Passive in Greenlandic Jeanette Sakel  950366 Specialeopgave på kandidatoverbygningsuddannelsen Vejleder: Peter Bakker Ekstern bedømmelse Institut for Lingvistik Aarhus Universitet, februar 1999 Table of contents page: Acknowledgements i Abbreviations ii The presentation of the examples iii Tables iv 1. Introduction 1 2. Terminology 2 2.1 Semantic roles and grammatical relations 3 2.2 Control and animacy 5 2.3 Valence and transitivity 6 2.4 Markedness 7 3. Voice 8 3.1 Different voice structures 8 3.1.1 Active 8 3.1.2 Passive 9 3.1.3 Antipassive 10 3.1.4 Middle 11 3.1.5 Inverse 14 3.1.6 Focus systems 15 3.1.7 Causative 16 3.1.8 Applicative 19 3.1.9 Other structures 21 3.2 Different definitions of voice 22 3.3 Voice and alignment 27 3.4 Summary 30 4. Passive 30 4.0 Introduction 30 4.1 The structure of the passive 32 4.1.1 The markedness of the passive 32 4.1.2 Different forms of passive marking 33 4.1.3 The promoted element 36 4.1.4 The demoted element 37 4.1.5 Impersonal passives 37 4.1.6 Stative and dynamic passives 38 4.2 Passives in ergative systems 40 4.3 Structures with the same formal marking as passives 40 4.4 The development of passives 41 4.4.1 Development of passives from inactive auxiliaries 42 4.4.2 Development of passives from NPreflexives 42 4.4.3 Development of passives from causatives 43 4.4.4 Development of passives from generalized subject constructions 44 4.4.5 Development of passives which do not arise via grammaticization 44 4.5 Development of ergative structures from passives 45 4.6 The functions of the passive 45 4.7 Summary 46 5. An introduction to the structure of Greenlandic 48 6. Voice in Greenlandic 55 6.1 Different voice structures 55 6.1.1 Active 56 6.1.2 Passive 56 6.1.3 Antipassive 56 6.1.4 Causative 57 6.1.5 Applicative 61 6.2 Voice in Greenlandic in terms of valence 62 6.3 Combinations of voice morphemes 63 6.4 Summary 65 7. Passive in Greenlandic 65 7.0 Introduction 65 7.1 The formal marking of the passive 65 7.1.1 neqar 66 7.1.2 tit 67 7.1.3 saa 71 7.1.4 sima 74 7.1.5 sariaqar 75 7.1.6 si & gi 76 7.1.7 Concluding section: the formal marking of the passive 76 7.2 The promoted element 78 7.3 The demoted element 79 7.4 Impersonal passives 81 7.5 Causatives functioning as passives? 82 7.6 The development of the passive 84 7.7 The functions of the passive 86 7.8 Summary 87 8. Conclusion 89 9. Appendix 93 10. References 94 Acknowledgements I want to thank all those who have contributed to this study either by their help as consultants on the Greenlandic language or by professional, financial or personal support. My first thanks go to the speakers of Greenlandic: the students, teachers, and staff of the Knud Rasmussen Højskole in Sisimiut, especially to the principal of the school, Lars Olsvig, who gave me a warm welcome. In particular, I want to thank my main consultant David Poulsen and my other informants Hans V. Kreutzmann, Mette Petersen, Anda Uitsatikitseq, Poul Eliasarsen and Johan Eliasen for sharing the secrets of their language with me. I also want to thank my informant in Denmark, Ebbe Qvist. For the purpose of my fieldwork in Greenland, I got financial support from the following institutions: the Faculty of Humanities at the University of Aarhus, which enabled me to purchase the plain ticket to Greenland for my fieldwork, and the Linguistic Institute at the University of Aarhus for the payment of my consultants. Finally, I want to thank Peter Bakker for an excellent professional support. Moreover, I want to thank Morten who always was there for me, which was the most important thing of all when writing this thesis. Abbreviations A transitive subject AB.PART abstract participle ABL ablative case ABL2 ablative 2 (see footnote 23 in 6.3) ABS absolutive case ACC accusative AD adessive case AGN agentive nominalization ALL allative case ANTI antipassive APPL applicative ASP aspect AUX auxiliary C consonant CAU.MOOD causative mood CAUS causative CL7 class 7 of Bantu noun classes CL8 class 8 of Bantu noun classes DAT dative DEF definite DESUBJ desubjective DET determiner, article ERG ergative F feminine gender FUT future tense ID indefinite IMP.PASS impersonal passive IMPF imperfect/ imperfective aspect INAN inanimate INDIC indicative mood INF infinitive INSTR instrumental case INSTR.FOCUS instrument focus LOC locative case/ locative adposition M masculine gender N neuter gender NOM nominative O (direct) object PASS passive PASS/CAUS passive or causative PASS.PART passive participle PAST past tense PART participle PERF perfective aspect PL plural POR possessor PUM possessum POSS possession (more general than POR and PUM) RECIP reciprocal REFL reflexive pronoun, reflexive marking S intransitive subject SG singular T tense T/ASP tense/ aspect The presentation of the examples All examples are marked for where they come from, e.g. Fortescue (1984). The examples from my fieldwork are marked as by the initials of my informants (to be presented in the introduction). The spelling of the Greenlandic examples is in the ordinary standardized spelling of Greenlandic, introduced in 1973. The notation of my examples contains some symbols whose meaning I briefly want to explain. ‘’ this symbol stands between separate morphemes in the Greenlandic examples. These morphemes are translated morpheme by morpheme in the notion below e.g.: ateqar namehave ‘.’ this symbol divides morphemes into their different grammatical contents. This is the case when a single morpheme in the Greenlandic example contains more than one grammatical category or more than two lexical elements in the translation. The different categories are thus separated in the notation by this symbol. This is shown in the following example with the verbal inflectional ending in Greenlandic, which is a fused ending; i.e. it contains more than one grammatical category: ateqarpunga namehave1SG.INDIC The Greenlandic verb inflects for both the subject and object of a transitive clause. In the notation, the person and number of the subject are mentioned before the object e.g.: takuara see1SG.3SG.INDIC In order to facilitate analysis of the forms, I list the personal inflection suffixes in the appendix of chapter 9. Tables page: 1 The animacy hierarchy 5 2 Passive 10 3 Antipassive 11 4 Middle voice 12 5 Inverse 14 6 Hierarchy of direct/ inverse marking in Cree 14 7 Instrument focus 16 8 Analytic causative 16 9 The case of the causee 18 10 Morphological causative from an intransitive active 18 11 Morphological causative from a transitive active 18 12 Applicative from an intransitive active 20 13 Applicative from a transitive active 21 14 Structures treated under the heading of voice in different theories 26 15 Accusative and ergative alignment 27 16 Dik’s (1997) ‘semantic function hierarchy’ 36 17 Consonants in Greenlandic 48 18 The structure of nominals and verbs in Greenlandic 51 19 Formal case marking in Greenlandic 54 20 Voice structures in Greenlandic 55 21 The cases of agent phrases in different passive and causative constructions 79 22 The different passive affixes 88 Aliortukkamut ini 003miit eqqaassutissaatippara (Dedicated to the ghost in room 003) 1. Introduction The present study is my MA thesis on passive in Greenlandic. My aim is to investigate the passive in Greenlandic in relation to the forms of this category and their uses. The passive is a form of ‘voice’. The first time a voice structure was described for a language was probably around 500 BC, when Panini described the grammar of Sanskrit. Since then, voice has played an important role in a variety of grammatical theories. According to Klaiman (1991), the term voice is directly translated from Latin vox since it refers to “the different ways a verb might be (that means ‘sounded’)” (Klaiman 1991: 2). Voice is a verbal category, which means that it is expressed in the verbal phrase and that it modifies the form and meaning of the verb. There are several different voice structures, such as middle, antipassive, and passive. The passive voice plays an important role in a number of grammatical theories, some of which try to explain the similarities between active and passive clauses in terms of transformation of arguments. Zellig Harris (1951) introduced the notion of transformation in his theory of grammar, which later was elaborated by his student Noam Chomsky (1957 and 1965). In this way, transformations became an important part of ‘Generative Grammar’. ‘Relational Grammar’ by D. Perlmutter & P. Postal (1977) developed from Generative Grammar and therefore also in this framework the passive plays an important role. All the approaches mentioned so far have the passive as an important part of their theory. However, these approaches usually only describe the simplest cases of the passive. In addition, also other theories sometimes only involve the simplest cases of the passive. This is the case in functional grammatical theories, such as ‘Functional Grammar’ by Simon Dik (1997) and ‘Role and Reference Grammar’ by W. Foley & R. Van Valin (1984). These focus on the functions of the different components of grammar. In this way, Foley & Van Valin describe passive as a category which structures discourse by changing around the positions of the arguments in the clause. If one wants to find a full description of the structure of the passive, one has to reach for language typology, which has the aim to describe the diversity of categories, but in this way also their unity in the languages of the world. I want to put in the remark that functional grammars often are based on insights in language typology. Thus, Foley & Van Valin’s Role and Reference Grammar is built on data from languages of different parts of the world, albeit primarily focussing on the functions of these structures. Other descriptions of the passive in work in language typology look at this structure from morphological or syntactical points of view: Haspelmath (1990) describes the passive in terms of morphology, whereas Klaiman (1991) and Siewierska (1984) primarily describe it from the point of view of syntax. The present study is a contribution to the typological study of passive and voice phenomena. I will discuss different kinds of typological descriptions of the passive and other voice structures mostly for the purpose of investigating the passive in West Greenlandic. The passive is not very deeply investigated in the descriptions of Greenlandic. Most grammatical descriptions only devote a few lines to this structure. The first grammatical sketches and word lists of Greenlandic by Egede (17225) do not include any remarks at all about the passive. His colleague, Albert Top (1727), however, mentions the passive use of a certain morpheme (tit, to be discussed in 7.1.2). This morpheme is first mentioned again in Fortescue’s (1984) grammar, and no other grammatical descriptions seem to include it. Other descriptions of Greenlandic, such as those by Kleinschmidt (1851), Rasmussen (1888), and SchultzLorentzen (1945: 95), only mention the passive very briefly and all of them mention the same two passive markers (-neqar- and -saa-, to be discussed in 7.1.1 and 7.1.3). Fortescue (1984) treats the passive on two pages of his grammar, mentioning six different morphemes used for this structure. I will take my point of departure in Fortescue’s description of the passive. The analysis of my own fieldwork data, however, will result in an alternative description of the forms and functions of the passive in Greenlandic. Thus, the present study is partly based on already existing descriptions of Greenlandic and partly on my own fieldwork. Apart from my own fieldwork data, I use as many examples as possible from reliable grammatical studies. I checked most of the examples from the literature with my informants in Greenland. I did my fieldwork in Sisimiut, Greenland, in November 1998. My informants were primarily students from the West Coast of Greenland who were staying at the ‘Knud Rasmussen Højskole’ in Sisimiut. Mostly, this fieldwork was based on elicitation since I wanted to find out about the different forms and uses of the passive and since the texts I recorded hardly included any passive morphemes at all. According to Payne (1997), elicitation is not appropriate for an analysis of voice phenomena. However, considering the limited time, elicitation fitted best for my aim to find out about the forms and uses of the passive. My fieldwork in Greenland is comprehensively described in my paper ‘fieldwork on passive in Greenlandic’ (Sakel 1998c). Furthermore, I had done fieldwork in Århus, Denmark, in 1998 and I did supplementary fieldwork in Denmark in 1999. When citing examples from my fieldwork in Greenland and Denmark, I always give the initials of my informants: DP is David Poulsen from Qeqertarsuatsiaat in Greenland, HK is Hans V. Kreutzmann from Sisimiut in Greenland and EQ is Ebbe Qvist, living in Århus, Denmark. I am grateful for the kind permission of my informants to use the language data in the present thesis. The aim of the present study is to investigate the forms and uses of the different passives in Greenlandic. Before the description of the passive in Greenlandic, I will discuss the definitions of voice and passive in several typological approaches to these structures. Furthermore, I describe the different voice structures in Greenlandic. This study is therefore divided into two parts: 1. The general theoretical background, including chapter 2 about terminology, chapter 3 about voice, and chapter 4 about the passive. 2. The Greenlandic part, including chapter 5, which is an introduction to Greenlandic grammar, chapter 6 about voice in Greenlandic and finally the comprehensive and important chapter 7 about the passive in Greenlandic. My results are summarized and discussed in the conclusion in chapter 8. 2. Terminology Before proceeding to the definition of voice, I will explain a number of concepts which are important in the description of this category. These concepts are semantic roles and grammatical relations, control, valence, transitivity, and finally markedness. 2.1 Semantic roles and grammatical relations Voice typically affects the different links between semantic roles and grammatical relations. Therefore, semantic roles and grammatical relations are important notions which I briefly want to explain here. According to Comrie (1989: 656), grammatical relations are notions such as subject, object, indirect object, and oblique arguments. Subject and object, and indirect objects (in languages that have these), are typically called ‘core arguments’ since they are purely grammatical and express the relationship of the argument with the verb. The core arguments can be arranged in different ways, and this will be reflected in their case marking. In socalled ergative languages (to be discussed in 3.3), an object is marked with the same case as an intransitive subject while the transitive subject is marked differently. Therefore, linguists usually distinguish between intransitive subjects (S), transitive subjects (A) and objects (O). Between brackets, I gave Dixon’s (1994) abbreviations for these relations. Comrie (1989) uses these abbreviations as well, but instead of ‘O’, he uses ‘P’ for the object. I will use these abbreviations in the following chapters of this study. Semantic roles are crosslinguistically identified as roles expressing a certain meaning: the role of the agent expresses a person volitionally acting on somebody or something, the role of the patient expresses someone or something being acted upon, etc. These roles often coincide with grammatical relations such as subject, object, or indirect object. Thus, the agent is typically the subject and the patient is typically the object of a clause. However, semantic roles are not morphosyntactic categories themselves. In this way, a patient may also be for example in subject relation in a clause (see examples at the end of this section). In this discussion of semantic roles, I will mainly refer to Payne’s (1997) wellarranged definition of different semantic roles. He includes the following roles: agent, force, instrument, experiencer, recipient, and patient According to Payne (1997: 49), the semantic roles he lists are from Comrie (1989: 5253). Actually, the page reference in Comrie is wrong. Furthermore, Comrie (1989: 58) only lists the roles agent, force, instrument, experiencer, and patient; i.e. Comrie does not include the role of the recipient. , which I want to explain briefly. Payne (1997: 49) defines the agent in the following way: “A prototypical agent is conscious, acts with volition (on purpose), and performs an action that has a physical, visible effect. It is a powerful controller of an event.” The agent is a typical role in subject function, but also other semantic roles may be subjects or the agent might be expressed as an oblique. In John hits Bill, John is the agent since he consciously acts on Bill with a physical effect. A role which is closely related to the role of the agent is force. A force is inanimate, such as storm in the storm broke our roof. The difference between agent and force is that the agent acts consciously while the force is not conscious about the action. Payne (1997: 49) defines force as “an entity that instigates an action, but not consciously or voluntarily.” The instrument lies semantically in between the agent and the effect of an action. Payne (1997: 49) defines instrument as “an entity that instigates an action indirectly”. Knife is the instrument in I opened the letter with a knife. The instrument can also be expressed with subject function as in the knife opened the letter. According to Payne (1997: 50), “an experiencer neither controls nor is visibly affected by an action. Normally an experiencer is an entity that receives a sensory impression, or in some other way is the locus of some event or activity that involves neither volition nor a change of state.” Thus, the experiencer has to be animate since it is sentient. In John saw the robber, John is the experiencer with subject function. Like agent and experiencer, a recipient is usually animate. According to Payne (1997: 50), “a recipient is the typically animate destination of some moving object.” An inanimate entity receiving something is called destination in Payne’s (1997: 50) terminology. He gives the following examples with the recipient Mary (1) and the destination France (2): (1) I sent the book to Mary. (2) I sent the book to France. Finally, the patient is the typical undergoer of an action, which Payne (1997: 51) expresses in that it usually “undergoes a physical, visible change of state”. In the example I gave to describe the agent above, John hits Bill, Bill is the patient since he undergoes a visible change of state. All of these semantic roles can play a role in voice since they can be expressed in different grammatical relations. According to Comrie (1989: 59), the semantic roles are not principally distinct terms but are on a continuum in terms of control. Thus, the agent has control over the situation while the force is acting but has no control. In the next section, I will shortly introduce the notion of control. Apart from the semantic roles described above, there are other semantic roles, such as location, direction, setting, purpose, time, and manner. I do not discuss them in detail here, as they play only a minor role in voice distinctions. These additional semantic roles are often expressed by oblique phrases but may in some cases be expressed as subject or object as well. Payne (1997: 49) has the following examples of the location bed in subject position in (3) and the setting Norway in object position in (4): (3) This bed was slept in by Che Guevara. (4) We did Norway last summer. Usually, the different semantic roles are encoded in certain typical ways linguistically. Often the agent is the subject of a clause but also other roles may be grammatical subjects. For example the patient is often the object of a clause but it can be subject as well in passive constructions like in John was beaten, where John has the semantic role of the patient and is the subject of the clause. Usually, verbs have inherent semantic roles. Thus, the verb receive typically has a recipient as the subject, a patient as the object, and the agent in oblique position. The verb kill typically has an agent as its subject and a patient as its object. The assignment of semantic roles to the arguments of a verb may, however, be changed by the application of voice. 2.2 Control and animacy Predicates might differ in terms of control. Klaiman (1991: 114) gives the following definition of control, following Lacey (1979: 11): Control relates to a state of options or choices an individual may have in a given situation. A person is said to control an outcome if and only if the person can make a voluntary response that increases the possibility of occurrence of that outcome, whereby ‘voluntary response’ is meant a response that increases in probability when rewarded and decreases when punished. In terms of control, predicates can be divided into control and noncontrol predicates, i.e. predicates expressing control, such as hit, and predicates expressing no control, such as bleed. Thus, control predicates have to do with actions while noncontrol predicates usually do not involve actions but states. Control predicates can be divided into agent locus and undergoer locus. In the agent locus, the subject has control over the action; a verb of this kind is hit. In the undergoer locus, the subject has no control, which means that the subject is not an agent. The undergoer locus contains verbs such as spread, melt and all verbs which are derived by a passive. Noncontrol predicates may have an inanimate subject such as the verb bloom, or animate subjects such as the verbs bleed and hiccup. The difference between animates and inanimates is that animates are intentional although they do not have control while inanimates are nonintentional. Animacy is an important notion in grammar. In many languages, grammatical structures work in terms of animacy. One such structure is case marking, where animate NPs or NPs higher in animacy than other NPs can be marked differently from e.g. inanimate elements. According to Comrie (1989: 186), “the existence of a separate accusative case frequently correlates with higher degree of animacy”. In this way, the personal pronouns in Danish, being typically high in animacy, have nominative, genitive, and accusative forms while other NPs only have nominative and genitive case forms. The animacy hierarchy, first introduced by Silverstein (1976), shows which elements are more likely to be treated as ‘higher in animacy’ than others. This hierarchy, of which slightly different versions have been proposed elsewhere, is presented in the following way by Dixon (1994: 85): Common nouns 1st person 2nd person Demonstratives, Proper Human Animate Inanimate pronouns pronouns 3rd person nouns pronouns < more likely to be in A than in O function Table 1: the animacy hierarchy The more to the left an element or a group of elements is on this scale the higher the element is in degree of animacy. Animacy can be seen not only in case marking but also in other structures of grammar. Thus, elements higher on the hierarchy are more likely to function as transitive agents than elements lower on the hierarchy. According to Comrie (1989: 185), in many languages the notion of animacy involves only differences between animate – inanimate or human – nonhuman. Some languages have more fine-graded differences. Animates, for example, can be divided into ‘higher’ animates, such as pets, ‘intelligent’ animals, etc., and ‘lower’ animates, such as mosquitoes. One of the passive forms in Greenlandic appears to show an instance of this difference in animacy (to be discussed in 7.1.2). 