a result of movement. Whether the case marker is attached to the head (i.e. is a suffix) or not (i.e. a postposition) is language specific (cf. Caha 2009: 26).
11
possible (Caha 2009: 43). This also implies that if a certain case is expressed with suffixes, all cases lower in the tree will also be expressed with suffixes (Caha 2009: 30).
Let me illustrate the previous with an example from German. German displays four morphological cases: nominative, accusative, genitive and dative. The dative is thus the maximal landing site in German. More complex cases such as the instrumental and comitative are expressed with the preposition mit 'with', that combines with a noun in dative case4. So, in order to express a comitative, the noun moves as high as possible: to the left of the dative. As a result, the comitative is spelled out before the noun, and the dative is spelled out as a suffix. The relevant German data can be found in (17)5.
(17) a. NOM: der Hund
b. ACC: den Hund
c. GEN: des Hundes
d. DAT: dem Hund
e. INST: mit dem Hund
f. COM: mit dem Hund
In nanosyntactic theory, two options exist for the spell out of the tree structures. Firstly, atomic terminal nodes can be spelled out individually. Secondly, phrase structures can be spelled out together by a single phonological form. This latter mechanism is called 'phrasal spell out', and can account for suppletive (i.e. non-related) forms. Instead of, for example, the individual spell out of the stem and the nominative and accusative nodes, the whole accusative phrase can be spelled out at once. An example of a phrasal spell out rule is provided in (18).
(18) Phrasal spell out rule of Dutch accusative phrase
4 Note that these two cases are syncretic in German. This syncretism is predicted to be possible by the case hierarchy, as the two cases are adjacent in the tree structure.
5 Note that in German, the determiner instead of the noun bears case. Caha (2009) presents his case hierarchy both with an NP and with a DP as the structurally deepest element. In German, case is assumed to be built on top of a DP, leading to an inflected D (in these examples the definite determiner).
12 2.4 The analysis of Dutch pronouns
To summarize the sections above, what I have shown is that the *ABA-observation in the Dutch pronominal system is similar to Bobaljik's (2012) *ABA-observation for adjectives, comparatives and superlatives. This observation has led Don et al. (2015) to propose a containment hierarchy of pronouns, comparable to Bobaljik's (2012) containment hierarchy of comparative morphology. This containment hierarchy is partially similar to Caha's (2009) case hierarchy, which is implemented in a nanosyntactic framework.
The nanosyntactic account of Dutch pronouns proposed by Don et al. (2015) implements the containment hierarchy based on the *ABA-observations (section 2.1-2.2) in a nanosyntactic framework inspired by Caha's (2009) work on case systems. They formulated the following containment hypothesis for Dutch pronouns.
(19) [[[[[NOM] ACC] POSS] nominalized POSS NP] nominalized POSS DP]
This containment hypothesis describes the pronouns as complex elements, built from left to right in the paradigm in table 1. It correctly predicts the found patterns of suppletion, and the *ABA-pattern.
Remember, however, that the direct object and indirect object in Dutch are always syncretic, in other words: the accusative and dative are syncretic. Caha's (2009) case hierarchy predicts that if a language has a syncretic accusative and dative, the genitive is syncretic as well (otherwise, an ABA-pattern would arise). Therefore, it is assumed that Dutch has a genitive, that is syncretic with the accusative and dative.
This might seem problematic, since the genitive does not appear in Dutch. However, Caha (2009: 273-277) finds similar accusative-dative syncretisms in Old English pronouns and argues that the language also has a genitive. This does not appear as an independent genitive pronoun. 'Independent' here means a genitive that is used for other functions than the possessive alone, for example in fixed combinations with verbs or prepositions. The Old English genitive pronoun is thus not independent, but 'hidden': it is only used as the basis of the possessive pronoun.
Old English can be contrasted to Czech, a language with an independent genitive pronoun. The genitive pronoun is used in combinations with certain verbs and serves as the basis of the possessive pronoun. In Czech, the fact that the genitive is contained in the possessive is overtly shown (Caha 2009: 276):
(20) je-jí dítĕ
POSS-GEN.3SG.F child
13
Taking this into account, Don et al. (2015) rephrase the containment hierarchy of personal pronouns such that the possessive contains the genitive:
(21) [[[[[[[NOM] ACC] GEN] POSS] nominalized POSS NP] nominalized POSS DP]
In this containment hierarchy, the possessive is built from the genitive, which is built from (and in Dutch syncretic with) the accusative. From this possessive, two nominalized possessives are built. In the following, the nanosyntactic analysis of Dutch pronouns of Don et al. (2015) is provided with examples for the first person singular. These can of course easily be extended to the plural and other persons.
In the nanosyntactic structure underlying the containment hypothesis in (21), the head of the tree consists of abstracts features such as [1SG] and [3SG.M]6. On top of this, the case features are built in the same way as in Caha's (2009) work. In the tree structures given below, I will only show the relevant case features for specific phonological forms, and I will thus leave out the comitative, instrumental and dative in most examples.
The nominative is spelled out if the head moves to the left of the nominative position. This results in the form ik 'I' by means of the mapping rule in (22a). The accusative is suppletive and takes the form mij 'me'. This form results from phrasal spell out of the accusative phrase (through the mapping rule in (22b)) after movement of the features to the accusative.
(22) a. Spell out rule of Dutch [1SG] nominative
b. Spell out rule of Dutch [1SG] accusative
6 Note that attempts have been made in nanosyntactic framework to formalize the trees underlying these abstract features. In this works, the features 1st person, 2nd person, singular, plural, etc. are all atomic terminal nodes (cf. vanden Wyngaerd (2014/in progrss)).
14
As shown above, the possessive is built from the genitive. In order to form a possessive pronoun, the features first move to the genitive. This genitive is spelled out as mij via phrasal spell-out (see (23)), as it is syncretic with the accusative. This genitive pronoun moves further to the possessive phrase, and is there combined with the regular Dutch possessive (singular) suffix –n to form mij-n 'my'. This whole phrase is then moved to D, as the possessive in Dutch behaves grammatically as a determiner. The complete derivation of the possessive pronoun is shown in (24a), and the spell out rule of the possessive is shown in (24b).
(23) Spell out rule of Dutch [1SG] genitive
15 b. Spell out rule of Dutch possessive
Next to the pronominal construction, the possessive meaning in Dutch can also be expressed with a prepositional construction. As Caha (2009) correctly predicts, the preposition is combined with the most complex case contained in (the one just below) the genitive, i.e. the accusative (see (25)).
(25) Structure of the Dutch prepositional possessive (de jas van mij 'the coat of me')
In order to become a nominalized possessive NP, the possessive phrase moves into the higher nominal head, where it is combined with the nominalizer –e (cf. example (5b)). The head of D is then spelled out by a definite determiner7, as is shown in (26a) below.
7 Dutch determiners reflect the grammatical gender of the noun. They either have the form de 'the.COMMON' or het 'the.NEUTER'. At this moment, I do not go further into this, and use de in all examples.
16
(26) a. Structure of Dutch nominalized possessive de mijne
b. Spell out rule of the nominalizer
A final option for a nominalized possessive is to move from its position in the NP even further to DP. There, it is combined with the suffix –s, and since it occupies the D-position, no determiner-element is allowed.
17
(27) a. Structure of Dutch nominalized possessive mijnes
b. Spell out rule of the determinalizer
As a summary, let me illustrate the whole nanosyntactic structure that accounts for the Dutch pronominal system. This structure represents the same complex relations between forms as the containment hypothesis in (21) described: the nominalized possessive contains the possessive, which contains the genitive, containing the accusative, containing the nominative. In the analysis of Don et al. (2015), the form mijnes is analyzed as a complex pronoun, as is shown in (28).
(28)a. mij-n-e-s
18 b. Structure of Dutch pronominal
2.5 New questions, hypotheses, predictions
Now that I have presented a nanosyntactic account of Dutch pronouns, the question of cross-linguistic application easily rises. Caha's (2009) case hierarchy and Bobaljik's (2012) containment hypothesis for comparative and superlatives both account for cross-linguistic rather than language specific patterns. The research question to be answered here is therefore formulated as in (29).
(29) Research question
Can the containment hypothesis on Dutch pronouns, which describes the pronouns as structurally complex forms that contain structurally smaller forms, be cross-linguistically confirmed?
So, the research question in fact asks whether the analysis that the possessive contains the genitive, which itself contains the accusative, etc. can be typologically confirmed. It is hypothesized that this containment structure can indeed be confirmed cross-linguistically, or in
19
other words: my hypothesis is that the patterns found in Dutch are not a language specific coincident, but rather mirror a cross-linguistic pattern.
This hypothesis can be divided in several sub-hypotheses, making predictions about different containment relations of pronouns. The first hypothesis, the 'Containment Hypothesis of Pronominal Case', is based on the containment relation Caha (2009) formulates for case on DPs, and states that in all languages pronominal case features follow the containment hierarchy of Caha (2009). It is important to mention that the hierarchy only applies to languages with nominative-accusative case systems, and not to languages with ergative-absolutive systems (Caha 2009: 18). I will leave the question of ergative languages for the moment, and come back to it in section 5.1. If this hypothesis is correct, no ABA-patterns will be found in the domain of pronominal case. I therefore predict '*ABA' for the sequence NOM-ACC-GEN, as well as for the
sequence ACC-GEN-DAT. If these ABA-patterns are found, this means that the hypothesis is falsified and Caha's (2009) case hierarchy has to be reconsidered, at least for pronominal elements. Based on his own cross-linguistic study however, this finding would be highly surprising.