2.3 Valence and transitivity A central notion in the description of voice is ‘valence’. This term was introduced by Tesnière (1959) to cover the differing numbers of obligatory participants in clauses. A verb with one obligatory participant is called intransitive, a verb with two obligatory participants is transitive, and a verb with three obligatory participants is ditransitive. Obligatory participants in a clause are called ‘core arguments’ as opposed to ‘oblique elements’. The latter are not obligatory and can be freely omitted. Different grammatical models use different terms referring to the same distinction between strictly grammatical arguments. The verbs usually have a fixed relation between the core arguments and the semantic roles (see 2.1). In this way, a verb such as kill in Rose kills John is transitive because it involves two obligatory participants: the one doing the killing and the one being killed. Furthermore, one of these two participants has to be the agent which in this case is Rose, while the other one, in this case John, is the patient of the killing. Payne (1997) distinguishes between semantic and syntactic valence. The semantic valence of a verb never varies; i.e. a verb has always a fixed number of logically possible semantic roles which are assigned to this verb. Thus, the verb kill always has two semantically obligatory participants. When one of these is omitted, e.g. in a passive clause, the semantic valence of the verb is not changed. The verb still implies all possible arguments semantically even if these are not present. Thus in John was killed, the semantic valence of the verb is ‘two’, i.e. the verb is semantically transitive as the agent of the killing is implied in the verb. The syntactic valence, on the other hand, always has to do with the actual number of core arguments in a given clause. By application of the passive, the syntactic valence is usually decreased, since one less core argument is present in the passive clause than in the active clause (to be discussed in chapter 4). Thus, in the passive clause John was killed, the syntactic valence is ‘one’ since there is only one syntactic coreargument present. In the analysis of voice phenomena, usually only the syntactic valence plays a central role as it refers to the decrease or increase of the number of syntactic arguments. Thus, when speaking about valence in the present study, I refer to syntactic valence. A notion closely connected to valence is transitivity. A verb with two core arguments is traditionally called a transitive verb. An influential approach to transitivity is Hopper & Thompson (1980). They distinguish degrees of transitivity. Hopper & Thompson define transitivity according to the following ten parameters: 1. two participants or more, 2. action, 3. telic aspect, 4. punctuality, 5. volitionality, 6. affirmation, 7. realis mode, 8. agency, 9. the total affectedness of the object, 10. a highly individual object. In this way, Hopper & Thompson define transitivity as a matter of degree, dependent on how many of the above parameters are present. A verb is thus high in transitivity when it involves many or most of the parameters, and a verb is low in transitivity when it involves only few of these parameters. However, this definition of transitivity cannot always explain all relevant structures in grammar. According to Siewierska (1984: 16), sentences which differ in the degree of transitivity may, nevertheless, have the same morphological and syntactic structures, such as the possibility of passivization (see example (5)(7) below). Furthermore, clauses which are equally high in transitivity do not always behave in the same manner syntactically. Siewierska (1984: 1718) shows this with the following examples: (5) Sandra understands things. > Things are understood by Sandra. (6) John lacks courage. > *Courage is lacked by John. Both (5) and (6) involve two participants and do not seem to differ for the other parameters in these clauses either. However, (5) can be passivized while (6) can not. In this way (5), which is low in transitivity, can be passivized in the same way as highly transitive clauses, such as (7) from Siewierska (1984: 16): (7) John killed Betty. > Betty was killed by John. Thus, transitivity described by Hopper & Thompson (1980) does not cover all aspects of grammar, especially not those involving the passive. In this study, I will therefore primarily use the notion of transitivity in its traditional sense: a clause is transitive when it has two obligatory arguments and intransitive when it has only one obligatory argument. In Greenlandic, as I will discuss in detail in chapter 5, the difference between transitive and intransitive clauses is marked in the inflectional endings of the verb: intransitive clauses crossrefer to a single argument, whereas transitive clauses are crossreferred to two arguments. 2.4 Markedness Markedness is an important notion in typology. According to Croft (1990), the term markedness was first introduced by the members of the Prague School in the 1920s and 1930s. The notion of markedness is used to distinguish between two different kinds of structures: some structures are very frequent and marked overtly, as opposed to other structures which are not very frequent and have additional marking. The first group of forms can thus be described as ‘basic’ or ‘unmarked’ while the second forms are ‘marked’. Markedness can be described in terms of formal marking, such as a longer form, and functional marking, such as the frequency of the structure. These are expected to correlate. In the present section, I want to give a definition of markedness following Comrie (1988) and Greenberg (1966). Comrie (1988: 1929) defines markedness in terms of raw frequency, formal complexity, degree of productivity, and discourse distribution. I will shortly explain how markedness is related to these terms: Raw frequency: unmarked structures are more frequent in discourse than marked structures. Formal complexity: the marked structure is formally more complex than the unmarked structure, i.e. the marked structure contains more morphemes or phonemes than the unmarked structure while the unmarked structure often is expressed by zero marking. Degree of productivity: the unmarked structure is usually more productive than the marked structure. Discourse distribution: marked structures occur in specific positions within a discourse, whereas unmarked structures are less restricted. Greenberg (1966) has eight criteria for morphosyntactic markedness, seven of which are applicable to all grammatical categories. One of his criteria involves only number and is therefore not discussed here. In addition to the criteria for markedness which Comrie has, Greenberg discusses the following: Unmarked forms often involve syncretization. According to Greenberg (1966: 27), “distinctions existing in the unmarked member are often neutralized in the marked categories”. Unmarked forms may in certain contexts also take the function of marked forms and not the other way around. Greenberg (1966: 28) calls this ‘facultative expression’. Finally, unmarked forms have often more allomorphs or irregular forms than marked structures. Markedness is relevant to my study since all voice structures can be described as marked as opposed to the ‘active’ voice. This will be discussed in the following chapter about voice. 3. Voice In chapter 2 I have dealt with some concepts which are important in the analysis of voice structures such as passive. I will now turn over to discuss voice in general. There are many approaches to voice. They differ as to which structures they include under the heading of voice and in their definition of this category. I will discuss both in the present chapter. 3.1 Different voice structures Before I describe different theories of voice, I want to introduce a number of structures which often are described under the heading of voice. Not all of these structures appear in all theories, as I will show in section 3.2. 3.1.1 Active The active is the basic form in a voice system. In many languages it is the most frequent form in discourse, it is formally less complex, more productive, and finally less restricted in its use than the other voice structures. Thus, the active is the unmarked voice (cf. the discussion of markedness in 2.4). The other voice structures to be discussed in this chapter may be described as marked. With the voice structure ‘inverse’ (to be discussed in 3.1.5 below), the active is traditionally called ‘direct’. In socalled focus systems the active may be called ‘agent focus’. The other voice structures can often be derived from the form of the active voice. Usually, the other voice structures are marked morphologically by an affix or analytically by an auxiliary. In this way, other voice structures are more complex than the active which is often realized by zero marking. In some languages, however, there might be equally complex marking and equally high text frequency for all voice structures. This is the case in the focus voice system of the languages of the Philippines, according to Shibatani (1988: 5) (to be discussed in 3.1.6). In these languages, both agent focus and patient focus are equally frequent and marked with the same complexity. Thus, in this type of language, the ‘active’ agent focus cannot be described as more basic than the patient focus. For almost all other languages, however, the active is clearly unmarked 3.1.2 Passive There are not only very different definitions of voice, there are also many different theories about the passive. I will discuss different theories of the passive in chapter 4. In the present section, I want to give only a brief introduction to the structure ‘passive’, one which would be accepted by the proponents of most of the theories to be discussed in chapter 4. Passive is a structure that in most cases decreases valence. The agent which is usually the subject of an active transitive clause is demoted to oblique status while another participant, usually the patient in object position, is promoted to subject status. The transitive verb in the active form has two core arguments, the subject and the object, while the verb in the passive only has one core syntactic argument, the subject. In this way, the valence of the active clause is decreased in the passive clause. The passive is usually marked morphologically in the verb or analytically in the verb phrase. If it is marked by an affix, the structure is called a ‘morphological passive’. When it is marked by an auxiliary in connection with the verb, the structure is called ‘analytic passive’. In both structures, the passive involves more complex marking than the active. Usually, the form of the passive verb can be directly deducted from the active verb, in which case one can say that the passive is derived from the active. Some theories include socalled lexical passives which consist of two suppletive verbal forms. One verb covers the semantic range of an active while the other one covers that of a passive. In case of these lexical structures, the passive is neither marked in relation to the active, nor derived from the active. I will come back to lexical passives in chapter 4. The following example from English shows an active clause (8) and the corresponding passive clause (9): (8) Bill kissed Monica. (9) Monica was kissed by Bill. In example (9), the passive is expressed analytically by the auxiliary was and the past participle form of the verb kiss. The patient Monica is the object in the active clause (8) and the subject in the passive clause (9), while the agent Bill is subject in (8) and demoted to an oblique, i.e. an argument which is in noncore relation to the verb, in (9) and marked with a preposition. The typical passive ‘derivation’ from an active to a passive can be shown in the following table: active: A O V passive: S V-PASS (OBL) Table 2: passive It can be seen in this table that the subject of the active clause is demoted to an oblique constituent and a nonsubject, in this case the object, is promoted to subject position. Thus, the passive has both a backgrounding, i.e. demoting, and a foregrounding, i.e. promoting function, as well as a function of inactivization of the situation. According to Haspelmath (1990), in the ‘Gramcats’ sample by Bybee et al. (published in 1994) 31 out of 80 languages (38,8%), have a passive. In the sample by Nichols (1992), 79 out of 174 languages (45,4%) have a passive. The two figures are nearly identical. Thus, one may conclude that somewhat fewer than half of the languages of the world seem to have a passive. 3.1.3 Antipassive Like the passive, the antipassive can be defined as a valence decreasing operation. But as opposed to the passive, the antipassive changes a transitive subject to an intransitive subject while the direct object of the active clause is demoted to an oblique constituent, or it is not expressed at all. In such cases, the verb is marked for antipassive voice. Shibatani (1988: 5) defines the antipassive in the following way: “The antipassive voice denies grammatical prominence to the patient nominal by either encoding it as oblique constituent or not syntactically encoding it at all. A typical consequence of antipassivization is the promotion of an agent to the most grammatically prominent constituent.” The following example of an antipassive is from Chamorro (Western Austronesian) from Cooreman (1994: 59): Cooreman (1994) presented the example in this way. I have no explanation for the forms of the verbs in this example. (10) Unpatek i ga’lago. ERG.2SGkick the dog ‘You kicked the dog.’ (11) Mamatek hao gi ga’lago. ANTI.kick 2SG.ABS LOC dog ‘You kicked at the dog.’ In this example, the transitive subject, marked by the ergative case, is changed into an intransitive subject in the absolutive case (ergativity will be explained in 3.3 below), while the object of the active clause (10) becomes an oblique element with the locative preposition gi in (11). At the same time, the verb is morphologically marked for the antipassive. Morphological marking of the antipassive seems to be the only possibility, since I have only found antipassives that are morphologically marked in the literature on the subject and no analytic antipassive, in contrast to the existence of analytic passives above. The two sentences differ in meaning. In example (10), the O is totally affected by the action, while in (11) the O is only partially affected. The antipassive is used to express the partial affectedness of the O. In Chamorro, the antipassive also has other functions; for example, when an object is indefinite the antipassive is applied. A schematic presentation of a typical antipassive ‘derivation’ is shown in the following table: active: A O V antipassive: S VANTI (OBL) Table 3: antipassive Usually the antipassive is found in ergative languages (see 3.3) or languages which have a special focus on the patient, as will be discussed in 3.3. 3.1.4 Middle The middle voice, or mediopassive, is well known from some IndoEuropean languages, such as Sanskrit, Classical Greek, and Romani. The category is also found in many other language families in the world, such as Bantu languages (Africa) or Austronesian languages (Asia). The term ‘middle voice’ indicates that this structure originally was thought of as lying between the active and the passive. The reason for this is that the middle voice has a participant with both active and passive characteristics. This can be seen in the following example from Classical Sanskrit (IndoEuropean) as given by Klaiman (1991: 24): (12) Devadattah katam karoti. Devadatta.NOM mat.ACC make.SG.ACTIVE ‘Devadatta makes a mat.’ (13) Devadattah katam kurute. Devadatta.NOM mat.ACC make.SG.MIDDLE ‘Devadatta makes (himself) a mat.’ In the middle construction (13) the agent both performs  an active characteristic  and benefits from the action  a passive characteristic. This differs clearly from the active clause (12) where the agent simply performs the action, and the action has no impact on the agent. The middle structure in the above example can be shown in the following table: active: A O V middle: A O VMIDDLE Table 4: middle voice I call this type of middle construction the ‘reflexive middle’ since it resembles reflexive structures; i.e. structures in which the same participant is in both subject and object position in a clause. Below I will introduce other types of middles. In the above examples (12) and (13) the active and middle are expressed morphologically by different inflectional paradigms in the verb. The middle voice may also be expressed analytically as in the following example (14) (Kemmer 1993: 147) from German, or lexically as in English, shown in example (15) (Kemmer 1993: 2): (14) Dieses Buch liest sich gut. this.N.NOM book.NOM read.3SG REFL well. ‘This book reads well.’ (rough approximation) (15) The door opened. The middle voice is expressed analytically by the use of the reflexive pronoun sich in example (14) and lexically, or rather semantically, in (15) by the use of the same verbal form opened as in the transitive active. However, the clause in (15) is interpreted as a middle and not as an active since the only argument the door, being inanimate, could not have caused the action and since there is no second argument in the clause. In example (13), the agent or initiator of the action is benefiting from it at the same time, but in example (14) and (15) there is no such coreference between the participants in the context. In (14), the book is read by someone who is implied but not expressed because this person is pragmatically deemphasized. Kemmer (1993: 149) calls the latter type of middle construction a ‘passive middle’ since it is very close to the passive voice in having an implied agent. The difference between the passive middle and the passive is, according to Kemmer, that the implied agent in the passive middle is more deemphasized than it is in the passive. In example (15), no causer of the action is mentioned and the meaning of this clause is that the door simply opened while nobody knows how. Kemmer (1993: 142) calls this type of middle construction ‘spontaneous events’. Spontaneous events are sometimes also called ‘anticausatives’. This name reflects that spontaneous events are logically opposite causative constructions because of the deemphasis of the agent. Payne (1997: 218) describes anticausatives in the following way: “instead of starting with a noncausative verb and adding a morpheme to make it causative, a middle construction [i.e. anticausative, J.S.] starts with a causative verb and results in a noncausative verb”. This can be seen in (15), where the causer of the opening of the door is not expressed, while in the corresponding active clause The man opened the door or The wind opened the door the verb opened has a causative sense since there is an overt agent or force that causes the action. There are thus three types of middles: reflexive middles, passive middles, and spontaneous events. All these types can be expressed morphologically, lexically, or analytically. According to Kemmer (1993:197), the most common type of middles is the reflexive middle which might have to do with the fact that most middle structures develop historically from reflexives. This might also be the reason why the reflexive pronoun sich is used to mark the passive middle in (14). Even if many middles develop from reflexives, there are a number of differences between these two structures. Payne (1997: 218) describes these differences in terms of process and action: “middle constructions express the scene as a process whereas reflexives and passives express the situation as an action”. This means that middles focus on the process which is undergone by the patient, while reflexives focus on the action which is carried out by the agent. I briefly want to explain how the three types of middles discussed above can be described in terms of processes. In clauses with a ‘passive middle’ or ‘spontaneous events’, no overt agent is present while a patient or another nonagent participant is the subject. In these clauses, the focus is on the process undergone by the nonagent participant since there is no agent present. In case of the ‘reflexive middle’ there is an overt agent, which is coreferent with another participant in the clause. The focus in a reflexive middle is on this coreference of a participant with the agent of the clause. The situation may thus be described as a process from the viewpoint of the coreferent participant instead of an action carried out by the agent. The coreferent participants can be implied, as in the example from Sanskrit (13) above. In active constructions, on the other hand, the focus is on the action carried out by an agent. In addition, in reflexive structures the action is focussed. According to Kemmer (1993: 208), all the types of middle voice can be defined in terms of low degree of elaboration. In all types of middles, usually one participant less is expressed than semantically possible: in the passive middles and spontaneous events only a nonagent is expressed while the same participant is found in two relations in reflexive middles. The low degree of elaboration in middles makes it sometimes difficult to describe these structures in terms of transitivity. Passive middles and spontaneous events can clearly be identified as intransitive since they only involve one participant. Reflexive middles, on the other hand, always involve two or more core arguments and are thus syntactically transitive or ditransitive structures. From the point of view of the number of participants, on the other hand, reflexive middles can often be described as intransitive since only one participant is present in both subject and object position. Because two types of middles can be described as formally intransitive while another type is formally transitive, whether these are considered transitive, or intransitive depends on the type of middle voice a certain theory focuses on. Payne (1997) and Foley & Van Valin (1984) focus primarily on spontaneous events and anticausatives in their definitions of middle voice. In this way, these authors claim that middles involve detransitivization. Klaiman (1991), on the other hand, focuses on reflexive middles and claims that middles do not involve any changes in the valence. Klaiman (1991: 24) even calls the middle the ‘basic voice’ since there are no changes in the valence or in the grammatical relations when this voice structure is applied. Kemmer (1993: 210) includes all three types of middles in her definition and distinguishes them mainly in terms of their low degree of elaboration: “the general correlation of middle morphosyntax with formal intransitivity that has been observed is not due to the fact that MMs i.e. middle markers, J.S. are meaningless markers of intransitivity, but rather to the property of low degree of elaboration of events that is inherent in middle semantics”. Thus Kemmer describes the middle voice as a structure involving fewer participants than theoretically possible due to low degree of elaboration and not because of detransitivization. 3.1.5 Inverse Inverses are known from Cree (Algonquian), but also other languages in the world have inverse structures. In some definitions of voice (to be discussed in more detail in 3.2), inverses are treated as voice structures. Inverses are structures which change the semantic roles in a clause without adding or removing any arguments. This can be shown in the following example from Plains Cree (Algonquian) from Dahlstrom (1991), cited in Payne (1997: 211): (16) Nise:kihawak. 1SGfrightenDIRECT3PL ‘I frighten them.’ (direct) (17) Nise:kihikwak. 1SGfrightenINVERSE3PL ‘They frighten me.’ (inverse) There are languages where both the direct and the inverse are marked and languages where only the inverse is marked. In Cree, both direct and inverse are morphologically marked. Direct is marked by the affix a in (16) and inverse is marked by the affix ik in (17). This marking is the only indication for which argument is acting on which in the clause. The structure of a typical inverse is shown in the following table: direct: A O V(DIRECT) inverse: A O VINVERSE Table 5: inverse According to Payne (ibid.), in Cree direct marking is hierarchically ordered in the following way and is thus grammatically fixed: second person > first person > third person proximate > third person obviative Table 6: hierarchy of direct/ inverse marking in Cree When the agent in a clause is higher on the hierarchy than the patient, the direct marking is used and when the agent is lower on the hierarchy the inverse marker is used. Thus, when a 1st person acts on a 2nd person the verb will always be marked in the inverse, while when a 2nd person acts on a 1st person the verb will be marked as direct. There are proximate and obviative third persons. The third person appearing first in a discourse is marked as a proximate while the other third persons are marked as obviatives. The proximate referent is thus the topic of the action while the obviative third persons are less focused on. According to Payne (ibid.), inversesystems often work with this difference between two third persons, called obviation. Since obviation has to do with the viewpoint of participants in discourse, Klaiman (1991) calls inverses ‘pragmatic voice’. According to Givón (1994: 17ff), there are analytical structures parallel to the inverse, which he calls word order inverses. However, these structures are not marked in the verb. The inversion is found rather in the variation of the constituent order dependent on the topicality of the arguments. Since this structure has no additional marking in the verb, I will not consider it as a voice structure in the present study. Apart from the inverses, there is another structure which operates on grounds of pragmatic salience: focus voice systems, which will be discussed in the following section. 3.1.6 Focus systems Focus voice systems are known from Mayan languages (Central America) and Philippine languages (MalayoPolynesian, Austronesian). In focus voice systems, a participant is focused depending on the relative pragmatic salience of this participant. The agent is usually the basic option for focus. Some languages always mark the focus of the agent morphologically on the verb, while other languages do not mark agent focus. When other participants such as patient or instrument are focussed, the verb is always marked by a focus marker, while the focussed participant is moved to clause initial position. Klaiman (1991: 33) defines focus systems in the following way: “Alternations of verbal marking reflect the relative importance or prominence of nominal referents in the information structure of the discourse”. How focus systems work can be seen in the following examples from Ixil (Mayan) by Ayres (1983: 42), cited from Klaiman (1991: 34): (18) Ak’oni in ta’n uula. 2SG.ERGshoot 1SG.ABS with sling ‘You shot me with a sling.’ (19) Uula ak’onib’e in. sling 2SG.ERGshootINSTR.FOCUS 1SG.ABS ‘With a sling you shot me.’ (Chajul dialect) Since there is no overt focus marking in example (18), the agent, being the basic option for focus, is understood as focused. In example (19) the instrument is focused: in (18) the instrument is in clause final position and stands with the preposition ta’n; in (19) it appears without this preposition in clause initial position and the verb is marked for instrument focus by the suffix b’e. According to Klaiman (1991: 34), the “ObliqueInstrumental nominal is the locus of information salience in the clause”. When other nonagent participants are focussed, another suffix is used. Ixil differs in some respect from Philippine languages. In Ixil, agent focus is unmarked, whereas in Philippine languages agent focus as well is marked in the verb. Furthermore, agent focus is as common as other focuses, such as patient focus. In Philippine languages, the agent focus is thus not basic as it is in Ixil in (18). The typical change in the focus as in the examples (18) and (19) above can be shown in the following table: agent focus: AERG V OABS OBL(instrument) instrument focus: INSTR AERG V-FOCUS(instr.) OABS Table 7: instrument focus Focus voice systems are often embedded in other grammatical structures in the language. According to Klaiman (1991: 35) “in Mayan and Philippine languages alike, information salience is usually assigned to nominals that are relativized or questioned”. 