The second hypothesis is based on Don et al. (2015). The 'Containment Hypothesis of Possessives' states that the possessive is built from the genitive case of the pronoun, and thus contains the genitive (and hence the accusative and nominative). This hypothesis predicts that no ABA-patterns will be found in the domain of possessive pronouns. I therefore predict '*ABA' for the sequence ACC-GEN-POSS.If these ABA-patterns are found, the Containment Hypothesis of Possessives is falsified and the idea that the possessive is built from the genitive has to be reconsidered.
The 'Containment Hypothesis of Nominalized Possessives' describes the last part of the complete hierarchy and states that the nominalized possessive (like de mijne, mijnes 'mine') are built from the possessive and contain the possessive and all elements contained in the possessive. If this hypothesis is correct, this means that no ABA-pattern is found in the domain of nominalized possessives. I therefore predict '*ABA' for the sequence GEN-POSS-NMLZ POSS, and if
this ABA-pattern is found, the hypothesis is falsified. If many languages lack a nominalized possessive, this means that I cannot be decisive on this hypothesis.
Remember from section 2.4 that Caha (2009) and Don et al. (2015) argue that in languages with a syncretic accusative-dative but a phonologically different possessive, there is also a genitive. I will call this a 'hidden' genitive: it is a genitive that is only used as the base of the possessive, and is not used independently in (fixed) combinations with certain verbs or prepositions. This means that the prediction is that a language with a syncretic accusative-dative lacks an independent genitive. Rather, the genitive is hidden and syncretic with the
accusative-20
dative pronoun, and the possessive is based from this syncretism. This is called the 'Independent Genitive Hypothesis'.
All hypotheses are listed in (30)-(33) below. As in Caha (2009) and Bobaljik (2012), the typological study in this paper is used to confirm a claim about a universal structure. Both the containment hierarchy and the possibility of its elements to move in the structure are assumed to be universal, or in other words part of Universal Grammar (UG). This means that each counter-example is a potential falsification of the hierarchy, and should be examined carefully. In contrast to much current typological work, this study does not aim at finding tendencies, but at establishing a cross-linguistic generalization. I will now turn to section 3, in which I will explain the methodology used to test these hypotheses and predictions. In section 4, results of this cross-linguistic study will be shown.
(30) Containment Hypothesis of Pronominal Case
For all nominative-accusative languages, pronominal case categories are ordered in a containment structure that has the following form:
[[[[NOM] ACC] GEN] DAT]
(31) Containment Hypothesis of Possessives
In all languages, the possessive pronoun and pronominal case categories are ordered in a containment structure that has to following form:
[[[[NOM] ACC] GEN] POSS]
(32) Containment Hypothesis of Nominalized Possessives
In all languages, the nominalized possessive and possessive are ordered in a containment structure that has to following form:
[[[[[NOM] ACC] GEN] POSS] NMLZ POSS]
(33) Independent Genitive Hypothesis
If a language has a syncretic accusative-dative pronoun, it lacks an independent genitive pronoun as well. In these languages, the genitive is only used as part of the possessive pronoun.
3 Method
3.1 Sampling procedure
A cross-linguistic comparison between fifty languages has been made to test the hypotheses formulated in the previous section. Since this project aims at studying a large variety of
21
languages in order to distinguish cross-linguistic patterns, a variety sample based on the sampling procedure of Rijkhoff & Bakker (1998) has been compiled.
The rationale behind this sampling procedure is that the more genetic variety a language family or subfamily exhibits, the more of its languages should be in the sample (Rijkhoff & Bakker 1998: 268). In other words: the internal complexity of a language family determines the proportion of this family in the sample. By this procedure, a variety sample that is free from genetic bias can be created. To determine how much variety a language (sub)family exhibits, the 'Diversity Value' (Rijkhoff & Bakker 1998: 269) is calculated for each family and subfamily (and sub-subfamily, etc.). Two parameters are used to calculate this Diversity Value (henceforth DV): the width and depth of the specific family.
The depth of a language family describes the number of levels between the top-node (i.e. the family name, for example Indo-Hittite) and the terminal nodes (i.e. the individual languages, for example English, Spanish and Persian). The top-node is not counted as a level, as all language families have such a top-node and this nodes adds no information on the diversity of the family. The terminal nodes or individual families are not counted as a level to restrict the influence of the actual number of languages. In this rationale it is not important how many languages a family contains, but how much genetic variation (Rijkhoff & Bakker 1998: 269).
The width of a language family describes the number of nodes on a single level, plus the number of pre-terminal nodes at higher levels (Rijkhoff & Bakker 1998: 269-270). To illustrate this, an example of a genetic language tree is provided in figure 1. This figure shows the genetic structure of the Eskimo-Aleut family (according to Ruhlen 1987: 329). The width of the first level (the level down the top-node) is two, since this level contains two nodes (Aleut and Eskimo). The node of Aleut has no further sub-nodes, only terminal nodes (individual languages), and this node is therefore extended, depicted in the figure by the 'x'-sign. The width of the second level is three: the two new nodes (Inuit and Yupik) plus the extension of the first level pre-terminal node Aleut. The width of the third level is four: two new nodes (Alaskan and Siberian) plus the extension of the pre-terminal nodes from higher levels (Aleut and Inuit).
22
In the calculation of the DV of a certain (sub)family, higher nodes are assigned greater significance. Since they represent diachronically older splits, higher nodes are assumed to have a greater impact on linguistic diversity than lower level nodes (Rijkhoff & Bakker 1998: 270). In the example of figure 1, the difference between Aleut and Eskimo is expected to be larger than the difference between Alaskan and Siberian, as the latter are lower (and more recent) split-ups. Once the DV of each language family and sub(-sub etc.) family is calculated, this DV determines by how many languages each family should be represented in the sample. Since the samples compiled with this procedure must account for linguist diversity across language families, it is a prescription of the procedure that each language family is represented in the language by at least one language (Rijkhoff & Bakker 1998: 268, 272). Each language family is thus represented by one language, and the DV of the family determines how many more of its languages should be in the sample to create a (genetically) balanced sample.
Rijkhoff & Bakker's (1998) procedure can be followed independent of the genetic classification system that one uses. In other words: any classification system can be used to compile a sample. I have worked with the classification system of Ruhlen (1987/1991), which distinguishes a relatively small number of language families. As mentioned above, the sampling procedure requires a minimum of one language per family and with this requirement in mind, Ruhlen's (1987) classification provides a good basis for a relatively small sample. I have chosen for a relatively small sample, containing fifty languages, for both theoretical reasons and reasons of feasibility.
In theoretical respect, a relatively small sample is the preferable choice for a pseudo-probability sample (Bakker 2010: 121; Van Lier 2009: 111), which are "relatively small variety samples with a relatively high degree of independence between the selected cases" (Van Lier 2009: 110). Pseudo-probability samples are variety samples, but composed with a procedure that prevents genetic biasing (such as the DV-procedure) and preferably also areal biasing. The number of fifty languages is a reasonable number for pseudo-probability samples (cf. the samples of Hengeveld et al. 2004 and Van Lier 2009).
The classification by Ruhlen (1987) is notable for its small number of language families: 19, and besides that a group of unclassified languages. These families are listed in table 2 below8. Note that in this classification, pidgins & creoles and language isolates are treated as a language family. Perhaps 'family' is not the right term for these groups of languages, as they are not genetically related in the way 'real' language families are. I will come back to this point later.
8 Ruhlen (1987) only classifies spoken languages. Sign languages are not included, and are also not included in my sample, how interesting they might be.
23 Language family Khoisan Niger-Kordofanian Nilo-Saharan Afro-Asiatic Caucasian Indo-Hittite Uralic-Yukaghir Altaic Chukchi-Kamchatkan Eskimo-Aleut Elamo-Dravidian Sino-Tibetan Austric Indo-Pacific Australian Na-Dene Amerind Language Isolates Pidgins & Creoles
Table 2. The families distinguished by Ruhlen (1987)
As Ruhlen (1987: 14) describes, creating a genetic classification is a reflection of history: the genetic classification describes the common origin of (sub)families. The most used 'evidence' for dividing the world's languages in language families is with a list of basic vocabulary (containing words such as numbers, body parts, kinship terms): languages with similar lexical items in terms of both phonology and semantics are grouped together as a family (ibid: 9-14). Once it has been established which languages belong to a certain family, they are grouped together in subfamilies based on their shared (lexical) innovations. Languages with (a cluster of) shared innovations are assumed to have the same ancestor and therefore belong to the same subfamily.
Ruhlen's (1987) classification is currently controversial because of its small number of language families. There has been debate about whether these languages are really related to each other, and in more recent (and currently more accepted) classifications many of Ruhlen's (1987) families are broken down into more families that are not related to each other. Also in a more recent version of Ruhlen (1991), two of the families mentioned in table 2 were broken
24
down in two families. Firstly, the Caucasian family was split up between the North Caucasian ('Caucasian' in Ruhlen 1991) and South Caucasian ('Kartvelian' in Ruhlen 1991) families. Secondly, the branch Korean-Japanese-Ainu was removed from the Altaic family and seen as a distinct language family (Ruhlen 1991: 397-380).