3.1.7 Causative Causative voice is the most common of the voice phenomena in the languages of the world, according to Nichols (1992). In her sample, 63% of all languages have a causative. As the name suggests, the causative is a structure that expresses causation. A causative structure is derived from an active clause by addition of a causer, which can be seen in the following example from French (from Haspelmath 1990: 47) in (21), with the corresponding active clause in (20) (constructed example): (20) Mahmoud mange la soupe. Mahmoud eat DET.F soup ‘Mahmoud eats the soup.’ (21) Marie fait manger la soupe à Mahmoud. Marie make eatINF DET.F soup to/by Mahmoud ‘Marie has Mahmoud eat the soup.’ In the active clause (20), Mahmoud is the agent and subject and soupe, ‘soup’ is the patient and object. In the causative derivation (21), Marie is added as the causer. The causative sentence consists of a main clause and an embedded clause. The main clause contains the subject Marie, the verb of causation fait, ‘make’ and the complement, which is the embedded clause. This embedded clause expresses the meaning of the active clause and is called ‘effect clause’ by Comrie (1989) since it expresses the effect of the causation. The effect clause contains the effect verb, the object, and the subject of the active Mahmoud, which is backgrounded and stands in oblique position, here as a prepositional phrase. The subject of the active is also called ‘causee’ in the causative construction. Comrie (1989: 165) describes the structure of a causative in the following way: a causative contains two “component situations, the cause and its effect (result)”. The structure of the causative in French can be shown in the following table: active: A1 V O causative: A2 ‘make’ (O V by A1) Table 8: analytic causative There are, however, some causative structures in which the causee is not backgrounded in the embedded clause but where it is in A position. English has such a causative structure, as is shown in the following example (from Haspelmath 1990: 47): (22) Marie makes Mahmoud read the book. In this example, the causer Marie and the causation verb are added to the active clause Mahmoud reads the book. In the structure of the causative in English, the embedded clause contains the original structure of the active clause except that the verb is in the infinitive as there can only be one finite verb in a grammatical sentence. The causative structure in (22), where the causee is not backgrounded, is typologically uncommon according to Haspelmath (1990). Usually, the causative involves the backgrounding of the causee. In this way, causative and passive resemble each other in that the agent of the active clause is usually backgrounded in the respective voice structures. The causative in French is only one of several different causative structures, namely an analytic causative. Analytic causatives mark the verb with an auxiliary to express causation. The causative verbal phrase in (21) is thus fait manger, ‘makes eat’ as opposed to the active mange, ‘eats’. Other causative structures are morphological and lexical causatives. Lexical causatives are pairs of suppletive verb forms, one of which expresses a causative meaning, e.g. die and kill, where kill is the causative form. Lexical causatives are usually not included in theories about causative. Palmer (1994), for example, excludes lexical causatives from his theory because these differ from actives only in meaning and not in structure. Morphological causatives express causativization through productive affixation. The verb is marked by an affix for causative voice and a causer is added to the clause. An example of a morphological causative is the following from Turkish (Comrie 1989: 176): (23) Müdür mektubu imzalad. director letterACC signPAST ‘The director signed the letter.’ (24) Dii mektubu müdüre imzalatt. dentist letterACC directorDAT signCAUSPAST ‘The dentist got the director to sign the letter.’ In example (24) the causer is dii, ‘dentist’ and the causative affix is t (-t is an allomorph of -d) . When looking at the structure of this morphological causative there is no embedded causative clause as in the French and English analytic causative above; rather the causation is expressed directly in the verb. The causative clause in (24) differs from the active clause in (23) in that there is a new participant, namely dii, ‘dentist’. This new participant is the transitive subject of the causative clause and the object mektubu, ‘letter’ of the causative clause is the same as the object in the active. The causee müdür, ‘director’ is demoted to a dative argument in the causative clause. Comparing morphological and analytic causatives, we can say that only morphological causatives are valence increasing since they add a new participant to the clause, while analytic causatives consist of two predicates with their respective arguments. In example (23) above, the causative was applied to a transitive clause, but causatives can also be applied to intransitive or ditransitive clauses. In all cases, a causer is added and stands in A position in the causative clause. In this way, the subject of the active clause, i.e. the causee, has to be removed from its original subject position. According to Comrie (1989: 176) the causee is either omitted or it is moved to the highest slot that is not occupied on the following hierarchy of demotion: subject > direct object > indirect object > oblique object Table 9: the case of the causee This hierarchy implies that when an intransitive clause is derived by a causative, the causee will become a direct object since this relation is not occupied in intransitive clauses. Thus, the active intransitive clause is transitivized. This structure is shown in the following table: active : S V causative: A O VCAUS Table 10: morphological causative from an intransitive active When a transitive clause is derived by causative, the causee will, according to Comrie (ibid.), be moved to the place of the indirect object, as is shown in the following table: active: A O V causative: A O IO VCAUS Table 11: morphological causative from a transitive active There are, however, exceptions to this hierarchy. Comrie mentions Sanskrit, which allows two accusative objects so that the causee in the derivation shown in Table 11 becomes an accusative object instead of an indirect object. I will now turn to the semantics of causatives. Apart from the place of the causee in the causative clause, the causee may have different case markings in some languages according to how much control the causee has over the situation. An inanimate causee usually has no control, while animate causees may have different degrees of control (discussed in 2.2). Comrie (1988: 174) has the following examples from Hungarian where the causee appears in the accusative case when it has no control and in the instrumental case when it retains some control over the situation: (25) Én köhögtettem a gyereket. I caused.to.cough the childACC ‘I made the child cough.’ (“e.g. by slapping him on the back and thereby inducing him to cough whether he wanted or not”) (26) Én köhögtettem a gyerekkel. I caused.to.cough the childINSTR ‘I made the child cough.’ (“e.g. by asking the child to do so”) In some languages, causatives can both express causing and permitting with the same form, in other languages the concepts are expressed by two separate forms. Concepts which semantically do not express causation, such as believe, say, want, and ask, may also be coded in the same way as causatives, i.e. they have the same structure as the causatives discussed in Table 8, 10 and 11 above. Causatives can furthermore be divided into direct and indirect causatives. These types differ in how closely the cause is associated with the effect. For example, if a young man pushes an old lady to make her fall the causation is direct, whereas the causation is indirect if the young man throws a banana skin on the pavement and in this way causes the old lady to fall. Some languages express direct and indirect causation differently. Furthermore, some languages may have a ‘passive’ causative. According to Comrie’s (1989) hierarchy above, the causee comes into indirect object position when the underlying active clause is transitive (Table 11) and into direct object position when the active clause is intransitive (Table 10). There are, however, languages where the causee always comes into oblique position, e.g. French, as can be seen in the following example from Comrie (1989: 179): (27) Jean fait manger les pommes par Paul. Jean make.PAST eat.INF DET.PL apple.PL by Paul ‘Jean made Paul eat the apples.’ In this example, the causee Paul is in oblique position marked with the preposition par, ‘by’ (the same as is used for agents in passives) and it is not the indirect object of the clause which Comrie’s hierarchy would predict. Comrie suggests a ‘passive analysis’ for this phenomenon. The embedded clause les pommes manger par Paul, ‘the apples eat by Paul’ can be analyzed as passive as the patient les pommes, ‘the apples’ is the subject of this structure and the agent Paul is coded as an oblique. Haspelmath (1990) disagrees with an analysis of this structure as a ‘passive’ causative, since there is no overt passive marking in the verb. On the other hand, according to Haspelmath (1990), passives develop from causatives in some languages and therefore these two structures may resemble each other in certain ways. I will come back to this when discussing the passive in Greenlandic. 3.1.8 Applicative Applicatives are found in many languages of the world. These structures add an O to an intransitive or to a transitive clause. Thus, applicatives (in many cases) increase the valence of a verb. The active clause is derived by an applicative in the following way: if the active clause is intransitive, the active intransitive subject becomes a transitive subject and an object is added; at the same time the verb is marked by an applicative affix. Payne (1997: 187) shows this in the following example from Yagua (PebaYanguan, Amazon): (28) Saduu ráviimú. 3SGblow INANinto. ‘He blows into it.’ (valence = 1) (29) Saduutára. 3SGblowAPPLINAN.O ‘He blows it.’ (valence = 2) In example (28) the clause is intransitive and the inanimate pronoun rá, ‘it’ is an oblique element, marked by the suffix viimú, ‘into’. In (29), on the other hand, the pronoun is added as the object of the clause which can be seen in the crossreference suffix ra in the verb. Apart from that, the verb is marked by the applicative suffix tá. In the clause with the applicative (29), the valence is increased in relation to the active clause in (28). The derivation by an applicative is schematically shown in the following table: active: S V (OBL) applicative: A O VAPPL Table 12: applicative from an intransitive active When an applicative is applied to an intransitive active clause, an object is added. A transitive active clause already contains an object. When such a clause is derived by an applicative, this object is either demoted to an oblique element, removed from the clause or it becomes an indirect object and thereby remains a core argument in the clause. In the latter case, one can speak of applicatives as valence increasing structures, while in the cases of object demotion or deletion the valence of the applicative clause is the same as the valence of the corresponding active clause. In these cases the grammatical roles are changed but not the valence: according to Payne (1997: 1867) “the applicative simply ascribes a new, formerly peripheral, semantic role to the direct object”. In the following example from German (Bussmann 1990: 91), the prefix be can be analyzed as an applicative marking element. The object of the active clause Buch, ‘book’ becomes an oblique element in the clause with the applicative (31) while the indirect object of (30) becomes the direct object in (31): (30) Der Vater schenkte dem Sohn das Buch. DET.M.NOM father givePAST DET.DAT son DET.N.ACC book ‘The father gave the son the book.’ (31) Der Vater beschenkte den Sohn mit einem Buch. DET.M.NOM father APPLgivePAST DET.M.ACC son with DET.ID.FDAT book ‘The father gave the son the book.’ The structure of an applicative with a corresponding transitive active clause can be seen in the following table: active: A O (IO) V (OBL) applicative: A O IO/ OBL/ Ø VAPPL Table 13: applicative from a transitive active I have found only morphological applicatives in the literature on this structure. However, it should be logically possible to find analytic or lexical applicatives, in the same way as there are lexical and analytic causatives. A language can have more than one form of applicative. Payne (1997: 188ff.) gives as an example the language Nomatsiguenga (PreAndine Maipuran Arawakan from the Eastern Peruvian foothills), which has “at least 9 applicative suffixes that express a variety of semantic roles.” A structure closely related to applicatives is ‘dative shift’, which is shown in the following example from German: (32) Der Mann schrieb einen Brief an das Mädchen. DET.M.NOM man write.IMPF DET.ID-M.ACC letter to DET.N.ACC girl ‘The man wrote a letter to the girl.’ (33) Der Mann schrieb dem Mädchen einen Brief. DET.M.NOM man write.IMPF DET.DEF.N.DAT girl DET.ID-M.ACC letter ‘The man wrote the girl a letter.’ The oblique an das Mädchen, ‘to the girl’ in (32) becomes an indirect object in the dative case in (33) through dative shift. Since an oblique element becomes an indirect object, the valence is increased. The oblique elements that become core elements through dative shift are usually benefactives and recipients while applicatives usually advance instruments or other obliques. Furthermore, dative shift constructions and applicatives differ in that there is no marking in the verb for dative shift, whereas applicatives are marked. 3.1.9 Other structures There are a number of other structures which are sometimes described under the heading of voice, especially in theories which define voice in terms of valence (see 3.2 below). These structures are reflexives, reciprocals, object omission, subject omission, object demotion, object incorporation, and possessor raising. For the sake of completeness, I briefly want to describe these structures. In reflexives, one participant occupies both the subject and object relation, i.e. agent and patient or agent and recipient roles in a clause. A reflexive clause is syntactically transitive but from the view of the number of participants it can be described as intransitive. In Greenlandic (to be discussed in chapter 5) reflexive clauses are inflected intransitively. However, in other languages reflexive clauses are formally transitive. In reciprocals, on the other hand, two participants act upon each other, in this way both participants are agents and patients. The following example from Payne (1997: 220) shows object omission in English: (34) The hunter shot the deer. (35) The hunter shot. The object of the transitive clause (34) is omitted and (35) is therefore an intransitive clause. Object omission is sometimes treated under the heading ‘antipassive’, as by Heath (1976: 203), according to Dixon (1994). Dixon (ibid.), on the other hand, does not treat object omission as an antipassive since there is no antipassive marking in the verb. In the same way as with object omission, some languages may omit a subject without any verbal marking. In this case, the O of the transitive clause becomes an S in the clause where subject omission is applied. As there is no passive marking in the verb of a clause with subject omission, this structure cannot be described as a passive. Object demotion is close to object omission since in both structures an object is moved away from core status. In object demotion, however, the object becomes an oblique element instead of being omitted. The following example from Payne (1997: 220) shows the equivalent of clause (34) above with a demoted object: (36) The hunter shot at the deer. In object incorporation, an object is incorporated into the verb. This can be seen in the verb babysit, where the object baby is incorporated into the verb sit. A former transitive clause becomes intransitive in this way, as the object is expressed overtly in the verb. The last structure I want to describe in this section is possessor raising. This structure is also sometimes called external possession. Possessor raising is valence increasing. A dative element is added to the clause as the possessor of the direct object. This can be seen in the following example from colloquial German where the external possessor is highlighted by the italics: (37) Ich suche dem Hans sein Haus. 1SG.NOM search.for DET.M.DAT Hans his house ‘I search for Hans’ house.’ Another example of possessor raising is the English clause He bites him the leg. Not all of the voice structures discussed in this chapter are always treated under the heading of voice. Furthermore, there may be theories of voice which include structures that are not discussed here. In any case, those relevant for a discussion of the Greenlandic passive have been mentioned here. In the following section, I will look at how several approaches to voice define this category and how these theories differ in their definition of voice. 3.2 Different definitions of voice In the preceding section, I have described several structures that are treated under the heading of voice in different theories. In this section, I want to discuss some of these theories. I will describe how approaches to voice differ in their definition of this category, i.e. which of the structures discussed in section 3.1 these theories include. In preceding decades, voice was often defined in a very restricted way. Thus, Barber (1975: 16), as cited in Klaiman (1991: 260), defined voice as “a strategy to move NPs in and out of subject position”. This definition covers a narrow structure in a restricted set of languages. The structure is narrow because this definition of voice does not include structures where an NP is moved in or out of object position, such as antipassives. When comparing passive and antipassive, their structures seem to be parallel with the exception that passive affects subjects while antipassive affects objects. It would thus be natural to include these two structures under the heading of ‘voice’. Furthermore, Barber’s definition is restricted to languages which have the grammatical notion of subject. Not all languages work in terms of subjects, such as the languages of the Philippines although they do have voice structures, such as passive or focus voice. In this way, Barber’s definition is very restricted. Other theories of voice take the other structures mentioned above into account as well. Nevertheless, recent theories also differ a lot in their definition. I want to discuss the approaches to voice by Foley & Van Valin (1984), Klaiman (1991), Payne (1997) and Nichols (1992) in the present section. These approaches will give a broad picture of how voice can be defined. Foley & Van Valin’s (1984) definition of voice is broader than that of Barber for two main reasons. First, they account for languages which do not make use of the concept of subject and second, they include antipassives in their definition of voice. Foley & Van Valin do not define voice in terms of subject and object, but in terms of actor and undergoer, which are concepts close to the semantic roles agent and patient. The typical actor is, according to Foley & Van Valin (1984), an agent and the typical undergoer is a patient. Furthermore, other roles which typically appear as the object of an active clause and as the subject of a passive clause are contained under the heading of ‘undergoer’. In the same manner, semantic roles that can be in subject position in an active clause are treated under the heading of ‘actor’. In this way, Foley & Van Valin can generalize over languages without the notion of subject marking, as they show in their analysis of the Philippine languages. Since Foley & Van Valin’s grammar is a functional grammar, they deal more with voice according to its discourse function than with the structural realization of this category. They see passive and antipassive primarily as ways of reorganizing the structure of a clause for discourse reasons by backgrounding and foregrounding certain constituents. Foley & Van Valin operate with socalled pivots, which can roughly be understood as subjects (to be discussed in 3.3) but which, nevertheless, form a different category. Often languages have pivot restrictions in structures such as deletion of a coreferent element in clause combination. In a language such as English, this means that there has to be coreference between the subjects of two combined clauses before the coreferent element can be deleted. This is shown in the following examples: (38) John hit Rose and Rose cried. (39) *John hit Rose and cried. (under the interpretation that Rose cried) (40) John hit Rose and cried. (under the interpretation that John cried) In (38), the two clauses John hit Rose and Rose cried are combined in coordination. The object Rose in the first clause is the same as the subject in the second clause, i.e. these two elements are coreferent (with the precondition that ‘Rose’ in the first clause and ‘Rose’ in the second clause refer to the same person). In (39), the coreferent element Rose is deleted in the second clause. However, the meaning of the clause is not that Rose cried, but that John cried (40). This is because there are pivot restrictions on deletion in clause combining in English so that only the second of two coreferent subjects may be deleted. When an O is coreferent with an S in adjacent clauses in English, the passive can bring the O into pivot position and the pivot restrictions are fulfilled. Languages with pivot restrictions, such as English, usually have a passive that can be used for this purpose. Coming back to example (38), Rose can be deleted in the second clause only when the passive is applied in the first clause to bring the O into S position: (41) Rose was hit by John and cried. In this example, the object of (38) is in subject position and the pivot constraints are fulfilled. In this way, Rose can be deleted from the second clause. Thus, Foley & Van Valin (1984: 149) define passive as “the occurrence of a nonactor argument, normally the undergoer, as the pivot of the clause, and the peripheral status of the actor.” According to Foley & Van Valin, the function of voice is foregrounding or backgrounding of certain constituents which then fulfil pivot conditions. In their view, discourse pragmatics has its roots in grammar. Foley & Van Valin’s (ibid.) definition of voice includes passive and antipassive as the most important voice structures. Others are the agentbackgrounding middle voice (‘passive middles’ and ‘spontaneous events’, discussed in 3.1.4) and focus systems which foreground and background elements to fulfil pivot conditions. However, they do not include causatives and applicatives in their definition of voice and this may have to do with the fact that causatives and applicatives are not used in the fulfillment of pivot conditions. Klaiman’s (1991) definition of voice accounts for the structure of voice, rather than for discourse pragmatics. Her definition is therefore very different from that of Foley & Van Valin (1984), although it deals with roughly the same structures, with the exception that she also includes inverses. According to Klaiman (1991: 261), voice is “a category encoding alternations in the configurations of nominal statuses with which verbs are in particular relationships”. Klaiman defines voice in terms of ‘nominal statuses with which the verbs are in particular relationships’ instead of subject and object. In this way, she accounts for voice structures involving elements other than subject and object. Klaiman divides voice into three types: derived voice, basic voice, and pragmatic voice. The derived voice has to do with valence change, in which case the semantic roles of the active are assigned to different grammatical relations in the derived voice. Thus, a syntactic ‘derivation’ takes place between the active and the derived voice structure. Klaiman includes the passive and antipassive under the heading of derived voice. She calls the middle voice ‘basic voice’ (discussed in 3.1.4) since there is no change in valence or in the ascription of semantic roles to grammatical relations. Her definition of middle voice contains mainly reflexive middles. Under the heading of pragmatic voice, Klaiman treats inverse and focus voice systems. Klaiman does not include causatives or applicatives in her theory of voice. However, these structures can be described as parts of Klaiman’s definition of voice, since causatives and applicatives also involve ‘alternations in the configurations of nominal statuses with which the verbs are in particular relationships’. Thus causatives and applicatives could be described under the heading of Klaiman’s derived voice which in this way would include not only valence decreasing structures, such as passive and antipassive, but also valence increasing structures, such as causative and applicative. The reason why Klaiman does not treat causatives and applicatives as voice structures may be that causative and applicative clauses, having an extra participant, do not describe exactly the same events as their corresponding active clauses. Passive or antipassive clauses, however, describe the same events as the corresponding active clauses, only with a different perspective. Even if all of these four voice structures change the situation in the clause, the situation change is somewhat bigger by the causative and applicative. One definition of voice containing causatives and applicatives is the one by Payne (1997: 169), who treats voice phenomena in terms of valence even if valence is not part of his definition of voice: “operations that adjust the relationship between semantic roles and grammatical relations in clauses”. Payne (ibid.) divides voice structures into valence increasing structures and valence decreasing structures. In valence increasing operations, a peripheral participant is usually upgraded to a central NP. Valence increasing structures discussed by Payne are causatives, applicatives, and possessor raising. In valence decreasing structures, a central participant is demoted to peripheral status or omitted. Payne includes passive, antipassives, reflexives, reciprocals, middles, inverses, subject omission, object omission, object demotion, and object incorporation under the heading of valence decreasing structures. I have a number of comments on Payne’s division of voice phenomena and on the structures he treats under the heading of voice. Payne defines middles as valence decreasing devices, while Klaiman (1991) treats middles as ‘basic’ structures which do not change the valence. The reason for this discrepancy between the two approaches to middle voice is that Klaiman concentrates on reflexive middles, while Payne defines middles in terms of spontaneous events (I discussed the distinction in 3.1.4). They are dealing with different phenomena under the same name. According to Payne (1997: 20910), inverses are valence rearranging. However, Payne (1997: 210) treats inverses under the heading of valence decreasing structures, arguing that: “the semantic effect of a prototypical inverse construction is to downplay the centrality of the agentive participant with respect to the more patientlike participant. Therefore, inverses operate in the functional domain of valence reduction”. In my opinion, the backgrounding of an agentive participant cannot be understood as a reduction in valence if this participant still has core argument status in the clause. I think that the treatment of inverses under the heading of valence decreasing structures is a compromise, since Payne wants to describe all voice structures in terms of valence. Unfortunately, this is not usually possible for inverses. As a result, his treatment of voice structures is too broad and it appears idiosyncratic. Furthermore, Payne includes a number of structures which are not usually treated under the heading of voice. These structures are reflexives, reciprocals, subject, and object omission, object demotion, object incorporation, and possessor raising. Payne (1997: 169) argues for the inclusion of these structures “because of their functional similarity and because many languages treat them in structurally comparable ways”. The similarities between these structures and other structures usually treated under the heading of voice often have to do with the diachronic development of voice structures, to be discussed for the passive in 4.4. Most of the approaches of voice do not include subject and object omission, object demotion, object incorporation and possessor raising because these structures usually do not involve any obligatory voice marking on the verb or in the verb phrase. Reflexives and reciprocals, on the other hand, are often marked morphologically in the verb or analytically in the verb phrase, but they usually do not encode ‘alternations in the configurations of nominal statuses with which the verbs are in particular relationships’, which is part of the definition of voice by Klaiman (1991: 261). Reflexives and reciprocals are rather a special situation in the active voice where the same participant occupies the place of the subject and the object, i.e. usually the roles of agent and patient. This does not involve any alternations in the nominal statuses. Actually, Payne’s own definition of voice would not include reflexives and reciprocals as voice structures since these structures do not ‘adjust the relationship between semantic roles and grammatical relations’. The final definition of voice I want to discuss in this section is that by Nichols (1992). She also divides voice in terms of valence, but includes mainly passive, antipassive, causative and applicative as voice structures. Nichols’ aim is not to give a definition of voice but rather to investigate the relationship between grammatical categories and structural types in the languages of the world. In the following section, I will write about her findings and therefore I treat her definition of voice in the present section. Nichols includes only morphologically marked structures, whereas most other approaches to voice include both morphological and analytic marking of voice. Furthermore, Nichols defines voice in terms of valence increasing and valence decreasing structures which either affect the subject or the object of a given clause. Voice structures affecting the A, such as passive and causative, are called Aaffecting processes and structures affecting the O, such as antipassives and applicatives, are called Oaffecting structures by Nichols. Taking valence reduction or increase into account, Nichols (1992: 154) calls passive –A, since it decreases valence by demoting the A of a clause. In the same manner, the antipassive is described as –O, the causative is +A, and the applicative is +O. All the theories discussed in the present section have in common that they define voice as a strategy to alternate the structure of the arguments in a clause. Nevertheless, the theories differ very much in their treatment of varying structures under the heading of voice, which can be seen in the following summarizing table: Foley/ Van Valin Klaiman Payne Nichols Active unmarked unmarked unmarked unmarked Passive voice derived valence decr. -A Antipassive voice derived valence decr. -O Middle voice basic valence decr. Reflexive valence decr. Reciprocal valence decr. S/A omission valence decr. O omission valence decr. O demotion valence decr. O incorporation valence decr. Inverse pragmatic valence decr. Focus-systems voice pragmatic Possessor raising valence incr. Causative valence incr. +A Applicative valence incr. +O Table 14: structures treated under the heading of voice in different theories All approaches shown in Table 14 have in common that they include active, passive, and antipassive under the heading of voice. In addition, middles are commonly defined as voice structures although Nichols (1992) does not include them. Furthermore, voice is usually treated as a grammatical category. This means that voice structures form a category which is formally marked as opposed to other structures, which differ solely in lexical terms. As a grammatical category, voice can be applied under certain pragmatic conditions, which means that the application of voice is largely the choice of the speakers. According to Payne (1997: 213), grammatically fixed structures are sometimes treated under the heading of voice, but he considers that a mistake. Such a structure is the inverse in Cree (Algonquian), discussed in 3.1.5. In this language, the application of the inverse in the first and second person is grammatically fixed according to a certain hierarchy. Thus, a 2nd person acting on a 1st person is always expressed as direct, while when a 1st person acts on a 2nd person the inverse is applied. Nevertheless, pragmatic factors are decisive in the marking of the third person as proximate or obviative. In this way, Payne argues that the inverse in Cree is a voice structure only due to its use in the 3rd person. The functions of voice are treated with different importance by the theories referred to in the present section. Foley & Van Valin (1984) define voice primarily from the point of its function: the foregrounding and backgrounding of participants. Klaiman (1991), Payne (1997), and Nichols (1992), on the other hand, mainly look at the structure of voice and do not describe any central function of voice. 3.3 Voice and alignment Alignment covers the different ways in which grammatical relations are coded. There are many different kinds of alignment. The two most common kinds of alignment are ergative and accusative alignment. As these are important in the following discussion, I will describe these two structures in detail. Ergative alignment is found in about one fourth of the languages of the world, according to Dixon (1994). Languages can be morphologically or syntactically ergative. Usually the term ergativity refers to morphological ergativity as syntactic ergativity is very rare. Structures are called morphologically ergative when they treat S and O in the same way, as opposed to A. One way to combine S and O is case marking: S and O are marked in the socalled absolutive case while A is in the ergative case. In addition, other clauseinternal structures, such as verbal agreement and word order, can show an ergative pattern. In accusative alignment, on the other hand, S and A are marked in the same way as opposed to O. The following table shows case marking in an accusative language as opposed to case marking in a morphologically ergative language, in intransitive and transitive clauses respectively: _ accusative language ergative language _ Snominative Sabsolutive Anominative Oaccusative Aergative Oabsolutive Table 15: accusative and ergative alignment The nominative and absolutive cases, apart from being the most frequent in their respective alignment type, have often zero case marking. These cases are therefore the unmarked cases in these structures (examples with ergative case marking and verbal morphology can be found in Dixon 1994). Apart from ergative and accusative structures within a clause as discussed here, there are structures which work between two or more adjacent clauses. This has to do with ‘pivots’ as discussed above. Structures working between two or more adjacent clauses can be syntactically ergative or syntactically accusative (Dixon 1994). Usually syntactic ergativity involves coreference constraints when complex sentences are formed by coordination or subordination. These coordination constraints demand that Ss and As in two adjacent clauses are coreferent in languages with accusative syntax, whereas S and O have to be coreferent in languages with ergative syntax. The arguments that have to be coreferent are called pivots. In languages with accusative syntax, the pivot involves subjects (see also 3.2 above) and is called S/A pivot, whereas languages with ergative syntax have Ss and Os as their pivot, also called S/O pivot. Languages with syntactic ergativity usually also have morphologically ergative structures. Ergative alignment and the passive have many common traits: in an ergative language, the object is unmarked and since the patient usually is in object position it resembles the passive structure where the patient also is unmarked. The agent, on the other hand, is in A position in an ergative language and thus marked by the ergative case while also the agent of a passive is in a marked case. In the passive, however, the agent is marked as an oblique and may be omitted, whereas the agent in an ergative construction is by definition a core case. Furthermore, languages with ergative structures do not automatically have a special marking on their verbs as opposed to passives. Apart from these differences, the similarity between passive and ergative structures is not an incident but can be seen in the light of the fact that at least some ergative structure have developed from a passive (see 4.5). Not only the diachronical development connects voice and alignment but certain types of alignment may prefer certain kinds of voice systems. Often passive is said to be connected with accusative languages and antipassive with ergative languages. Thus, Dixon (1994: 147) writes: “passives are typically found in accusative and antipassives in ergative languages.” In addition, Foley & Van Valin’s (1984) definition of passive and antipassive reflect that passives have to do with accusative languages and antipassives with ergative languages. According to Foley & Van Valin, passive and antipassive bring a secondary participant into pivot position: the passive brings Os, which are secondary participants in accusative languages, into S, i.e. pivot position. The antipassive brings As, which are secondary participants in ergative languages, into S position. In this way, Foley & Van Valin treat passives with accusative alignment and antipassives with ergative alignment. Nichols (1992) investigated the connections between several different grammatical phenomena, such as voice, and structural types, such as alignment. She includes only morphologically marked structures (as discussed in 3.2). Her findings are highly relevant to the present discussion. Nichols found that the frequency of Oaffecting voice structures compared with Aaffecting voice structures is very low: the causative is the most frequent voice structure which is found in 110 of the 174 languages of her sample. Passive is the next most frequent structure and is found in 79 out of the 174 languages. Antipassive and applicative voice are not very common: 39 out of the 174 languages have an applicative and 20 out of the 174 languages have an antipassive in Nichols’ sample. Many languages have more than one voice structure; therefore, the total number of structures (248) is higher than the total number of languages (174). Altogether, there are three times more Aaffecting structures in the languages of the world than there are Oaffecting structures. At the same time, a fourth of the languages of the world are ergative while the majority of languages are accusative in their structure. Thus, the frequency of alignment types and voice structures may have some connection. Nichols (1992) also investigated if voice structures preferably appear with certain alignment types. She found that there is no real connection between Aaffecting structures and accusativity. On the other hand, there is a connection between Oaffecting structures and languages having a certain status for their Os such as ergative languages. The languages in Nichols’ sample which had Oaffecting structures as their only voice phenomenon were all ergative, which shows that antipassive is indeed typically connected with ergative structures. In addition, some languages with accusative alignment and Oaffecting structures appeared in Nichols’ sample. However, these languages all have crossreference to both subject and object in their verbs. Nichols (1992: 158) summarizes her findings in the following way: Antipassives are associated with languages in which the O is central to clause morphosyntax: ergative languages, where the O takes the unmarked case, and those accusative languages in which there is agreement with the direct object. (…) It is thus observationally true, as traditionally maintained, that the antipassive is associated with ergativity; but the causal factor is evidently not the ergative alignment per se but the special structural status of the O which is basic to ergativity. Aaffecting structures, on the other hand, are equally common in both ergative and accusative languages. According to Nichols (1992: 159) “Aaffecting operations are the unmarked type, common and apparently unrestricted; O operations are less common and more restricted”. Aaffecting structures are thus both unmarked with regard to which type of alignment they appear with and they are by far the most frequent structures. Nichols (1992: 159) suggests that Aaffecting structures are unmarked because most of the languages in the world are syntactically accusative. Because of this, she concludes that voice primarily affects and depends on the syntax of a language. Before I will to proceed to the concluding section of the present chapter, I want to present a broader view on the structures discussed in the present and preceding sections. Above I discussed pivot constraints in clause combining which have to do with syntactically accusative and ergative structures. According to Dixon (1994) there are languages which use structures other than pivot constraints in clause combining. There are two such structures: on the one hand switch reference, and on the other hand no syntactic mechanisms, which means that clauses can be combined freely, as long as it is semantically acceptable. Switch reference is a category that marks the verbs of two adjacent clauses for coreference of their participants. In this way, languages with switch reference do not need pivot constraints in clause combining since all clauses may be combined and at the same time marked for coreference. Since languages with switch reference do not involve pivot constraints, passive and antipassive do not have the function of fulfilling pivot conditions in these languages. According to Dixon (1994: 154): Languages with switch reference typically lack passive and antipassive or  if they do include these  they do not have any pivotfeeding function (…). It can be argued that there is no syntactic need of passive or antipassive, since it is not necessary to bring an underlying O or A NP into derived S function, to facilitate NP omission without any chance of ambiguity. See also 7.7 below. In 7.7, I will come back to this, as Greenlandic has both a passive and a form of switch reference, and is therefore exceptional. 3.4 Summary In the present chapter, I have described how different theories define voice and which voice structures these theories include. Furthermore, I have discussed the relationship between voice structures and alignment. In the present section, I want to give my definition of voice in reference to the theories discussed in this chapter. In my opinion, the theories of voice discussed in the present chapter are either too broad or too narrow in their definition of this category and in their treatment of different structures under the heading of voice. Therefore, I want to give my own definition of voice for the remainder of the present study. My definition is more practical than principal, as it is inspired by a Greenlandic perspective. The theories above have in common that they define voice as an alternation in the argument structure of a given clause. Furthermore, most theories include formal marking of the verb, i.e. the verb of an active clause is morphosyntactically less complex than the verb of another voice structure. I will combine these two parts in my approach to voice. In this way, I define voice as a category which changes the structure of the arguments in a clause by at least formal marking in the verb. In my definition, I exclude a number of structures which Payne (1997) treats under the heading of voice, namely subject, and objectomission, object demotion, object incorporation, and possessor raising. These structures are excluded in my definition since they are not formally marked in the verb. On the other hand, my definition includes more structures than that of Foley & Van Valin (1984) who define voice in terms of a foregrounding and backgrounding function. I do not define voice in terms of its pragmatic function. Nevertheless, the structures treated under the heading of voice in my definition have a common function: they change the situations in the clause in different ways. At the same time, the verb is marked for the change in the situation. In this way, the semantic valence of the verb is constant while the syntactic valence may or may not be changed (see 2.3). Furthermore, my definition of voice includes voice structures which Klaiman (1991) does not include, although my definition is very close to that of Klaiman. Although she does not mention them, Klaiman’s definition of voice might also include causatives and applicatives as I have shown in section 3.2. I include causatives and applicatives in my definition of voice since these structures are formally marked and change the structure of the arguments in a clause. Thus, the structures included in my definition of voice are passives, antipassives, middles, inverses, focus systems, causatives, and applicatives. 4. Passive 4.0 Introduction Having described voice in general, I will now turn to the voice structure this study is primarily concerned with, namely the passive. There are many approaches to the passive and these result in very different definitions of this structure. One such definition of the passive is that by Dixon (1994). His view on voice, and especially passive and antipassive, is roughly the same as Foley & Van Valin’s (1984, summarized in 3.2 above) since Dixon as well sees the fulfilling of pivot conditions as one of the main functions of the passive. According to Dixon (1994: 146), the passive formally: (a) applies to an underlying transitive clause and forms a derived intransitive. (b) the underlying O NP becomes S of the passive (c) the underlying A NP goes into a peripheral function, being marked by a noncore case, preposition, etc.; this NP can be omitted, although there is always the option of including it. (d) there is some explicit formal marking of a passive construction (generally by a verbal affix or else by a periphrastic element in the verb phrase  such as English be ... en  although it could be marked elsewhere in the clause). Dixon’s definition is typical for many recent approaches to passive. It describes a kind of ‘basic passive’ structure that exists in many languages which have a passive. There are, however, other structures which do not fit in Dixon’s definition but which are clearly very similar to passive structures. It could thus be argued that the definition of passive should be broader to include other structures. I will shortly present some of the structures that are excluded by Dixon’s definition. These structures will be treated more exhaustively in the following sections. Dixon’s definition of passive consists of four parts. His first claim (a) is about transitivity, since he claims that the active clause is transitive and the corresponding passive clause is detransitivized and thus intransitive. In many passive structures, this kind of detransitivization takes place indeed, but the socalled ‘impersonal passives’ (to be discussed in 4.1.5) can be applied to intransitive active clauses. Such an impersonal passive is German es wird gelaufen, literally ‘there is running’, which is derived from the intransitive verb laufen, ‘run’. In (b), Dixon claims that the object in the active clause is foregrounded to the subject of the passive construction. In some languages, however, oblique constituents can become the subjects of a passive construction as well (discussed in 4.1.3). Dixon claims in (c) that the A of the active clause is backgrounded, which means that it becomes an oblique constituent or that it is simply omitted in the passive clause. Dixon claims that even if the A can be omitted there is always an option to include it. There are, however, languages with socalled ‘agentless passives’, i.e. passive structures which cannot express the agent of a passive construction (to be discussed in 4.1.4). I agree with Dixon’s last claim (d) that the passive construction is always explicitly marked. Nevertheless, even this claim is disputed since some scholars include structures that have no formal marking for voice as a part of their definition of passive and voice. As may be clear from the discussion above, many different structures can be described under the heading of passive. Approaches to passive differ in the way the passive is described and in which of these structures they include. In the present chapter, I want to define passive by looking at this structure in different ways. First, I will describe the formal side of the passive. Then I will look at some structures connected with the passive which will lead me to a scenario for the diachronical development of this structure. After that, I will describe the functions of the passive according to different theories. At the end of the present chapter, I will formulate a definition of passive that I want to use in the remainder of this study. I will come back to the different structures of the passive described in the present chapter when describing the Greenlandic passive (in chapter 7). 4.1 The structure of the passive In the present section, I want to give a definition of the formal structure of the passive according to different approaches. My concern will be the markedness of the passive, the different ways of marking passive, and the nature of the promoted element and the demoted element. I will furthermore describe different kinds of passives, such as impersonal passives and stative versus dynamic passives. 4.1.1 The markedness of the passive Passive verbs do not appear in the majority of sentences in discourse. Rather, the use of passive is very restricted. According to Dixon (1994: 152) passives “are only used when specific syntactic and/or semantic and/or pragmatic and/or discourse conditions are satisfied.” The active voice, on the other hand, appears in most of the sentences in a discourse. It seems thus that the active can be described as unmarked and the passive as marked, following Comrie (1988) and Greenberg (1966) (see 2.3). In order to show that the passive is marked and the active unmarked, I want to look at Comrie’s four criteria for markedness (raw frequency, formal complexity, degree of productivity and discourse distribution) and see how they fit with the passive and the active. Active constructions are more frequent than passives. Apart from that, passive structures have some further morphological or analytic marking than active structures; passive constructions are thus formally more complex and longer than active structures. Finally, actives are crosslinguistically more productive than passives. In many languages, verbs can occur as actives but not as passives and actives can occur in more construction types than passives. Thus, passives are marked and actives are unmarked in terms of all Comrie’s (1988) four criteria. The verb in the passive construction is usually defined as having morphological or analytic formal marking. In this way, the passive verb differs from the active verb. Some theories, however, include passive structures in which the verb is not marked differently. These passive structures are usually referred to as lexical passives. Lexical passives are structures in which the verb takes a patient as its subject and where an agent is understood as carrying out the action of the verb. According to Payne (1997: 205), lexical passives are crosslinguistically very rare. He gives the following example from Yagua (PebaYanguan, Amazon): (42) Sabááryimáá. 