Once one has decided on the genetic classification system to be used, the DVs of the families and subfamilies have to be calculated. This is a challenging task, especially in terms of time. Bakker (p.c.) has developed an automatized application which will generate a basic sample. For this basic sample, the application calculates the number of languages that is needed per family and subfamily, based on the total number of languages that will be used. For my study, the application uses the updated version of Ruhlen (i.e. Ruhlen 1991) and groups together pidgins & creoles and language isolates9. This is an option in the sampling procedure of Rijkhoff & Bakker (1998: 290-292) to prevent the sample of being filled with individual language isolates. The DV of these two subgroups is set on 1.5, and as a result most samples (except the very large ones) contain one pidgin or creole and one language isolate.
The application provides a basic sample, describing the number of languages per family and subfamily. However, in some cases less languages are needed than the family has subfamilies. In a fifty-language sample for example, six Indo-Pacific languages are used, but the family contains thirteen subfamilies (Ruhlen 1987: 297-298). In these cases, the researcher has to select the subfamilies by himself. The actual languages that will be used also have to be selected by the researcher himself.
From the basic sample that Bakker's application (p.c.) calculated, I created a complete sample, following several steps and guidelines. Firstly, I only included languages that are not extinct in Ruhlen's classification (1987)10. This means that I did not include an Anatolian language (part of the Hittite family, Ruhlen 1987: 325-327), but included a second Indo-European language instead. I excluded the Germanic subfamily, as Dutch belongs to this family. Secondly, I did not include a language from the group 'unclassified'. As the reason for this label lies in the fact that these languages are not described thoroughly enough to classify them, finding data is hard for these languages. I used a language from the group 'pidgins & creoles' instead.
In cases where the application-created sample does not select a subfamily and the researcher has to do so, Rijkhoff & Bakker (1998: 276) suggest three options for this selection. They recommend to randomly choose a subfamily, but with taking into account its DV11. I
9 Since the application is based on the procedure of Rijkhoff & Bakker (1998), it can in principle use any classification. I have chosen to work with Ruhlen (1991).
10 Note that this classification is relatively old. Languages might be extinct by now, although Ruhlen does not specify them as such.
11 By doing so, each subfamily has the chance of being selected, but the subfamily with the highest DV has the biggest chance of being selected.
25
however choose a different option, and selected the sub(sub)family with the largest number of languages and with the largest diversity.
Each decision of including or excluding a language also depended on the availability of good quality descriptions. I only selected languages that were described detailed enough with respect to their pronouns, possessive constructions and case-system. Besides this practical aspect, I tried to take into account geographical location and size of the languages, in order to balance for these factors as well. However, I did not do this in a statistical manner.
The procedure explained above results in the language sample in table 3. Appendix 1 contains a second list of the sample, describing the geographical location and primary reference of each language. Because Ruhlen (1987/1991) is controversial, this appendix also mentions its genus according to the World Atlas of Language Structures (WALS, Dryer & Haspelmath 2013). Appendix 2 contains a map showing the locations of all languages included in the sample.
Language family Subfamilies Language in sample
Khoisan (1) Nama Austric (7) Austro-Tai (5) Austronesian (4) Malayo-Polynesian (1) Paiwanic (1) Atayalic (1) Tsouic (1) Daic (1) Austroasiatic (1) Miao-Yao (1) Chamorro Paiwan Atayal Tsou Lao Kharia Hmong Njua
Indo-Pacific (6) West Papuan
East Papuan Torricelli Sepik-Ramu Trans New-Guinean Tidore Lavukaleve Bukiyip Alamblak Kobon Tauya Niger-Kordofanian (4) Niger-Congo (3) Niger-Congo Proper (2) Central (1) West Atlantic (1) Mande (1) Kordofanian (1) Koromfe Kisi Bambara Katcha
Caucasian (1) North (1) Lezgian
Amerind (9) Ge-Pano-Carib (2) Ge-Pano (1) Macro-Carib (1) Northern Amerind (2) Bororo Tiriyo Koasati Oklahoma Cherokee
26 Equatorial-Tucanoan (2) Equatorial (1) Macro-Tucanoan (1) Chibzan-Paezan (1) Central Amerind (1) Andean (1) Wari' Hup(da) Teribe Kiowa Imbabura Quechua Sino-Tibetan (2) Tibeto-Karen (1) Sinitic (1) Lahu Mandarin Chinese
Nilo-Saharan (2) East Sudanic Central Sudanic Lango Ma'di Australian (3) Pama-Nyungan Tiwi Bunaban Dyirbal Tiwi Gooniyandi Afro-Asiatic (3) Chadic Berber Cushitic Hausa Tamasheq Afar Indo-Hittite (2) Indo-European (2) Anatolian (0) Nepali Albanian Na-Dene (1) Sarcee Korean-Japanese (1) Japanese Altaic (1) Evenki
Eskimo-Aleut (1) Central Alaskan Yupik
Chukchi-Kamchatkan (1) Chukchi
Pidgins & Creoles (1) Saramaccan
Elamo-Dravidian (1) Koya
Language Isolates (1) Ket
Uralic-Yukaghir (1) Votic
Kartvelian (1) Georgian
Table 3. Language sample used in this study12
3.2 Data collection
In the collection of the data, several decisions with respect to definitions had to be made. There is much theoretical debate about the differences between pronouns and agreement (see e.g. Fuss 2004: 168-181 for a discussion of criteria to distinguish the two), and for example clitics have been argued to be pronouns by some but agreement by others. It is also not clear how nanosyntax defines the difference. To avoid this complicated theoretical discussion, I choose to only include independent pronouns in the data collection. Clitics and affixes were therefore excluded.
12 The numbers after each (sub)family name indicate how many of them should be in the sample according to the application (Bakker p.c.) based on Rijkhoff & Bakker (1998).
27
For possessive constructions, I did not differentiate with respect to their grammatical status. All types of possessive constructions have been taken into account: independent pronouns, pronominal affixes, particle constructions etc.. This was done for two reasons. First of all, the decision of whether an element is possessive is not as controversial as the distinction between pronoun and agreement. Secondly, many language have affixal possessives and if these are excluded, the group of languages to select in the sample is very reduced. And especially with respect to the hypothesis on nominalized possessives it is interesting to see whether there is a difference in nominalization strategy between independent possessives and affixal possessives.
A second important decision deals with the definitions of case categories. In some grammars, pronouns are described as bearing case (nominative, accusative, etc.), while in others, they are described as expressing syntactic functions (subject, direct object, etc.). However, I decided to lump these terms together. I thus analyse a subject-form as being the form in the nominative case. The same holds for direct objects and accusative case, and for indirect objects and dative case. Note that I do not lump possessives together with the genitive case, as they are expected to behave differently based on the theory in section 2.
Of course not all languages display a nominative-accusative alignment, i.e. not all languages show a system in which the single argument of an intransitive clause is marked in the same way as the agens-argument in a transitive clause. Ergative-absolute alignment, in which the single argument of an intransitive clause is marked in the same way as the patiens-argument in a transitive clause, is also a possible strategy. As was described in section 2.5, Caha's (2009) work does not lead to a hypothesis for these languages. I have not excluded ergative languages from the sample, and I will come back to their relation to the hypotheses in section 5.1.
4 Data description
In this section, I will describe the collected data in general. I will first describe issues on data collection and analysis in individual languages, followed by information on the basic pronominal patterns in the languages in the sample. In the next section, these data will be analysed with respect to the hypotheses and predictions that were formulated in section 2.3.
Firstly, it is worthwhile to mention that of the total of 50 languages, all languages actually use personal pronouns. Sarcee (Na-Dene; Canada) might be regarded an exception to this, as personal pronouns occur seldom in this language and subject and object are expressed with verbal prefixes. Personal pronouns are only used as emphatic pronouns, and as such they can be used in all syntactic functions (Cook 1984: 62-63).
In many of the 50 languages, pronouns are not obligatory expressed. Sometimes the optionality of pronouns only applies to the subject, a phenomenon that is known as 'pro-drop' (Rizzi 1986; Neeleman & Szendroi 2007). In some other languages, such as Chinese
(Sino-28
Tibetan; China), pronouns in direct and indirect object function can also be omitted, which is known as 'radical pro-drop' in the literature (Neeleman & Szendröi 2007).
In this paper I will not make a difference between (radical) pro-drop and non-pro-drop languages. Most pro-drop languages have the opportunity to express the personal pronoun, and many of them also have different forms for different syntactic functions. In the analysis, I do not make a distinction between obligatory and optional pronouns. However, in some radical pro-drop languages it is a difficulty to gather data on pronouns: as the pronouns are not obligatory, many example sentences lack pronouns all together. As a result of this, it is very difficult to conclude whether a certain form can be used in accusative or dative function in these languages. My sample contains three of these heavily pro-dropping languages: Bororo (Amerind; Brazil; Crowell 1981), Koasati (Amerind; USA; Kimball 1991) and Alamblak (Indo-Pacific; Papua New Guinea; Bruce 1984).
I solved this difficulty by analysing these languages as pro-dropping, and I therefore analysed the accusative and/or dative function as 'not expressed with pronouns'. For Alamblak and Bororo, both accusative and dative were analysed as such. For Koasati, only the dative was analysed as 'not expressed with pronouns'. I made the same analysis for the dative in Tauya (Indo-Pacific; Papua New Guinea; MacDonald 1990).
Of the whole dataset, 14 language descriptions do not give information on nominalized possessives (i.e. equivalents of de mijne 'the mine'). These languages are listed in table 4. For some of these languages, the process of nominalization is described, but no examples of nominalized possessives are provided, neither is it explicitly mentioned that the nominalization process can be applied to possessives. The general nominalization process, if described, is mentioned in table 4 as well, together with comments on its applicability to different word classes.