3SGbe.killed.in battlePERF ‘He was killed in battle.’ Payne does not give the corresponding active verb for kill in Yagua which would have shown that the verb in (42) is a suppletive variant of the active verb, rather than being regularly marked for passive. Other structures are sometimes mentioned as forms of passives: passive structures without overt passive marking. Thus Siewierska (1984: 36, 40) argues that Palauan (Austronesian), along with a number of other languages, has a passive structure without passive marking (Siewierska’s glossing): (43) A sechelik a silsebii a blai. my friend stemPASTburn3SG house ‘My friend burned down the house.’ (44) A blai a lesilsebii. house 3SG(hypothetical)stemPASTburn3SG ‘The house was burned down.’ Haspelmath (1990: 26ff), on the other hand, refers to studies where passive structures without any formal marking are uncovered not to be passive structures after all. Thus, the Palauan ‘passive’ is analyzed as a special topic construction by De Wolf (1988: 167 ff.) cited in Haspelmath (1990: 27). Haspelmath argues that all verbs in passives must be marked either morphologically or analytically. Payne (1997), on the other hand, includes also lexical passives in his definition of passive. 4.1.2 Different forms of passive marking In the preceding section, I have mentioned three different kinds of formal marking of the passive, namely morphological, analytic and lexical marking. In the present section, I want to look at the morphological and analytic marking of the passive. Some scholars distinguish only between morphological and analytic passives, but actually, the marking can be further divided into smaller units. Haspelmath (1990), who studied the passive crosslinguistically, divides the different kinds of passive marking into the following seven structures and lists these according to their frequency: 1. additional stem affix, 2. auxiliary and participle, 3. particle, 4. extrainflectional affix, 5. differential subject person markers, 6. alternate stem affix and 7. sound replacement. I will briefly discuss these different kinds of passive marking: The passive marking by an additional stem affix is the most frequent type of morphological marking. The affix that expresses passive is in this case added directly to the verbal stem before any tense, aspect, and person marking. The passive marking by an auxiliary verb in combination with a passive or perfect participle form of the verb is the most common type of analytic passive, but it is not as common as morphological marking by an additional stem affix. IndoEuropean languages commonly have this type of passive marking while the structure is rare in other language families. Often the auxiliary is an equivalent of the verb be or become, but there are various other uses in different languages: go in Hindi (IndoEuropean), receive and get in English, suffer and undergo in the languages from South East Asia, come in Italian (IndoEuropean) and eat in Sinhalese (IndoEuropean). Even if this kind of passive is called analytic passive, it typically involves morphological marking as well, since the verb is morphologically marked as a participle. However, I will treat passive marking by an auxiliary and a participle as an analytic passive in the remainder of this study. These are by far the two most common markings. The following types of marking are only used by few languages to mark the passive: In some languages, the passive is marked analytically by a particle which stands in combination with the verb. This type of passive is found in Margi (an AfroAsiatic language from Nigeria), shown in the following example from Haspelmath (1990: 29): (45) ábá kr. understand PASS ‘Be understood.’ Another kind of morphological marking of the passive is carried out by an extrainflectional affix, which means that the passive affix is added to the inflected verb after the inflection. The passive marking follows all other inflectional affixes, such as tense, aspect, and person marking. This type of passive marking can be found in Danish, which Haspelmath (1990: 29) shows by the following example: elske, ‘to love’, elskes ‘to be loved’, elskede, ‘loved (past)’ and elskedes, ‘was loved’. Passive can also be marked morphologically by differential subject person markers, i.e. different crossreference forms in the verb. This type of voice marking is found in Latin (IndoEuropean) and Classical Greek (IndoEuropean) where there are different inflectional paradigms for active and passive voice. The passive can furthermore be marked morphologically by an alternate stem affix, which means that a morpheme in the active form is replaced by another morpheme in the passive form. According to Haspelmath (1990: 31), the transitive suffix –i in Kefa (a language from Papua New Guinea) is replaced by the passive suffix –e in the passive clause: cucci, ‘tie’ and cucce, ‘be tied’. The last type of morphological passive marking which Haspelmath lists is sound replacement. In this type of passive marking, sounds in the active verb are regularly replaced by other sounds in the passive. Sinhalese (IndoEuropean, spoken in Sri Lanka) has this type of passive marking. In Sinhalese, passives are formed regularly from actives by the fronting of the vowels. Haspelmath (1990: 31) shows this fronting of vowels in the following example: bala, ‘see’ and bäle, ‘be seen’. Keenan (1985) observed that passives are never expressed by reduplication or tone from which he deducts that these two structures never would be used for passive marking. However, according to Haspelmath (1990: 31), passives formed by reduplication or tone are theoretically possible. First, there are languages in which passive participles are formed by reduplication. When a passive structure is formed from these participles, they may also use reduplication. Second, passives formed from tones are possible, since passive affixes may develop into tones which is known from affixes in other structures. In the preceding section about the marking of the passive, I claimed that passives are usually more complex in marking than their active counterparts. Here, I want to elaborate on this claim. In the types of passive marking above, the formal types listed under 5., 6., and 7. mark active and passive in an equally complex way. Thus the passive form is not necessarily more complex in all structures than the active one. On the other hand, these three structures are not very frequent; thus, the vast majority of the types of passive markings are actually more complex than their active counterparts. As I already indicated, the seven types of passive marking listed above are of very different frequency. Haspelmath (1990) investigates the frequency of the different types of passive marking by using the ‘Gramcats’ sample from Bybee et al. (published in 1994). In this sample two thirds of all markings of passive are by additional stem affixes and thus morphological passives. The next most frequent marking of passive is the analytic passive where an auxiliary is combined with a passive participle. This marking is found in one fifth of the languages in the sample which Haspelmath used. The other types of passive marking are quite infrequent. Thus, almost 90% of the markings of passives are morphological passives. Haspelmath (ibid.) deducts from this that passive is a primarily morphological verbal category. He explains this with Bybee’s (1985: 13) principle of relevance: A meaning element is relevant to another meaning element if the semantic content of the first directly affects or modifies the semantic content of the second. If two meaning elements are, by their content, highly relevant to one another, then it is predicted that they may have lexical or inflectional expression, but if they are irrelevant to one another, then their combination will be restricted to syntactic expression. Passives are relevant to the verb since they change its semantic content by inactivization of the situation. Thus, passives are expected to be expressed lexically or morphologically according to Bybee’s (ibid.) principle of relevance. Haspelmath (1990) shows that morphological passives actually are by far the most frequent type of passives while only few passives have an analytic expression. Furthermore, the most frequent type of morphological passives is marked by an additional stem affix. Haspelmath also explains this by Bybee’s principle of relevance. In a survey of the languages of the world, Bybee found that valence affecting categories and voice Bybee (1985) takes Barber’s (1975) definition of voice (discussed in 3.2). Therefore, Bybee treats valence-affecting categories such as causative and voice as separate categories. are much more relevant than aspect, tense, mood, and agreement markers. According to Haspelmath, this is the reason why the marking by an additional stem affix, i.e. marking directly after the stem and before aspect, tense, mood, and agreement markers, is the most frequent kind of passive marking. In chapter 7, I will show that also the Greenlandic passive is marked by an additional stem affix. Before going over to the other parts of the passive construction, I want to draw attention to the fact that languages often have more than one formal marking of passive. This marking may be of the same or different types. Danish, for example, has two passives, one of which is marked morphologically by the extrainflectional affix s, and another one which is marked analytically by the auxiliary blive, ‘become’ and the perfect participle form of the verb. There are also languages which mark the passive strictly morphologically, or strictly analytically. According to Keenan (1985), in languages with several forms for passive, the different passive morphemes usually have different meanings. This is also the case in Greenlandic, which has several passive affixes (to be discussed in chapter 7). 4. 1.3 The promoted element In many cases, the element which is promoted to subject position in the passive clause is the patient and stands in object position in the active clause. There are, however, semantic roles other than the patient and relations other than the object in the active clause which can be promoted to subject status in the passive clause. Some languages can promote a benefactive, source, instrumental, locative, or other structure to the subject of a passive construction. Furthermore, the promoted element in the passive clause can be an indirect object or an oblique in the active clause. According to Dik (1997: 266), subject and object assignment is crosslinguistically carried out after a certain hierarchy which he calls ‘semantic function hierarchy’, which also plays a role elsewhere in typology: Agent > Goal Dik (1997: 122) calls the semantic roles of the patient ‘goal’ in his framework. > Beneficiary > Recipient > Instrument > Location > Time Subject + > + > + > + > + > + > + Object + > + > + > + > + > + Table 16: Dik’s (1997) ‘semantic function hierarchy’ When a goal, i.e. a patient, may be subject in a language, then also agents are subjects in this language. Languages with instrument subjects also have recipient, beneficiary, patient and agent subjects, and so on. The same is the case with objects, apart from the fact that agents cannot be objects. In the following example from Dalgish (1976), cited in Siewierska (1984: 71) from Olutsootso (Bantu, Niger-Congo) a locative in oblique position is promoted to the subject of the passive clause: (46) Esie en  deer  aanga eBi  taBo mushiiro. I 1SGbringT/ASP CL8 CL8 refers to class 8 and CL7 to class 7 of the nominal in Olutsootso. These are some of the Bantu noun classes.book LOCCL7market ‘I bring the books in the market.’ (47) Mu shi  iro mu  leer  uung w  a  mwo eBi taBo neende esie. LOCCL7market LOCbringT/ASPPASSASPLOC CL8book by I ‘In the market is brought the books by me.’ (rough approximation) In this language, the passive is expressed morphologically by the affix w. The verb deer, ‘bring’ in sentence (46) is active and the verb is passivized in sentence (47). In (47), it is not the object of the active sentence eBitaBo, ‘books’ which is the subject of the passive construction but the locative of the active clause mushiiro, ‘in the market’. In section 4.1.5 I will discuss impersonal passives which do not promote any element in the passive clause. Thus, the promotion of an element in the passive clause does not absolutely have to take place. 4.1.4 The demoted element In a passive clause, the agent is demoted to oblique status or omitted, but it is never expressed as a core argument. When the agent is expressed as an oblique constituent, this is usually called ‘agent phrase’. The agent phrase often contains the argument with the semantic role of the agent in an oblique case or with an adposition. According to Keenan (1985: 261), in a passive clause “an agent phrase will be an NP (with or without adpositions) which functions as the semantic but not syntactic subject of a verb in an expression derived from that verb (or verbphrase).” Actually, agent phrases do not absolutely have to contain an agent but whatever role may be subject of an active. Here, Keenan (1985: 261) gives the following example with the agent phrase by the church, which is a recipient rather than an agent: (48) Money is needed by the church. Agent phrases also exist in structures other than the passive, such as nominalizations and causatives; they are therefore not specific for passives. Furthermore, many languages with passives do not accept agent phrases. These structures are sometimes called agentless passives. A language with such a passive is Finnish. In this language, agents can never be expressed overtly in passive clauses (Vilkuna 1996: 142): (49) Ryöstäjä pidätettiin eilen aamulla. robber.NOM.SG catchPASS.PAST yesterday morningAD.SG ‘The robber was caught yesterday (in the) morning.’ In this way, the agent may be omitted from the passive clause in some languages, while it has to be omitted in other languages. In Greenlandic, as I will discuss in 7.3, the agent phrase may be present or omitted from the passive clause, as opposed to the agentless passive in Finnish. 4.1.5 Impersonal passives Impersonal passives are not always treated under the heading of passive, as can be seen in the definition of passive by Dixon (1994) in 4.0 above. Many scholars include impersonal passives, however, since these often have the same morphological or analytic marking as ‘personal passives’, i.e. all other passives. These structures are attested in many different languages of the world (see Siewierska 1984: 93). Impersonal passives are structures without any corearguments. Their meaning is always indefinite; therefore, they are called impersonal passives. Often the impersonal passive is formed from an intransitive active clause. The agent of this intransitive clause is demoted to oblique position or left out and the verb is passivized. As opposed to passives from transitive clauses, there is no patient or other participant present which could be promoted to subject position. The impersonal passive, however, is only backgrounding since it only involves the demotion of the agent. There is no argument promoted to subject position, therefore the subject slot is either not occupied or an indefinite pronoun or a socalled dummy subject occupies it. A dummy subject is an indefinite constituent which does not have the usual subject characteristics. It is, for example, not crossreferred in the verb. The impersonal passive is different from clauses with a general subject, such as man in German and on in French, since the latter clause type has no passive marking. The following examples from German show an impersonal passive without any core argument (50), and one with a dummy subject (51). The translations are rather literal and therefore of doubtful grammaticality in English: (50) Gestern wurde viel gelacht. yesterday be.PAST much PERF.PARTlaughPERF.PART ‘Yesterday much laughing was (done).’ (51) Es wurde gestern viel gelacht. dummy S ‘it’ be.PAST yesterday much PERF.PARTlaughPERF.PART ‘Yesterday it/there was much laughing (done).’ Until now, I have discussed impersonal passives as they are applied to intransitive verbs. Impersonal passives can also be applied to lexically transitive verbs. The impersonal passive clause has no core arguments like impersonal passives from intransitive clauses. Rather, all overt participants are expressed as obliques. An example of an impersonal passive applied to a transitive verb is the following from Irish (Celtic, IndoEuropean) from Keenan (1985: 275): (52) Bhuail si e. hit she him ‘She hit him.’ (53) Bhi se buailte aici. AUX he hit.PART at.her ‘He was hit by her.’ (54) Buaileadh (lei) e. hit.IMP.PASS (with her) him ‘There was hitting of him (by her).’ (rough approximation) In this example, the personal passive is expressed analytically (53), whereas the impersonal passive is expressed morphologically (54). Some languages can form impersonal passives only from transitive verbs and not from intransitive ones. According to Siewierska (1984: 98), Kannada (Dravidian) is such a language. 4.1.6 Stative and dynamic passives When languages have more than one form for the passive, it is often the case that one of the forms expresses a stative passive while the other forms are dynamic passives. On the other hand, when a language only has one form for the passive voice this form is usually a dynamic passive. Keenan (1985: 249) generalizes this in the following way as an implicational universal: “If a language has passives of stative verbs (e.g. lack, have, etc.) then it has passives of activity verbs.” Keenan’s ‘passives of activity verbs’ are what I call ‘dynamic passives’. The difference between dynamic and stative passives can be described in that dynamic passives focus on the action expressed by the verb, while stative passives focus on the state of the patient. Palmer (1994: 138) gives the following examples from German: (55) Als ich um fünf kam, when I.NOM at five come.PAST.1SG war die Tür geschlossen, … be.PAST.3SG DET.F.NOM door close.PERF.PART ‘When I came at five the door was shut, …’ (56) … aber ich weiss nicht, wann sie geschlossen wurde. but I.NOM know.1SG not when 3SG.F close.PERF.PART become.PAST. ‘… but I do not know when it was shut.’ In (55), the verbal form is war geschlossen, ‘was closed’ and in (56), the verbal form is wurde geschlossen ‘became closed’. In (55), the focus is on the state of the door, while in (56) the focus is on the action. The passive in (55) can therefore be described as a stative passive and the passive in (56) as a dynamic passive. When (55) and (56) are translated into English, there is no formal difference between stative and dynamic passive even if the difference exists semantically. This can be seen in the following equivalent taken from Jespersen (1924: 274): (57) When I came at five, the door was shut, … (58) … but I do not know when it was shut. In both (57) and (58) the verbal form is was shut, but the semantic content is very different: in (57) it focuses on the state of the door, while in (58) the focus is on the action. Thus, the difference between stative and dynamic passives may be expressed in languages even when there is no formal marking. In (55), the stative passive is expressed analytically. Actually, stative passives are usually expressed analytically and the auxiliary is often be. Siewierska (1984) claims that in languages with both morphological and analytic passives, the analytic passive is often more stative and perfective in meaning. The clause die Tür war geschlossen, ‘the door was closed’ can be analyzed as consisting of the grammatical subject die Tür and the verbal phrase expressing the passive war geschlossen. Another analysis is possible here, however. The clause can be analyzed as consisting of the subject, followed by the copula war and the adjectival predicative geschlossen. This is possible, since participial forms can either be analyzed as main verbs in the verbal phrase or as adjectives. Both the passive and the adjectival analysis are possible for stative passives since the focus is on the state of the patient and this state may be expressed by an adjective. For dynamic passives, on the other hand, the passive analysis is the most obvious since the focus is on the action expressed by the participle which in this way is part of the verbal phrase. Apart from the fact that stative passives can be analyzed as copula clauses, they usually do not have corresponding active clauses. All other kinds of passives have active counterparts. Thus, if one defines passives as having an active equivalent, stative passives would not be included in this definition of passive. Siewierska (1984. 127) argues that stative passives do not have corresponding actives because they are actives themselves and should be analyzed as copula clauses with an adjectival predicative. 4.2 Passives in ergative systems Since Greenlandic is an ergative language, I briefly want to discuss passives in ergative systems. As discussed in 3.3, passives are found in both ergative and accusative languages, whereas antipassives are restricted to languages having a special focus on the object, most of which are ergative languages. Passives in ergative systems differ from passives in accusative systems in the nature of cases assigned to the nominal constituents in relation to the verb. In accusative systems, the application of the passive changes the case marking of both A and O of the active clause. A is in the nominative case and becomes an agent phrase marked in an oblique case in the passive while the O is marked by the accusative case in the active and by the nominative in the passive. In languages with ergative structures, the A is marked by the ergative case in the active and by an oblique case in the passive while the O is in the absolutive case both in the active and in the passive. Thus the passive in accusative systems involves two ‘new’ case markings since both constituents are marked differently whereas the passive in ergative systems only involves a change in the case marking of the A. The reason why quite a few ergative languages have a passive is probably that most ergative languages involve S/A pivots in clause combining or these languages combine S and A in other ways, such as in switch reference systems (Nichols 1992). Thus, these languages are syntactically accusative (discussed in 3.3). In this way, O and S are in the same morphological case, i.e. the absolutive, but they cannot be coded as coreferent in clause combinations. The O has to be brought into subject position to fulfil restrictions in clause combination even if the language is morphologically ergative. 4.3 Structures with the same formal marking as passives In many languages, the same structure marking the passive is also used for other phenomena. The following example shows the Danish passive which is marked in the same way as the reciprocal. In (59) the –s marks the reciprocal, while in (60) it marks the passive: (59) Vi ses! We seeRECIP ‘See you!’ (literally: we see each other) (60) Jeg tror vi kan ses af vagten. I think we can seePASS by guardDET ‘I think we can be seen by the guard.’ According to Haspelmath (1990), the same morphological process as is used in the passive is primarily found also in reflexives, reciprocals, resultatives, anticausatives (middles), potential passives, fientives, reflexive causatives, deobjectives (antipassives), and desubjectives. I will briefly explain the terms resultative, potential passive, fientive, and desubjective, as some of them are not commonly used in linguistics. According to Nedjalkov & Jaxontov (1988: 6), “the term resultative is applied to those verb forms that express a state implying a previous event”. Nedjalkov & Jaxontov (1988) describe stative passives as ‘objective resultatives’ since stative passives express the result of an event from the view of the object. According to Haspelmath (1990: 33), the potential passive is a structure where “the subject is capable of undergoing an action”. An example of a language with a potential passive is Mwera (Bantu, NigerCongo) coma, ‘read’ and comeka, ‘be readable’ from Haspelmath (1990: 33). The fientive, also called ‘inchoative’, “derives a process of becoming from stative expressions”, according to Haspelmath (1990: 34). Thus the Mwera (Bantu, NigerCongo) adjective –tali, ‘long’ is talika, ‘become long’ in the fientive, according to Haspelmath. In a desubjective construction, the subject is removed from the clause without any other argument being promoted to subject status. This structure is found in Finnish in the following example from Comrie (1977), cited in Haspelmath (1990: 34): (61) Maija söi sen. Maija.NOM eat.3SG.PAST it.ACC ‘Maija ate it.’ (62) Syöttiin sen. eatDESUBJ.PAST it.ACC ‘’It was eaten.’ In the desubjective clause (62), the subject Maija is deleted, not simply missing because of ellipsis and the object sen, ‘it’ is not promoted to the subject but still in the objective accusative case. In the following section, I will discuss one of several reasons for passive and other structures having the same formal marking, namely that some of these structures have developed into passives. 