Language General nominalization strategy? Comments
Nama not found
in equational sentence, -à attached to second NP, no examples with possessives (Hagman 1979: 114-115)
Chamorro not found
only roots (not affixes) can change word class with morphological process (Topping 1973: 101-105)
Atayal not found
all nominalization processes are explicitly stated to nominalize only verbs or adjectives (Rau 1992: 118-125)
29
246-247) and adjectives
Lavukaleve suffix –e or –i (Terrill 1999: 347-354) only verbs can be nominalized Bukiyip suffix –lúli (Conrad & Wogiga 1991: 44-45)
Alamblak not found copular verb –e attached to nuclei, including possessive phrases
Kisi suffix –ɔ̀ɔ́ or noun class system suffixes
(Tucker 1995: 201-204) process to nominalize verbs Tiriyo not found only constructions translated as 'the one
who has X'
different nominalizations, no examples on possessives
Koasati not found
only nominalization process to derive "agentive nouns" is described (Kimball 1991: 281-286)
Kiowa suffix –dé/ –gó (Watkins 1980: 139) can be affixed to "any non-verbal stem to derive a noun"
Sarcee suffix –i (Cook 1984: 210)
Japanese suffix –sa (Martin 2004: 909) process to nominalize
adjectives Chukchi suffix -lʔ- (Dunn 1999: 144)
Table 4. Languages without information on nominalized possessives
Apart from the 14 languages without information on nominalized possessives, there are several languages of which the data are 'difficult' to interpret: they have constructions serving as the translational equivalent of mine that are however structurally rather different. They can therefore be used less straightforwardly in this cross-linguistic comparison. For Dyirbal (Australian; Australia), a nominalized possessive is described, but the author doubts his own analysis (Dixon 1972: 105-107). Kharia (Indo-Pacific; India) does not have a nominalized possessive, but a construction that I would rather define as a 'verbalized possessive'. An example is given in (34).
(34) Kharia (Petersen 2011: 107)
íɲ ho=kaɽ=te íɲ=aʔ=yoˤi
1SG 3SG=OBL 1SG=POSS=ACTPST.1SG
'I made him/her mine.'
What this example shows is that the possessive (marked by =aʔ) can be combined with the active past marker, to create the meaning 'to make it mine/yours/etc.'. With other past tense markers, the meaning 'to become mine/yours/etc.' can be created (Petersen 2011: 107). I have
30
analysed Kharia as not having a nominalized possessive, with the remark that it does have a verbalized possessive that would be interesting for further study.
Kobon (Indo-Pacific; Papua New Guinea) is another language in which the translational equivalent of 'mine' is in fact a rather different construction, as is illustrated by the following sentence.
(35) Kobon (Davies 1981: 60)
gaisam u gaisam ade (ke)
peanut that peanut 1SG self
'Those peanuts are my peanuts' (i.e.: Those peanuts are mine)
The construction shows that this Kobon sentence does not contain a nominalized possessive: the noun is not omitted. A closer look into Kobon equational sentences can explain this: the noun cannot be omitted in Kobon, and English sentences like 'This man is good' have the construction 'man this man good' in Kobon (Davies 1981: 26). The noun can only be omitted if the suffix –rö or –bo is attached to the other element in the NP (ibid: 53), but there are no examples of these suffixes on possessives. I have therefore analysed Kobon as a language that does not have nominalized possessives.
The third language without nominalized possessives is Central Alaskan Yupik (Eskimo-Aleut; USA). For this language, it is explicitly stated by the author that the language does not express constructions like de mijne 'mine' (Miyaoka 2012: 721.) So, in total the sample contains three languages (6%) of which we can be sure to lack nominalized possessives.
The final point about which information is somewhat difficult to interpret for certain languages, is the third person. Six languages in the sample (12%) do not have a real third person pronoun: Kiowa (Amerind; USA; Watkis 1980), Lezgian (Caucasian; Azerbijan; Haspelmath 1993) and Lavukaleve (Indo-Pacific; Solomon Islands; Terrill 1999) use a demonstrative instead; Tsou (Austric; Taiwan; Szakos 1994) uses articles instead and for Dyirbal (Australian; Australia; Dixon 1972) and Cherokee (Amerind; USA; Montgomery-Anderson 2008) no alternative strategies are described.
This issue was solved by means of a practical approach: in case that the strategies of expressing a third person (e.g. demonstratives) were described, the information was taken into the data collection. In cases where the description did not provide enough information, the third person was analysed as 'not expressed with pronouns'.
Now that practical issues of the dataset have been described, I will provide some general information on the pronominal systems investigated. In this, I will focus on the pronominal features, expression of grammatical case, possessive constructions and nominalized
31
constructions. Although the features in the pronominal systems differ between the languages to a great extent, several distinctive features can be described. If possible, I will compare the patterns found in my sample with the patterns described in the WALS (World Atlas of Language
Structures, Dryer & Haspelmath 2013).
Firstly, most of the sample languages do not make a distinction between inclusive and exclusive first person plural. In other words, they do not distinguish between 'we, including the hearer' and 'we, excluding the hearer'. There is a different for form inclusive and exclusive first person plural in 18 languages (36%) in the sample13. These numbers are comparable to those in the WALS, in which 63 out of 200 languages (31,5%) exhibit the inclusive-exclusive distinction (Cysouw 2013). The geographical and genetic distribution of the languages over the typological values is also similar to that described by Cysouw (2013). For example, no language in Europe expresses the distinction, and languages from the Dravidian family (in this sample Koya), Munda sub-family (Kharia; part of Austric family) and languages from all Australian sub-families except Pama-Nyungan (Tiwi and Gooniyandi) do distinguish inclusive and exclusive.
Secondly, 35 of the languages (70%) do not make a gender distinction at all in their pronominals. Of the remaining languages14, most make a gender distinction in the third person (11 languages, 22% of the total sample), some also in other persons. Koromfe (Niger-Kordofanian; Mali & Burkina Faso; Rennison 1997) makes a semantic distinction in the third person, with different pronouns for human and non-human referents. In the WALS-sample, this distribution is similar: 254 out of 378 languages (67,2%) do not have grammatical gender, and the third person is the most common person to express gender distinctions if present in the language (122 languages, or 32%). The geographic and genetic distribution is also comparable: European and African languages (except from the Nilo-Saharan family) are typical gender-distinguishing languages, whereas Austronesian and Austro-Asiatic languages are not (Siewierska 2013).
With respect to number features, it can be mentioned that all languages distinguish between singular and plural pronouns. Nine languages (18%) also distinguish a dual pronoun15, used to refer to exactly two people. The WALS does not contain information on the feature dual (for nouns nor for pronouns), so a comparison is not possible in this case.
Languages also differ to a great extent in their case systems. Four of the fifty languages (8%) have an ergative pronominal case system, or characteristics of it. These languages are Tauya (Indo-Pacific; Papua New Guinea; MacDonald 1990), Gooniyandi (Australian; Australia; MacGregor 1990), Yupik (Eskimo-Aleut; USA; Miyaoka 2012) and Chukchi (Chukchi-Kamchatkan; Russia; Dunn 1999). The percentage of languages with an ergative pronominal
13 These languages are listed in table 1 in appendix 3. 14 These languages are listed in table 2 in appendix 3. 15 These languages are listed in table 3 in appendix 3.
32
case system in the WALS is somewhat higher: 11,6% (20 out of 172 languages), but the geographical spreading is comparable (Comrie 2013).
In my sample, 20 languages (40%) do not distinguish for case in personal pronouns16. So, these languages use one syncretic pronominal form for nominative, accusative and dative case. The WALS-sample contains a slightly higher percentage of languages with neutral case: 45,6% (79 out of 172 languages; Comrie 2013).
Some languages show variation in their pronominal paradigm, such that they do not distinguish case in parts of their paradigm but do distinguish case in other parts. Tidore (Indo-Pacific; Indonesia) only has a distinct nominative in the first person singular (van Staden 2000: 92-93), Saramaccan (Creole; Suriname) only has a distinct nominative in the third person singular and second person plural (McWhorter & Good 2012: 94-95) and Georgian (Kartvelian; Georgia) only has distinct cases in the third person singular and plural (Hewitt 1995: 76-77). Further information on case syncretisms will be provided in the next section, related to the hypotheses tested in this paper.
With respect to the possessive pronouns, it is remarkable that all languages in the sample have a construction to express possession: 36 languages (72%) have independent possessive pronouns17, all other languages express possession with a possessive affix on the noun denoting the possessed item. In Tidore and Japanese, the independent pronoun is accompanied by another element expressing possession. The phrases below illustrate the difference between independent (36a) and affixal (36b) possessive marking.
(36) a. Alamblak (Bruce 1984: 184) nan-ho kuñ-t 1SG-POSS house-3SGF 'my house' b. Chamorro (Topping 1973: 108) karetá-hu car-1SG.POSS 'my car'
Five languages that have independent pronouns, also have the possibility of using affixes. These different strategies relate to a distinction between alienable and inalienable possession in Kharia (Indo-Pacific; India; Petersen 2011), Tauya (Indo-Pacific; Papua New Guinea; MacDonald 1990) and Bororo (Amerind; Brazil; Crowell 1981). So, these languages make a distinction between
16 These languages are listed in table 4 in appendix 3. 17 These languages are listed in table 5 in appendix 3.