4.4 The development of passives I have discussed in the preceding section that the use of passive markers is not restricted to passive structures but that these same markers are also found in other structures. This may give a clue to the emergence of passives, which is the concern in the present section. I want to look at the development of passive morphology and analytic structures, taking most of my information from Haspelmath (1990). According to Haspelmath, grammaticization is the primary path of development of passive morphology. The term ‘grammaticization’, or usually ‘grammaticalization’ According to Bybee et al. (1994: 4, footnote 2), the terms ‘grammaticization’ and ‘grammaticalization’ are used interchangeably in the literature. I use the term ‘grammaticization’ in this study, because I refer to Haspelmath (1990), who uses this term. was first introduced by Meillet (1912b), according to Bussmann (1990). It covers a process where grammatical elements develop from an autonomous lexical source. The lexical source often loses its syntactic independence, reduces its form, and becomes an affix. Furthermore, it becomes obligatory, which is a feature of grammatical categories as opposed to lexical items. Haspelmath (1990: 57) describes the direction of grammaticization as unidirectional However, Haspelmath (1990: 578) lists some exceptions to unidirectionality in grammaticization.; i.e. the development only goes in one direction. Lexical items are generalized when being involved in grammaticization and they cannot develop back into concrete lexical items or in items which are more lexical than the present structure. Reflexives, for example, can be identified lexically. According to Haspelmath, reflexives can develop into passives, which are more general in meaning, through grammaticization. Passives, on the other hand, cannot usually develop into reflexives because reflexives are more specific (lexical) in meaning than passives. In 4.5, I will discuss that the unidirectional development of grammaticization can proceed even further, since passives can develop into ergative systems which are even more general. Above I discussed passives that develop from reflexives, but there are also other structures from which passives can develop. Haspelmath (1990) list the following four major sources for passives: inactive auxiliaries, NPreflexives, causatives, and generalized subject constructions. I will look at each of these structures respectively. 4.4.1 Development of passives from inactive auxiliaries Inactive auxiliaries develop from nonagentive verbs, i.e. verbs that have the patient as their subject. Most nonagentive verbs are intransitive, such as be and become, but there are also transitive nonagentive verbs, such as suffer and receive. According to Haspelmath (1990: 38), these main verbs develop into inactive auxiliaries which build up analytic passive structures. These auxiliaries may further develop into affixes. Haspelmath (1990: 39) has the following example from Korean ggeggeji, ‘be broken’, where the passive is expressed by the affix –ji. This affix is derived from the inactive auxiliary ji, ‘fall’. 4.4.2 Development of passives from NPreflexives Reflexive pronouns such as himself or in some languages reflexive ‘nouns’, such as head, body and soul, can develop into passive morphemes. According to Haspelmath (1990), the intermediate state in the grammaticization of NPreflexives to passives is the anticausative, i.e. spontaneous middle (discussed in 3.1.4). When reflexives develop into passives, the first thing that happens is that the subject loses its agentivity. The structure which arises from the loss of agentivity is an anticausative. This can be shown in the following example from Haspelmath (1990: 44): (63) John hurts himself. This sentence is ambiguous in that John could have hurt himself intentionally, in which case John is the agent and the structure is a reflexive, or John did not hurt himself intentionally. In the latter case, no agent is presupposed and this structure is an anticausative. The use of the reflexive as an anticausative marker can be extended to clauses with inanimate subjects. Since inanimate subjects cannot be agents themselves, a clause with reflexive morphology and an inanimate subject is automatically understood as an anticausative, as can be seen in the following example from German from Haspelmath (1990: 45): (64) Die Erde dreht sich. In this example, the meaning is anticausative since the earth, as an inanimate item, cannot spin itself. The anticausative can develop into a passive structure when an agent is added into oblique position or when an agent is implied. An example from Russian of a reflexive which developed into a passive is given below (from Siewierska 1984: 162). According to Haspelmath, the reflexive morpheme –sja can express the passive in the imperfective aspect: (65) Rebenok umylsja. child.NOM wash.PERFPASTREFL ‘The child washed himself.’ (66) Devocka myla pol. girlNOM wash.IMPFPAST floor.ACC ‘The girl was washing the floor.’ (67) Pol mylsja devockoj. floor.NOM wash.IMPFREFL girlINSTR ‘The floor was being washed by the girl.’ In (65) –sja is used to mark the reflexive, while in example (67) –sja is used to mark the passive. Siewierska (1984) calls those passives which develop from reflexives ‘passive reflexives’. 4.4.3 Development of passives from causatives Passives and causatives are almost opposite in meaning since causatives focus on the causation of an action, while passives focus on the state of the patient or other semantic role. Furthermore, causatives are usually valence increasing since they add a new participant, while passives are usually valence decreasing because of the demotion of the agent. Still, passives have developed from causatives. According to Haspelmath (1990), socalled reflexive causatives are the intermediate stage between causative and passive. This can be seen in the following example from Keenan (1985: 263) who claimed that the English ‘getpassive’ developed from a causative: (68) John got Bill fired (causative) (69) John got himself fired (reflexive causative) (70) John got fired (passive) (71) The window got broken (passive) The reflexive causative in (69) develops into a passive (70) by the loss of agency: while John is both agent and patient in (69), he is only patient in (70). The kind of passive in (70) can be extended to all possible uses of passives by analogy as in (71). In this example, the subject of the clause is inanimate and therefore it cannot be the agent of a causative construction. The present discussion about passives developing from causatives is important in the description of the passive in Greenlandic, and I will come back to it in chapter 7. 4.4.4 Development of passives from generalized subject constructions Generalized subject constructions are clauses with an indefinite pronominal subject. An indefinite subject is often a 3rd person plural pronoun or an indefinite pronoun, such as German man. Since the agent of generalized subject constructions is by definition indefinite, one can say that it is backgrounded. According to Haspelmath (1990), the subject crossreference on the verb can be reinterpreted as a desubjective marker since the subject is indefinite. In this way, the patient may gain subject status and the general subject construction becomes a passive. Givón (1979: 188, 211), cited in Haspelmath (1990: 4950), gives the following two examples from Kimbundu (Bantu, NigerCongo): (72) Nzua, amumono. Nzua 3PL.A3SG.Osee ‘Nzua, they saw him.’ (73) Nzua amumono kwa meme. Nzua PASS3SG.Ssee by me ‘Nzua was seen by me.’ Example (72) shows the construction with the generalized subject in the 3rd person plural. In (73), an agent phrase is added. Nzua, the object in (72), is reinterpreted as the subject. The crossreference marker of the generalized subject has become reinterpreted as a marker for the passive. 4.4.5 Development of passives which do not arise via grammaticization Most of the passive marking arises through grammaticization but there is another source for markers of passive. According to Haspelmath (1990: 51), “It appears that the next most important source for general grammatical morphemes (after grammaticization) is the lexical expansion of initially idiosyncratic derivational morphemes.” According to Haspelmath, this is the case in Greek, where passive affixes arose through lexical expansion of morphemes which originally marked the verbal stem for ‘inactive’ meaning. 4.5 Development of ergative structures from passives In the preceding section, I discussed grammaticization and other processes whereby passives develop. Even if passives may develop from very different sources, the resulting passive structures are very similar in their function. This is because the sources from which passives develop become generalized. The generalization from a number of different structures leads to one structure: the passive. Passives can furthermore be generalized and develop into ergative structures. There are other sources for ergativity as well, but one of them is certainly the passive. A passive can develop into an ergative structure when the passive is very productive and regularly used in the language. The passive structure may then become generalized and is used as the unmarked clause structure in the language. According to Haspelmath (1990), the grammaticization precedes in the following way: the agent phrase of the passive becomes an important and obligatory participant. Then it gains subject properties, such as being crossreferred in the verb. The originally oblique case marking of the former agent phrase becomes the ergative case marker. The patient is still in the basic case form nominative, or rather absolutive, as this case is called in ergative structures. In this way, the passive structure which is marked becomes an unmarked structure. According to Comrie’s (1988: 1920) conditions for markedness (discussed in 2.4) the passive structure is marked because it is not the most frequent structure in the language, whereas the ergative is unmarked, being the most frequent structure in transitive sentences in an ergative language. Furthermore, the passive is formally more complex because there is passive marking on the verb, whereas there is no such marking in the ergative structure. Since the ergative structure is the typical alignment structure in an ergative language, it is very productive. The passive is not so productive, since, according to Comrie (1988: 20) many verbs, especially intransitive verbs, have actives but no passives. Finally, the passive is restricted to certain discourse conditions which the ergative is not. Thus, the passive is marked in discourse distribution as well. According to Haspelmath (1990), the ergative structure is the endpoint of this grammaticization. Passives develop from lexical structures, or at least from structures with lexical traits. In relation to these structures, passives are generalized. However, the passive only appears under certain conditions and has some lexical meaning left, namely ‘inactivity’. Ergative structures, on the other hand, contain no lexical information at all but are obligatory structures. Haspelmath (1990: 56) expresses this in the following way: “thus the observed endpoint of the grammaticization changes discussed here coincides with its logical endpoint: zero; i.e., no meaning is left, the ergative construction is the unmarked clause construction with maximal generality.” Ergative structures are the endpoint of this type of grammaticization, but this does not mean that ergative languages cannot change their alignment type; languages commonly change from ergative to other alignment types (Dixon 1994). The change of alignment is another process than the grammaticization discussed here of a lexical item to a grammatical morpheme, the ergative marker. 4.6 The functions of the passive Often the functions of the passive are described in terms of foregrounding of a nonagent and backgrounding of the agent. Thus, Keenan (1985), Dixon (1994), and Foley & Van Valin (1984) include foregrounding and backgrounding as central functions of the passive. According to Foley & Van Valin (ibid.), the backgrounding function of the passive is found in all kinds of languages, whereas the foregrounding function of the passive is found only in languages which have pivotrestrictions on S and A. These languages need a passive to bring an O NP into pivot condition. Siewierska (1984: 217ff.) discusses the functions of the passive, citing Givón (1981) who groups the functions in the following domains: topic identification, impersonalization, and detransitivization. Topic identification implies the foregrounding of a nonagent. Impersonalization, on the other hand, involves the backgrounding of the agent. Through detransitivization, the passive clause is lower in transitivity than the active clause. In the rest of her discussion of the functions of the passive, however, Siewierska looks only at analytic structures from some European languages which she describes in terms of foregrounding and backgrounding structures. Haspelmath (1990), on the other hand, does not see foregrounding and backgrounding as the primary functions of the passive. He claims that the primary and original function of the passive is the inactivization of the situation. Foregrounding and backgrounding can be described as secondary functions since, according to Haspelmath (1990: 60), “the two functions of agent backgrounding and patient foregrounding follow automatically from inactivization”: the agent becomes less important in an inactive situation and it is thus backgrounded. Since the subject slot is free, the patient can be promoted to subject status. Haspelmath (ibid.) investigates the function of the passive by looking at how passive morphology develops. The diachronical developments from certain structures to the passive, as discussed in 4.4, all involve inactivization. In the case where the passive develops from generalized subject constructions, the backgrounding of the agent plays a larger role than inactivization, however. Nevertheless, all in all the primary function of the passive seems to be the inactivization of the situation. However, Haspelmath (1990: 59) points out that this primary function of the passive may change: “As the construction becomes more and more grammaticized it also becomes more widespread and may be put to use for various pragmatic functions that differ widely across languages. In this way, a function may become dominant that is quite different from the original function.” According to Haspelmath, the passive is primarily a verbal category, involving changes directly in the verb instead of primarily changing relations in nominal statuses which is implied in the foregrounding and backgrounding functions of the passive. Haspelmath (1990: 69) expresses this in the following way: “that the basic function of the passive is to modify the situation expressed by the verb rather than to alter the prominence relations among the verb’s participants should also be clear from the fact that passive is a verbal category, not a category expressed on NPs.” 4.7 Summary I the present chapter, I have described the forms (4.13), the diachronical development (4.45), and the functions (4.6) of the passive. I have discussed different approaches to the passive, which vary in their definition of this structure. Certain structures are included under the heading of ‘passive’ in some definitions while not in others. Structures, which are sometimes excluded are lexically marked passives, impersonal passives, and stative passives. Many definitions of the passive do not include lexically marked structures. This is the case because lexical passives are not grammatical categories like the passives marked analytically or morphologically but simply lexical forms. However, Payne (1997) includes lexical passives into his definition. Impersonal passives are usually included in the definitions of passives. However, theories that define passives in terms of transitive structures which are detransitivized by application of the passive may not include impersonal passives since impersonal passives do not always involve transitive active structures. This is the case in Dixon’s (1994) definition of the passive (discussed in 4.0) where transitivity of the active clause is mentioned as the first part (a) of his definition. In this way, Dixon does not include impersonal passives. Stative passives lie in between passive structures and copula clauses with an adjectival predicative. Siewierska (1984) argues for the latter interpretation since stative passives often resemble actives in not having corresponding active clauses. I will follow Haspelmath’s (1990) definition of passive in the remainder of this study. This definition does not include lexical structures under the heading of ‘passive’. He does include impersonal and stative passives in his definition. Haspelmath defines passive as a morphological verbal category. Passive in his view is a morphological category since the most common formal marking of the passive is morphological. Haspelmath defines passive as a verbal category since it changes the semantic content of the verb and at the same time the passive is expressed in the verb: as an affix on the verb or by marking in the verbal phrase. In this way, not the restructuring of the nominal elements in the clause is the most important function of the passive, but the change in the meaning of the verb. However, the restructuring of the nominal elements takes place. Thus, the passive is defined as a formally marked (morphological) verbal category, which inactivizes the situation expressed by the clause. Haspelmath’s definition of one of the voice structures, namely the passive, fits in the alternative definition of voice I gave in 3.4. I defined voice as a category which changes the structure of the arguments in a clause by at least formal marking in the verb. According to Haspelmath, the passive changes the structure of the arguments in the clause, even if this is not the primary function of this voice structure. Furthermore, Haspelmath includes only structures which are formally marked in his definition of passive structures, which is also a part of my definition of voice. In section 3.3, I have discussed Nichols’ (1992) investigation of the relationships between grammatical categories and structural types in the languages of the world. She came to the conclusion that voice, and therefore also passive, is a syntactic rather than morphological category. She argues for passive being a syntactic category since the antipassive is usually found in languages with special focus on the object, while the passive is more ‘unmarked’ in respect to grammatical relations and exists in both morphologically accusative and ergative languages. The appearance of the passive in different kinds of languages can be explained by the fact that nearly all languages of the world, even if morphologically ergative, are syntactically accusative. Haspelmath (1990), on the other hand, describes passive in terms of morphology. Thus, Nichols (1992) and Haspelmath seem to have diverging views on the passive. However, these two approaches can be reconciled: according to Haspelmath (ibid.), voice is a morphological category since it is marked in the verb. Its function is primarily the inactivization of the situation. Haspelmath says that the functions may develop from the primary function of inactivization of the situation to other functions. However, in many languages the passive developed the function to fulfil pivot conditions. Since pivot restrictions are usually accusative in nature, i.e. S and A are treated alike in adjacent clauses, the passive is found in all kinds of languages, even those which are morphologically ergative. The passive has therefore often the function to serve syntactic accusative structures. I would not say that the passive is syntactic in nature, as Nichols (ibid.) does, but rather that the passive is a morphological, verbal category which has some functions in the syntax. 5. An introduction to the structure of Greenlandic In the present chapter, I want to give an introduction to the structure of Greenlandic. This introduction is not intended as a complete sketch of the grammar of this language, rather, I want to focus on those structures which will be relevant in the examples given in chapter 6 and 7. When referring to Greenlandic in this study, I mean more specifically West Greenlandic, the dominant variant in Greenland with about 38,000 speakers. It is spoken on the west coast of Greenland from Nanortalik in the South to Upernavik in the North. The other dialects are East Greenlandic on the east coast with about 3,000 speakers and Polar Eskimo in the Thulearea with about 700 speakers. South Greenlandic, which is usually referred to as a dialect of West Greenlandic, is developing in the direction of a grammatically and lexically distinct dialect. Greenlandic and Inuktitut from northern Canada and Alaska form the language group InuitInupiaq. InuitInupiaq and Yup’ik, a language spoken in Alaska and a few places in Siberia, form the group of languages referred to as Eskimo. Eskimo languages are further related to the language of the Aleut Islands. Together they form the language family EskimoAleut. In recent years, a genetic relationship between EskimoAleut and the language family ChukotkoKamchatkan in northeastern Siberia has been suggested. Moreover, people have suggested parallels to the Uralic language family to which Finnish belongs (among others Fortescue 1997). In the present chapter, I primarily want to introduce Greenlandic morphosyntax. Before that, I have some comments on phonology. The Greenlandic vowel system consists of three vowels /a/, /i/, and /u/. The vowels /i/ and /u/ have the allophones /e/ and /o/, which appear before the consonants /r/ and /q/. Some descriptions of Greenlandic (e.g. Fortescue 1984) omit the allophones in their examples, using a phonological transcription. Fortescue (1984) lists the following consonants for Greenlandic: labial labiodental alveolar postalveolar palatal velar uvular glottal Plosives p t k q Fricatives v s s According to Fortescue (1984: 334), younger speakers do not use this phoneme. j g r h Nasals m n  N Liquids l r According to Fortescue (1984: 335), this phoneme is only found in Nordic loan words. Table 17: consonants in Greenlandic Even if Greenlandic phonology seems quite simple, morphophonological rules can make the combination of morphemes rather complicated. Greenlandic has three different verbal stems: vowel stems, which end in a vowel, consonant stems, which end in one of the consonants except /r/ Fortescue presents the final consonant of these stems as a C, since this consonant always assimilates to the following consonant. I will also use this marking in the present study., and finally ‘r stems’. The r-stems and consonant stems may result in different forms of the following affixes. Furthermore, there are regressive rules, i.e. rules which operate backward, which involve different types of affixes and progressive rules, i.e. operating forward, in the assimilation of vowels. I will discuss some here. When affixes are attached to stems, changes in the form of the stem can appear; these changes depend on the kind of affix involved. According to Fortescue (1992), there are four kinds of affixes: truncating, additive, replacive, and fusing affixes, which I want to describe briefly. Truncating affixes have the effect of deleting the final consonant in the stem. An example of a truncating affix is –rujussuaq, ‘very big’. When this affix is added to timmisartoq, ‘airplane’ the last consonant /q/ is deleted: timmisartorujussuaq, ‘a very big airplane’ (Olsen & Hertling 1988: 68). Additive affixes are added directly to the stem but they often provoke sound changes. Thus, in the following example from Olsen & Hertling (1988: 39) the final /q/ of Kangeq (the name of a village) becomes /r/ in Kangermioq, ‘a person from Kangeq’ with the additive affix –mioq, ‘a person from’. All passive affixes (to be discussed in chapter 7) are additive affixes. Replacive affixes usually start with an /l/ and delete the final consonant (if there is any) and vowel of a stem while losing their own initial consonant. According to Fortescue (1984: 350), nerisassaq, ‘food’ + liorpoq, ‘she makes’ becomes nerisassiorpoq, ‘she made food’. Finally, fusing affixes start with a /g/ and behave in the same way as truncating affixes, apart from the fact that /g/ becomes /r/ when added to a uvular. Thus, arnaq, ‘woman’, ending with the uvular /q/ and gaa, ‘have it as’ becomes arnaraa, ‘she is his mother’ according to Fortescue (1984: 351). Progressive assimilation of vowels involves the vowel /a/. When /a/ is the final vowel of a stem, the vowels that follow it are assimilated to /a/. In this way, /i/ and /u/ are frequently changed into /a/. Thus, the truncating affix for the copula suffix u, ‘to be’ becomes /a/ when the last vowel in the preceding stem is an /a/, as nakorsaq, ‘doctor’ together with the copula in nakorsaavoq, ‘he is a doctor’. All this shows that the forms of an affix may be completely changed. This often makes it difficult for linguists to find out about the forms of the different affixes. Another example of an affix that is changed completely when added to a base is (ss)ut(i), ‘to him, with it’, which can turn into up or even ap (this affix will be discussed in 6.1.5). The affix (ss)ut(i) is truncating, and in many cases the /ss/ in the beginning is left out. When ut(i) appears directly before an (additive) inflectional ending, the final /i/ disappears, and the consonant /t/ is assimilated to the first consonant of the inflectional affix, which is /p/. In this way, the affix (ss)ut(i) becomes up, as seen in the following example (Fortescue 1984: 89): (74) Tikkuuppaa. point.outAPPL3SG.3SG.INDIC ‘He points something out for him.’ In some cases, this affix is changed even more since an /u/ can turn into an /a/ when the preceding vowel is an /a/ (see above). This results in the affix ap which in no way resembles the primary form (ss)ut(i) given in Fortescue (1984). This is shown in the following example from Fortescue (1984: 89): (75) Kamaappaa. be.angry.withAPPL3SG.3SG.INDIC ‘He is angry with him.’ I will now turn to the morphosyntactic structure of Greenlandic. Greenlandic is a polysynthetic language; and according to Mithun (1988), verbs are central in polysynthetic languages. This is true for Greenlandic, where the verb can be the only element in a sentence. Agreementmarkers in every verb crossrefer the subject in intransitive sentences and the subject and object in transitive sentences In the socalled contemporative mood (to be discussed below), however, only the object is referred to in the crossreference ending of transitive clauses.; therefore, nominal elements are present in the sentence even if verbexternal arguments can be omitted. According to Mithun, morphological complexity exists primarily in the verb in polysynthetic languages. This is also the case in Greenlandic even if this language also has nominal inflection and derivation. However, the inflection and derivation in the verbs is much more complex. Tense, which is a common inflectional verbal category in European languages, is not an inflectional category in Greenlandic. The future tense and different kinds of aspect may, however, be expressed by derivational affixes. Present and past tense are not grammatically distinct; they are expressed by the same inflectional endings in the verb. Therefore, the interpretation of a clause depends on its context, as in the following example from Fortescue (1984: 272) which can be translated in the past or present tense: (76) Aggerpoq. come3SG.INDIC ‘He is/ was coming.’ There are two basic clause types in Greenlandic. In the first type, the clause consists of an inflected verb with or without verbexternal NP constituents. The second type is an equational construction with the copula particle tassa, ‘it is’, followed by an NP. The clauses of the first type are much more common than those of the second type. In clauses of the first type, the word order is SV, i.e. intransitive subject and verb, in intransitive clauses and AOV, i.e. transitive subject, object and verb, in transitive clauses when all verbexternal elements are present. However, many clauses consist simply of a verb. Fortescue (1993: 269) expresses this in the following way: It is rather rare in texts for all possible nominal arguments of a verb to be explicitly present as independent constituents rather than crossreferred to by the verb’s inflection alone. Greenlandic words can be divided into three classes: nominals, verbs, and particles. Meanings covered by English adjectives are expressed in verbs in Greenlandic. See the following example from Janussen (1987: 24): (77) Angut kusanarpoq. man.ABS.SG be.handsome3SG.INDIC ‘The man is handsome.’ Other modifications, such as adverbials (what is expressed in e.g. English), are usually expressed by derivational means in the verb. There are more than 400 productive derivational affixes in Greenlandic, many of those with meanings like adverbials in languages such as English. There are no prefixes. Particles do not inflect, while both nominals and verbs can have derivational endings and nominals and verbs always have inflectional endings. The word structure of nominals and verbs is as follows: root (derivational affixes: up to 10) inflection (clitic) Table 18: the structure of nominals and verbs in Greenlandic Nominals and verbs always consist of at least a root and an inflectional ending. The inflectional ending always consists of a mood marker and crossreference markers. Derivational affixes and clitics are optional. The following constructed example shows the structure of a verb containing a root, a derivational affix, the inflectional ending, and a clitic: (78) Qimmeqarpungalu. doghave1SG.INDICalso ‘I have also a dog.’ In this example qimme, ‘dog’ is a nominal root which is verbalized by the derivational affix qar, ‘have’, followed by the inflectional ending –punga, ‘I’ and the clitic –lu, ‘and, also’. The root, with the derivational affixes form the stem, and this in turn is followed by the inflection. There is a strict difference between inflection and derivation in Greenlandic. Fortescue (1984) calls the affixes between the root and the welldefined inflectional ending all derivational. The derivational endings form complex verbal or nominal stems. The order of derivational affixes is mostly not fixed, but a different order of the derivational affixes can provoke differences in meaning, as shown in the following examples with the transitive verb orniC, ‘come to’, taken from Fortescue (1984: 313): (79) Ornikkusunneqarpoq. come.towantPASS3SG.INDIC ‘Somebody wanted to come to him.’ (80) Orninneqarusuppoq. come.toPASSwant3SG.INDIC ‘He wanted somebody to come to him.’ According to Fortescue (ibid.), “each successive affix modifies all of the word ‘to its left’”. This can be seen in the examples (79) and (80). In (79), the passive affix neqar stands right before the inflectional ending and it modifies the rest of the word to its left. Thus, when reading this clause morpheme for morpheme starting with the inflection in the end, the translation is ‘he was wanted (to) come to’ meaning ‘somebody wanted to come to him’. In (80), on the other hand, the passive affix only modifies the verbal stem ‘come to’, whereas rusuC, ‘want’ modifies the passive morpheme and the verbal root. The morpheme for morpheme translation of (80) is thus ‘he wanted to get come to’, meaning ‘he wanted somebody to come to him’. According to Fortescue (1980), the derivational affixes can be divided into classes, depending on how they affect the stem. He distinguishes between the following types of derivational affixes: verbalizing, verbextending, verbmodifying, nominalizing, nominalextending, nominalmodifying affixes, and sentential verbal affixes. Nouns can be verbalized by a verbalizing affix and verbal stems can be nominalized by a nominalizing affix. In this way, the difference between nominals and verbs is primarily seen in the inflectional ending as in example (78) above. I will first look at the inflectional forms of the verb and then I will describe the forms of the nominals. The inflectional ending of the verb consists of a mood marker and a crossreference marker. The latter refers to the S in intransitive clauses and to the A and O in transitive clauses: moodmarker – crossreference marker (S, or A and O) The crossreference ending is mainly structured on ergative grounds, which means that the forms for S and O are the same while A is marked in another way. The transitive crossreference marker consists diachronically of two different forms for A and O (Fortescue 1984, discussed in Sakel 1998b). Synchronically, however, it is not always possible to analyze the ending as two separate morphemes. Rather, the ending is analyzed as one portmanteau morpheme. The forms of the inflectional marking in Greenlandic are given in appendix 1. As both subject and object are crossreferred in the verb, the division between intransitive and transitive clauses is quite clearcut: verbs with crossreference to one participant are intransitive while verbs with crossreference to two participants are transitive. Some verbs in Greenlandic, however, may be both transitive and intransitive without any further marking. These verbs can behave in two different ways. The first possibility is that these verbs have the subject of their transitive form as the subject of their intransitive form; Fortescue (1984) calls these verbs ‘agentive verbs’. The second possibility is that these verbs have the object of their transitive form as the subject of the intransitive verb, and these verbs are called ‘non-agentive verbs’ by Fortescue. According to Fortescue (1984: 85), an agentive verb is neri, ‘eat’, which has the same argument as the subject in transitive and intransitive use. A nonagentive verb is napi, ‘break’, which has the object of the transitive clause as the subject of the intransitive clause: napivoq, ‘it is broken’ and napivaa, ‘he broke it’. Furthermore, strictly transitive verbs may be inflected intransitively which results in a reflexive meaning. The strictly transitive verb toquppaa, ‘he kills him’ has a reflexive meaning when inflected intransitively: toquppoq, ‘he kills himself’. NPs other than S or A and O are usually called ‘oblique NPs’ in the description of Greenlandic. Oblique NPs are not crossreferred in the verb, but must appear verbexternally when present. Therefore, there are no ditransitive verbs since all other participants in a clause are expressed by oblique cases. There are eight moods in Greenlandic, all of which are marked by a distinct moodmarker. What is called mood in Greenlandic grammar is usually called ‘mode’ in the literature, but since the term ‘mood’ is traditionally used in Greenlandic grammar (e.g. Fortescue 1984) I will also use it here. Four of these moods are independent, i.e. they appear in main clauses, while the other four are subordinate moods, i.e. they only appear in subordinated clauses. Since only one verb can be marked by an independent mood in a sentence, all other verbs are in subordinate moods; this is also the case in coordination. The independent moods are indicative, imperative, optative, and interrogative. The subordinate moods are contemporative, causative, conditional, and participial. I briefly want to explain the meanings of some of these moods. The optative expresses wishes, such as in takulara, ‘let me see it!’. The interrogative mood is used in questions. The contemporative and participial moods are idiosyncratic terms as used for Greenlandic grammar. The contemporative mood expresses that the verbal action of the subordinated clause is simultaneous with the main clause, with the structural restriction that the subjects of the subordinate clause and the main clause are coreferent. The participial ‘mood’ is a nominalization of the embedded verb. One of the reasons why this structure is called a mood in Greenlandic is that it interacts with the contemporative mood in one of the two switch reference systems in Greenlandic (see below and Sakel 1998a). Finally, the causative mood expresses a reason or indication of time. It expresses the subjective point of view of the speaker about the situation, and it does not change the situation expressed by the verb by addition of a causer, as is the case in the causative voice (discussed in 3.1.7 and to be discussed for Greenlandic in 6.1.4). In the remainder of the present study, I call the causative voice simply ‘causative’. The so-called causative mood is irrelevant to my study, and will not be discussed further. While the independent moods inflect for first, second, and third person, the subordinate moods inflect for three persons plus a fourth person. The notion of fourth person can be misleading since it does not refer to another person. On the contrary, it means that the person referred to by the fourth person suffix is coreferent with the subject of the main clause. This can be seen in the following example from Fortescue (1991: 534): (81) Tikikkami iserpoq. come4SG.CAU.MOOD enter3SG.INDIC ‘When he (1) arrived, he (1) went in.’ (82) Tikimmat iserpoq. come3SG.CAU.MOOD enter3SG.INDIC ‘When he (1) arrived, he (2) went in.’ In example (81), the subject of the main clause iserpoq, ‘he went in’ is coreferent with the subject in the subordinate clause which is marked by the socalled fourth person inflection in the latter. In (82), on the other hand, the subject of the main clause is not coreferent with the subject of the subordinate clause which can be seen in the third person marking in the latter. This phenomenon, involving coreferentiality in two adjacent clauses, is called ‘switch reference’ and was first introduced by William Jacobsen (1961). Greenlandic has two forms of switch reference. One involves the difference between the so-called third and fourth person as in the examples above. The other one involves the subordinate contemporative mood which is used only when the subjects of the subordinate and the main clause are coreferent. Berge (1997) suggests a third kind of switch reference in possessive structures. Even if this marking is parallel to the fourth person marking, I do not include it as a form of switch reference, since possessor marking is nominal, while switch reference is a verbal category. These two kinds of switch reference are only partially parallel, as I discuss in Sakel (1998a). The marking of coreferentiality by the fourth person or the contemporative mood is also called ‘same referent’ or ‘same subject’ marking (SS), while the marking of noncoreferentiality by the third person or the participial mood is called ‘different referent’ or ‘different subject’ (DS). The group of nominals contains nominal stems, pronouns, and numerals. There are personal pronouns for the first and second person while the forms of the third person pronouns are the same as those for demonstrative pronouns. All nominals inflect for number and case, and in possessive constructions the nominal stems inflect for possession as well. The category of number contains singular and plural. In other Eskimo languages, there is a dual number but this has disappeared from West Greenlandic. Greenlandic case marking follows an ergative pattern. There are eight cases. The core cases are absolutive, which is zero-marked, and ergative, which is marked by the suffix –(u)p. These two core cases are purely syntactic in nature: they denote grammatical relations. The other six cases are oblique cases, which are not crossreferred in the verb. These cases are instrumental, allative, locative, ablative, prosecutive According to Fortescue (1984: 226), “the prosecutive case may (…) indicate general location (…) in the sense of field/ sphere of action.” Furthermore, “the prosecutive case may have a sense of (movement) through or among” (Fortescue (1984: 234)., and equative The equative case is parallel to the English preposition as, such as in (Fortescue 1984: 170) Maalia Ammaaliatut ajortigaaq, ‘Maalia is as bad as Ammaalia’.. The following table shows the forms of the eight cases in Greenlandic in the singular and plural (Fortescue 1984: 206) with an indication of their meanings: singular plural approximate meaning core cases absolutive Ø -(i)t used for S and O ergative -(u)p -(i)t used for A oblique cases instrumental -mik -nik “with” allative -mut -nut “to” locative -mi -ni “at”, “in” ablative -mit -nit “from” procecutive -kkut -tigut “at”, “through” equative -tut -tut “as” Table 19: formal case marking in Greenlandic As can be seen in the table, the ergative and absolutive case have the same form in the plural, and therefore the difference between singular and plural nouns can only be seen in the verb inflection. I argued in Sakel (1998b) that plural NPs in Greenlandic involve neutral alignment instead of ergative alignment, i.e. S, A, and O are all marked by the same case. However, most structures are ergative in case marking and crossreference and thus, Greenlandic can be described as morphologically ergative (discussed in 3.3). Syntactically, however, Greenlandic follows an accusative pattern partly because the elements combined in switch reference are mainly subjects (but see Sakel 1998a). Possession is expressed in noun phrases by an inflectional ending on both possessor (POR) and possessum (PUM). Thus, Greenlandic can be characterized as ‘doublemarking’ possession in Nichols’ (1986) classification of languages in terms of head and dependentmarking. The possessum inflects for person and number of the possessor as well as for its own number. The possessor is marked by the suffix –p, which is the same exponent as that for the ergative case. The ergative case is therefore traditionally called ‘relative’ since it has both ergative and genitive functions. The following example from Janussen (1987: 23) shows a possessive construction: (83) Kaalip anaanaa KarlSG.POR motherSG.PUM.3SG.POR ‘Karl’s mother’ In this example, the –p marks the possessor, and the fused suffix –a marks both singularity of the possessum and the fact that the possessor is a third person singular. The present introduction to the structure of Greenlandic is far from complete. It is mainly concentrated on aspects that are relevant to the following discussion about voice and passive in Greenlandic. 6. Voice in Greenlandic 6.1 Different voice structures I have discussed theories of voice and passive in chapter 3 and 4, followed by a sketch of Greenlandic in the preceding chapter. Before going on to the description of the passive, I want to give a presentation of all identified voice structures in Greenlandic. These voice structures can be roughly divided into valence increasing and valence decreasing structures. While some verbs can be both used transitively and intransitively without any formal marking, most of the verbs are basically either transitive or intransitive. They can change transitivity, but then they have to be marked by either a transitivizer or a detransitivizer suffix. Detransitivizers are affixes which decrease the valence of a verb; transitivizers increase the valence of a verb. In Greenlandic, there are two valence decreasing structures, viz. passive and antipassive, and two valence increasing structures, viz. causative and applicative. Nichols (1992) divides voice structures further into structures affecting O and structures affecting A (see chapter 3). Table 20 shows how the Greenlandic constructions can be divided along these parameters: valence increasing valence decreasing A-affecting causative, others passive O-affecting applicative antipassive Table 20: voice structures in Greenlandic In this way, Greenlandic has morphological expressions of all the voice structures included in Nichols’ (1992) definition of voice. In terms of Klaiman (1991) who divides voice structures into derived voice, basic voice, and pragmatic voice (discussed in 3.2), Greenlandic seems to have only the derived voice. However, Klaiman’s definition does not include the two voice structures causative and applicative, both of which are found in Greenlandic. In the following sections, I want to introduce the different voice structures in Greenlandic. Here, I follow my definition of voice given in 3.4. 6.1.1 Active The active is zeromarked in Greenlandic. It is the form which is unmarked for voice and the other voice structures can be derived from it by addition of an affix. 6.1.2 Passive The passive is expressed by morphological means. There are several affixes that can have a passive meaning. Fortescue (1984) lists the following: neqar, tit, saa, sima, sariaqar and the combination sigi. These passive markers differ not only in their form, but also in meaning. The forms and functions of the passive morphemes will be comprehensively discussed in chapter 7. Here, I only want to mention that tit is called ‘pseudopassive’ by Fortescue (ibid.), since it has another function as a productive causative in Greenlandic. Furthermore, during my fieldwork, I encountered an impersonal passive with the affix toqar, which is not mentioned by Fortescue (1984). 6.1.3 Antipassive The antipassive is expressed by morphological means. Like the passive, the antipassive has several different affixes, which traditionally are called ‘halftransitivizing affixes’ in some descriptions of Greenlandic (e.g. Fortescue 1984). The affixes are: (s)i Bittner (1987) calls this one (ss)i., ller, nnig, and si. The affixes (s)i and si are diachronically distinct, and also synchronically their forms differ in certain surroundings. There is a fifth kind of antipassive which is expressed without an overt antipassive affix on the verb. In my definition of voice (see 3.4), I said I would only include structures with an overt formal marking, which means that, following my definition of voice, I should not include the antipassive with zero marking. However, I make an exception here. This structure has the same semantic content as the other antipassive forms. Furthermore, it is parallel in structure to the other types of antipassives since it involves instrumental marking of the demoted object. Therefore, I will include the antipassive with zero marking in my discussion. However, if there had not been any antipassive with formal marking, the type with zero marking would probably not have been identified as an antipassive by the grammarians of Greenlandic. The following examples from Bittner (1987: 1945) show the antipassive with si (85) and the antipassive with zero marking (87). The corresponding active clauses are in (84) and (86): (84) Jaakup ujarak tiguaa. JacobERG stone.ABS.SG take3SG.3SG.INDIC ‘Jacob took a stone.’ (85) Jaaku ujaqqamik tigusivoq. Jacob.ABS stoneINSTR.SG takeANTI3SG.INDIC ‘Jacob took a stone.’ (86) Jaakup illu sanavaa. JacobERG house.ABS.SG build3SG.3SG.INDIC ‘Jacob builds a house.’ (87) Jaaku illumik sanaØvoq. Jacob.ABS houseINSTR.SG buildANTI3SG.INDIC ‘Jabob builds a house.’ I have a short remark to the translation of the examples above: Fortescue (1984), Kalmár (1979) and others claim that antipassive clauses are indefinite while active clauses have a definite character. Bittner (1987) shows that there are actually many counterexamples to this claim and that the clauses usually do not differ in definiteness. I checked Bittner’s examples with my informants, and found out that the translations of both antipassives and actives were indefinite. When these clauses are translated definitely, una, ‘this one’ has to be added to the clause, as in the following example (DP): (88) Jaakup ujarak una tiguaa. JacobERG stone.ABS.SG this.one take3SG.3SG.INDIC ‘Jacob took the stone.’ According to Bittner, the different antipassive affixes are often paradigmatically exchangeable. Some verbal stems can take all five of the antipassive affixes while most of the verbal stems can take three or four of these affixes. This shows, according to Bittner (1987), that these are separate affixes and not suppletive forms as Fortescue (1984) claims. Furthermore, Bittner claims that there are aspectual differences between the antipassive affixes. The affixes si, (s)i and nnig imply imperfective aspect. According to Bittner (1987: 199), these affixes also “allow a frequentative interpretation with verbs whose transitive form obligatorily refers to just one event”. In addition, she claims that the antipassive marked by zero conveys an imperfective meaning. The remaining affix ller, on the other hand, “marks some kind of inceptive aspect” (Bittner 1987: 200). I think that Bittner’s ‘inceptive aspect’ is what is usually called ‘inchoative’. 6.1.4 Causative Before describing causatives in Greenlandic, I want to point out that the socalled causative mood (discussed in chapter 5) has nothing to do with valence changing structures discussed in the present section. Like the passive and antipassive, the causative is expressed by morphological means. I will discuss the causative in more detail, since it is directly relevant for passivization in Greenlandic. Fortescue mentions four affixes, all of which add an A to an intransitive or transitive clause. These affixes are traditionally called doubletransitivizers (e.g. by Fortescue 1984). One of these is the causative tit, which expresses both ‘let’ and ‘cause’ (see 3.1.7). The other forms are qqu,‘ask, want, tell to’, sar, ‘get to/ (try to) cause to’ and -tsaali-, ‘prevent from’. The latter I call tsaali (with the allomorphs taali, taaliur and tsaaliur) in the present study. In my fieldwork, it appeared that sar has disappeared from the spoken language and most of my informants did not know what sar meant. Therefore, I will not treat this affix in my study. The two other forms have the same structure as causatives even if they do not express causation. According to Payne (1997), forms expressing believe, say, want, ask, and let may have the same structure as causatives even if they do not have a causative meaning (discussed in 3.1.7). The affix qqu expresses two of the meanings mentioned by Payne: ‘ask’ and ‘want’, while the meaning of -tsaali-, ‘prevent from’ is not included in his list. However, since the verbal structures which contain the affixes tit, qqu and -tsaali- are exactly the same, they will be treated together under the heading of causative in the following discussion. When a causative affix is added to an intransitive base the underlying S becomes O and an A is added in the derived clause. In the following example from Fortescue (1984: 269), the causative affix tit is added to the intransitive clause in (89) and creates the transitive clause in (90): (89) Meeqqat aniput. childABS.PL go.out3PL.INDIC ‘The children went out.’ (90) Ilinniartitsisup meeqqat anisippai. teacherERG.PL childABS.PL go.outCAUS3SG.3PL.INDIC ‘The teacher sent the children out.’ ‘The children’, which is the subject of the active clause (89), is in object position in the causative clause (90). Furthermore, the causative clause has a new subject, which is ‘the teacher’. When a causative is applied to a transitive active clause, the object of the active clause remains the object in the causative clause. The causee is demoted to oblique position in the allative case while a causer with ergative case marking is added in the causative clause. In the following example, the affix qqu, ‘ask, want, tell to’ is added to a transitive active clause (DP): (91) Anaanap meeraq uffarpaa. motherERG.SG child.ABS.SG wash3SG.3SG.INDIC ‘The mother washes the child.’ (92) Ataatap meeraq anaanamut uffaqquaa. fatherERG.SG child.ABS.SG motherALL.SG washask.to3SG.3SG.INDIC ‘The father asks the mother to wash to child.’ Anaanap, ‘mother’, the subject of the active clause (91), is an oblique element in the allative case while the causer ataatap, ‘father’ is added in subject position in the causative clause. Both in the active and in the causative clause, the object is meeraq, ‘child’. The following example from Fortescue (1984: 269) in (94) shows the causative affix -tsaali-, ‘prevent from’ added to a transitive active clause, which is shown in (93) (DP): (93) Qimmit nerivaat. dogERG.PL eat3PL.3SG.INDIC ‘The dogs eat it.’ (94) Qimminut neritsaalivaa. dogALL.PL eatprevent.from3SG.3SG.INDIC ‘He prevented the dogs from eating it.’ In example (94), the core arguments are omitted in the causative clause; however, the structure of the causative in (94) is the same as the one in (92). The verb remains transitive. The causative derived from an intransitive active in Greenlandic has the structure which is shown in Table 10 in 3.1.7 for the morphological causative from intransitive actives in general. The structure of the causative from a transitive active, however, differs from the structure shown in Table 11 in 3.1.7 since the causee is not demoted to an indirect object, but to an oblique element. This may resemble the structure in Comrie’s (1989) passive analysis of the causative (discussed in 3.1.7). In that type the causee is not demoted to an indirect object which should have been the case according to Comrie’s principle of demotion (3.1.7). There are, however, two reasons not to treat the causative in Greenlandic in terms of Comrie’s passive analysis. The first reason is that causatives from intransitive actives follow Comrie’s principle of demotion since the causee is demoted to an object and not to an oblique element. The second reason is that Greenlandic has no formally marked indirect objects (discussed in 5). In this way, I argue that Greenlandic follows Comrie’s principle of demotion since the causee is always demoted to the next free slot on the hierarchy and this does not include indirect objects in Greenlandic. In my description of the structure of the causative above, the causative clause was described as transitive, containing at least causer and causee. However, all causative affixes can be attached to verbs which have crossreference to only one person, i.e. they are formally intransitive verbs. Usually, when a semantically transitive verb (such as a verb containing a causative affix) has an intransitive crossreference ending, the interpretation is reflexive, as can be seen in the following elicited example (DP): (95) Hanni toqutsaalivoq. Hanne.ABS dieprevent.from3SG.INDIC ‘Hanne prevents herself from dying.’ (e.g. curing herself by drinking herbal tea, etc.) Furthermore, one of these causative affixes can have a passive interpretation in intransitive use. The affix tit is often mentioned as a passive marker in Greenlandic and Fortescue (1984) calls it a ‘pseudopassive’ because it originally is a productive causative affix. I investigated this use of causative affixes in my fieldwork, and my findings will be discussed in 7.1.2 and 7.5 below. Now, I will turn to the discussion of the semantics of causatives. There are two structures which may express the difference between direct and indirect causation in Greenlandic (discussed in 3.1.7). According to Fortescue (1984: 84), the causative affix tit usually expresses direct causation. But when attached to nonagentive verbal stems, it expresses indirect causation. This can be seen in the following examples with the active clause in (96) and the causative clause in (97) (both from Fortescue 1984: 84, 85): (96) Matup assakka kiggippai. doorERG.SG fingermy.PL jam3SG.3PL.INDIC ‘The door pinched my fingers.’ (97) Assakka matumut kiggisippakka. fingermy.PL doorALL.SG jamcause1SG.3PL.INDIC ‘My fingers got jammed in the door.’ The translation of example (97) is rather ‘I got my fingers jammed in the door’ than the translation given by Fortescue. I suspect that Fortescue chose his translation to show that the causation is indirect, which means that the first person singular did not intentionally jam the fingers in the door. To see the difference between direct and indirect causation, compare example (97) with (90) above. In (90), the teacher sent the children out intentionally and the causation is direct while the causation in (97) is nonintentional and indirect. According to Fortescue (1984: 85), the difference between indirect and direct causation can also be seen in another structure, which involves more or less lexicalized forms. The causative affix tit functions as direct or indirect causative, depending on whether the preceding vowel causes the initial /t/ to assimilate regularly or not, as in the following example from Fortescue (ibid.): (98) Tikisippaa. come.toCAUS3SG.3SG.INDIC ‘He let it (e.g. an ordered item) come.’ (99) Tikitsippaa. come.toCAUS3SG.3SG.INDIC ‘He made him come.’ In (99), the initial /t/ of the causative affix tit assimilates regularly and the structure expresses direct causation. In (98), on the other hand, the /t/ assimilates in another way than expected with the verbal stem tiki, ‘come’ and the causation expressed is rather indirect. There are only a few forms where this difference between direct and indirect causation appears and they are highly lexicalized. The different translations ‘let’ and ‘cause’ of tit have nothing to do with indirect or direct causation, but vary according to context. 6.1.5 Applicative Like all the other voice structures in Greenlandic, applicatives are expressed by morphological means. According to Fortescue (1984), who calls them ‘transitivizers’, the applicatives in Greenlandic have the following forms and meanings: qatigi, ‘do together with’, (ss)ut(i), ‘do with/ for’, utigi, ‘have as means/ time of’, ssutigi, ‘have as reason for’, vvigi, ‘have as place/ time, etc. for’. According to Fortescue (1984: 269), there is another applicative affix –(s)i which is lexicalized and therefore unproductive. According to Fortescue, the affix -(s)i- functions as a productive antipassive marker (discussed in 6.1.3). The derivation of active clauses with applicatives can be seen in the following examples with qatigi from Olsen & Hertling (1988: 58) in (100) and (101) and (ss)ut(i) from Olsen & Hertling (1988: 91) in (102) and (103): (100) Sulivoq. work3SG.INDIC ‘He works.’ (101) Suleqatigaa The ending gaa is a contracted from givaa. The affix gi, ‘have as’ of the suffixes qatigi, utigi, etc., merges with the inflection vaa to the form gaa. worktogether.with3SG.3SG.INDIC ‘He works together with her.’ (102) Allappoq. write3SG.INDIC ‘He wrote.’ (103) Allaatigaa. writewith/about3SG.3SG.INDIC ‘He wrote about/ with it.’ Both active clauses (100) and (102) are intransitive. The subjects of these clauses are also subjects in the applicative clauses (101) and (103). In the applicative clauses, an object is added. In this way, the intransitive active clause is transitivized by the applicative. Whereas the causative affixes treated in the preceding section can be added to both intransitive and transitive verbs, applicatives can, according to Fortescue (1984: 270), only be added to intransitive verbs: “In fact all of the latter i.e. valence increasing affixes, J.S. except the ‘doubletransitive’ affixes (which may be attached to transitive bases) must follow intransitive/ intransitivized bases i.e. stems, J.S..” However, Fortescue also gives examples which contradict this claim elsewhere in his grammar. These counterexamples involve the applicative (ss)ut(i), ‘do with/ for’ of example (103). This can apparently be added to both intransitive and transitive verbal stems. In the following example from Fortescue (1984: 889), this affix is added to a transitive stem: (104) Niisi aningaasanik tunivaa. Niisi.ABS moneyINSTR.PL give3SG.3SG.INDIC ‘He gave (the) money to Niisi.’ or ‘He gave Niisi the money.’ (105) Aningaasat Niisimut tunniuppai. moneyABS.PL NiisiALL givefor3SG.3PL.INDIC ‘He gave the money to Niisi.’ The suffix up is a contracted form of (ss)ut(i) (see example (74) of chapter 5). In both examples, the subject ‘he’ is not expressed verb externally. In (104), the object is Niisi, while aningaasa, ‘money’, a plural noun, is an oblique element. In (105), however, aningaasat, ‘money’ is the object of the clause while Niisi is an oblique element. The use of (ss)ut(i), ‘do with/ for’ on transitive verbal stems is usually found with verbs which semantically take three arguments, such as ‘send’ or ‘bring’, but Fortescue (1984: 90) gives also examples with transitive verbs, such as ‘read’. 6.2 Voice in Greenlandic in terms of valence Passive and antipassive decrease the valence of a clause and causative and applicative increase the valence of a clause. Therefore, I divided voice structures in Greenlandic into valence increasing and valence decreasing structures in the beginning of this chapter. In fact, these structures do not always affect the valence. Therefore, the question is whether voice in Greenlandic can be treated in terms of valence or not. When a causative or applicative is added to an intransitive active clause, the valence is increased since a new coreargument is added to the clause. This can unequivocally be seen in the person markers of the crossreference endings in the verb: in the active clause, one argument is crossreferred, i.e. the clause is intransitive, while in the causative and applicative verbs two arguments are crossreferred, i.e. the clause is transitive. When a causative or applicative is added to a transitive active clause, however, the valence is not changed since only two arguments can be in core case and crossreferred in the verb. When there are three arguments in a Greenlandic clause, the third one will automatically be expressed in an oblique case or left out. An oblique element can be promoted to the object of a clause by an applicative construction, but at the same time the object of the active clause is demoted to an oblique element. In causatives, a new argument is added as a transitive subject while the causee is demoted to an oblique element. Thus, in both applicatives and causatives one argument is added to the clause while another one is removed. In this way, the grammatical valence is constantly that of two. Semantically, on the other hand, there are changes provoked by the applicative and causative. Even if the valence is not changed, the semantic roles assigned to the core arguments are new. Thus, an active clause with the semantic roles agent and patient as subject and object may become an applicative clause with the semantic roles agent and beneficiary as core arguments. Like causatives and applicatives sometimes do not increase the valence, passives do not always decrease the valence of a clause. This is the case in impersonal passives, which will be discussed in 7.4. Taking into account the structures where the valence is not changed, the question is whether to treat voice in Greenlandic in terms of valence change. All voice structures involve valence change to some degree. Applicative and causative change the valence of intransitive active clauses while passive and antipassive change the valence of transitive active clauses. Thus, all voice structures could be treated in terms of valence change. On the other hand, in many causatives and applicatives the valence is not changed. This leads to the conclusion that the voice system as a whole cannot be defined in terms of valence in Greenlandic. Rather, voice in Greenlandic results in a change of the structure and/or number of the arguments by formal marking of the verb. Semantically, this leads to a perspective change in the clause compared with the active (discussed in 3.4). 6.3 Combinations of voicemorphemes The voice morphemes discussed above do not only appear on their own but they are often found together with other voice morphemes in one word. In the following example from Fortescue (1984: 270), the causative tit is followed by the passive affix neqar: (106) Anisinneqarpugut. go.outCAUSPASS1PL.INDIC ‘We were sent out.’ In this example, the valence is first increased by the causative tit and then the valence is decreased again with the passive neqar. Furthermore, the intransitive active clause anivugut ‘we go out’ is modified by the causative, resulting in the clause anisippugut, ‘someone sends us out’. This clause is furthermore modified by the passive suffix in which case the causer, which was added by the causative, is demoted again and does not appear in the clause. There are several other combinations of voice morphemes. In the same way as tit, ‘cause, let’ other causatives can be passivized. This can be seen in the following example where the causative qqu, ‘ask, want, tell to’ is passivized with neqar (DP): (107) Jeanette Daavimiit atuaqquneqarpoq. Jeanette.ABS DavidABL2 In this example, the agent phrase is in the ablative 2, as I call it. According to Fortescue (1984), this case is a variant of the ablative with the ending –mit. readCAUSPASS3SG.INDIC ‘Jeanette was asked to read by David.’ As mentioned in section 6.1, the causative tit can also have a passive function. This affix can also be used instead of neqar in the above example also with no reported meaning difference (DP): (108) Jeanette Daavimiit atuaqqutippoq. Jeanette.ABS DavidABL2 readCAUSPASS3SG.INDIC ‘Jeanette was asked to read by David.’ However, my informant did not accept the clause when the causative affix qqu, ‘ask, want, tell to’ in (b) was replaced by tit in its causative function. According to Fortescue (1984: 270), two causative affixes may appear after each other in cases where the first affix is lexicalized and where the causation is indirect, as in his following example: (109) Inuulluaqqutippaa. live.wellCAUSCAUS3SG.3SG.INDIC ‘He sent his greetings to her (through another).’ According to Fortescue (ibid.), inuulluaqqu means ‘say hello to’ and this is the stem for the causative affix tit, ‘cause, let’. Otherwise, two causative affixes usually do not follow each other. Antipassives may appear together with causatives in one word. In the following example from Fortescue (1984: 270), the verb is derived by the antipassive affix ller, transitivized by the causative qqu, ‘ask, want, tell to’ and finally detransitivized again by the antipassive si: (110) Meeqqamut neqimik aalliqqusivoq. childALL.SG meat.INSTR.SG fetchANTICAUSANTI3SG.INDIC ‘She told the child to fetch (some) meat.’ Fortescue (1984: 270) has also the following example with an antipassive and two applicatives in the same clause: (111) Ilinnik oqaloqatiginnissutigaara. thouINSTR.PL speakAPPLANTIAPPL1SG.3SG.INDIC ‘I wanted to speak to you about it.’ In the following example from Fortescue (1984: 91), the applicative (ss)ut(i) is followed by the passive neqar: (112) Nassiunneqarpoq. sendAPPLPASS3SG.INDIC ‘It was sent.’ As can be seen in example (111), the same voice structures, in this case applicatives, may be added to a clause several times. In example (111), the transitivizing applicative is followed by a detransitivizing antipassive which is transitivized again by a new applicative. However, the same voice structures may also appear directly after each other, as can be seen in example (109) where two causatives stand side by side of each other. In my data, I do not have any clauses with an applicative together with a causative, or a passive together with an antipassive and neither are in the chapter about voice in Fortescue (1984). It may be that this combination is not possible since both either increase the valence or decrease the valence. However, it could be logically possible to have a causative together with an applicative since there are no restrictions on how many arguments can be added to one clause. Apart from these exceptions, all voice structures can appear together in the same clause. 6.4 Summary In the present chapter, I have described the voice structures antipassive, causative, and applicative in Greenlandic. I also dealt briefly with the passive. Following Nichols (1992), I divided these structures in terms of valence in the beginning of the present chapter (also discussed in 3.2). However, I concluded that not all of these structures provoke an alternation in the valence, such as impersonal passives or causatives and applicatives from transitive verbs. Therefore, the category of voice in Greenlandic cannot be described in terms of valence alone, although most voice structures involve a change in the valence. Even if the voice structures discussed in the present chapter are very different from each other, they have a number of common traits. They are all morphologically coded. Furthermore, all voice structures are represented by at least three affixes which usually differ semantically. In the following chapters, I will try to find an explanation for why there are so many different affixes for the passive. I will also discuss later the question why voice is represented by so many different affixes in Greenlandic. Before all that, I will describe the passive structure in more detail. 7. Passive in Greenlandic 7.0 Introduction In the preceding chapter, I have presented the voice structures antipassive, causative, and applicative in Greenlandic. The passive was discussed only briefly. In the present chapter, I will describe the different forms and uses of the passive in Greenlandic. In chapter 4, I defined the passive as a formally marked verbal category which has the function to inactivize the situation expressed by the clause. The inactivization of the situation results in a change of the status of the arguments in the clause. Thus, the patient or participant with another semantic role in object position in the active clause becomes the subject of the passive clause, while the role in subject position in the active clause, typically the agent, is demoted to an oblique element or is omitted in the passive clause. I will first look at the different forms of the passive followed by a description of the promotion and demotion of the arguments involved. After that, I want to describe impersonal passives. Finally, I will look at the development of the passive in Greenlandic, and I will briefly discuss the functions of this structure. This description of the passive in Greenlandic will build on the definition of passive I gave in chapter 4. 7.1 The formal marking of the passive The passive in Greenlandic is marked morphologically by affixes in the verb. All passive affixes are derivational affixes, and as expected they appear after the root and before the inflection in the verb (discussed in chapter 5). In terms of Haspelmath’s (1990) classification of passive marking (treated in 4.1.2), the passive in Greenlandic is marked by an additional stem affix. There are several different affixes which can express the passive. Fortescue (1984: 2656) mentions most of these affixes and I will take his division of passive affixes as the headings of the following sections. Nevertheless, I will mainly refer to the results from my own investigations in these sections. Apart from the six forms of the passive listed in Fortescue (ibid.), I encountered one additional passive affix during my fieldwork, which will be treated in section 7.4 below. The different forms of the passive will all be listed in the summary of the present chapter. 7.1.1 neqar The affix neqar is a productive passive morpheme and functions as a dynamic passive. It is built up by the abstract participle neq and the verbalizing affix qar (to be discussed in 7.6). The following example from Fortescue (1984: 265) shows an active clause (113) which is derived by the passive with the affix neqar in (114): (113) Inuit nanoq takuaat. peopleERG.PL polar.bear.ABS.SG see3PL.3SG.INDIC ‘The people saw the polar bear.’ (114) Nanoq (inunnit) takuneqarpoq. Polarbear.ABS.SG (people.PLABL) seePASS3SG.INDIC ‘The polar bear was seen (by the people).’ In the passive derivation shown by this example, the subject of the active clause (113) inuit, ‘people’ is demoted to an oblique element in the ablative case in the passive clause (114). The object of the active clause nanoq, ‘polar bear’, on the other hand, is promoted to the subject of the passive clause and the verb is formally marked by the passive affix neqar. The active clause is transitive with crossreference to both subject and object in the verb, whereas the passive clause is intransitive, having only one core argument in the verb. In this way, the passive affix neqar involves automatically detransitivization of the clause. Like in other languages, the demoted agent may be omitted from the passive clause. In fact, verb external core arguments can be omitted as well in Greenlandic so the omission of the agent phrase is not a structure unique to the passive. In this way, the examples (113) and (114) above could also be presented as the following elicited examples (DP): (115) Takuaat. see3PL.3SG.INDIC ‘They see it.’ (116) Takuneqarpoq. seePASS3SG.INDIC ‘It is seen.’ However, the omission of the agent phrase in a passive clause is not the same as the omission of core elements as verb external constituents, since the latter are still represented in the crossreference ending of the verb. The agent phrase in the passive, on the other hand, is totally omitted in (116) and the notion of a undergoer of the seeing is only implied in the clause. The passive affix neqar, forms a dynamic passive. It focuses on the action, as opposed to stative passives, which focus on the state of the undergoer of the action (discussed in 4.1.6). Apart from the stative affix saa, all the passive affixes in Greenlandic are dynamic. In section 7.1.3, I will discuss the difference between dynamic and stative passive in Greenlandic more thoroughly. 7.1.2 tit The passive morpheme tit with the allomorphs sit and tsit is called ‘pseudopassive’ by Fortescue (1984) because it is used mostly as a productive causative affix in Greenlandic. The passive use of tit, however, was already mentioned by Top (1727; 1986: 115): Endskiönt at Grönlænderne haver mange og adskillige ord og udtaler, som synis at kunde komme overeens med Verbo Passivo, saa befinder man dog, at saadanne icke ere regularia, hvilke ieg og derfore icke anförer, men det ord, som mand sickerst holder for Verbum Passivum, kommer af et andet, (…) sc. Pisipok; tj naar dette indlöber i andre ord, da udkommer Passivum eller dend rette Passivi bemerckelse, saalænge at intet suffixum kommer dertil, ex. gr. Ermiksípunga (…) : ieg toes og giöres toed (…), og da leggis gemeenligen disse particler til sc. Uaugnut : af mig, Illignut : af dig (…); men dersom samme Passivum antager noget Suffixum, da iföris det denne bemerkelse, ex. gr. Ermiksipaukit : ieg giör v. kommer dig til at toes (…). Dend forskiel, som og herefter er at merke, at nogle Passivi haver Sipok, andre Tipok til enden, veed ieg ingen synderlig rigtighed for, uden saa er, at Præteritum Indicativi maa giöre noget hertil (…). Free translation into English: Even if the Greenlanders have many and different words and pronunciations which may be compared to Verbo Passivo, these are not regular, and therefore I do not mention them; but the word which certainly is a Verbum Passivum comes from another one, the socalled Pisipok; because when this is included in other words, the meanings of these words are passive, as long as no other suffix is added, e.g. Ermiksípunga, ‘I was washed in the face’, and usually these particles are added Uaugnut, ‘by me’, Illignut, ‘by you’; however, when the same passive gets a suffix, it gets the following meaning, e.g. Ermiksipaukit ‘I make you wash your face’. I do not know any explanation for the difference between Sipok and Tipok in the end of passives, apart from that Præteritum Indicativi may have to do with it. Top mentions the passive use of the verb pisippoq, ‘be done, be captured’, which ‘can be added to other verbs, resulting in a passive meaning’. Of course, it is only the passive ending sippoq which is added to the other verbs and not the whole verb pisippoq. Top mentions the causative use of this affix when ‘standing with a suffix’, i.e. when having a transitive crossreference ending. Finally, he wonders about the different allomorphs of this affix. These allomorphs have primarily to do with the verbal stem, depending on whether it is a vowel stem or an r, or a consonant stem (discussed in 5). Since this affix already is mentioned by Top (ibid.), its use as a passive marker is not a new invention in the Greenlandic language. According to Fortescue (1984) and my own fieldwork experience, the affix tit is most commonly used in the spoken language to express the passive where it typically replaces the affix neqar. My informants usually used the titpassive when asked to translate a passive clause to Greenlandic. When asking my informants about the difference between a clause with a neqar passive and a passive with tit they usually said that there was no difference at all. However, in further investigations of these affixes it appeared that tit is structurally and functionally different from neqar. I will first give an example of the causative use of this morpheme (see also 6.1.4) and then turn to its passive function. Fortescue (1984: 269) shows the causative use of tit in the following pair of examples: (117) Meeqqat anipput. childABS.PL go.out3PL.INDIC ‘The children went out.’ (118) Ilinniartitsisup meeqqat anisippai. teacherERG.SG childABS.PL go.outCAUS3SG.3PL.INDIC ‘The teacher sent the children out.’ In the intransitive active clause (117), meeqqat, ‘children’, is the subject. When the causative affix tit is added to the verb, the clause is transitivized, as can be seen in the crossreference ending of (118). The new argument in the clause is ilinniartitsisu, ‘teacher’ which is in A position and has A case marking. The constituent in S position in the active clause, meeqqat, ‘children’, is moved to object position in the causative clause. The following examples show the passive use of the affix tit; I give the active clause in (119), which is an elicited example (DP), and the passive clause in example (120) from Fortescue (1984: 265): (119) Qimmip Hansi kiivaa. dogERG.SG Hans.ABS bite3SG.3SG.INDIC ‘The dog bites Hans.’ (120) Qimmimut kiisippoq. dogALL.SG bitePASS3SG.INDIC ‘He got bitten by a dog.’ The subject of the active clause (119), qimmeq, ‘dog’, is demoted to an oblique element in the allative case in the passive clause (120) while the object of the active clause (119), Hansi, is promoted to the subject of the passive clause. In the example of the passive clause (120), the subject Hansi is not expressed verbexternally but it appears in the crossreference ending of the verb. When tit is used as a causative, a new argument is added to the clause while in the passive use of tit, an argument is taken away from the clause. Such a link between causative and passive may seem strange at first sight. However, structures expressing both the passive and the causative are found in languages of different parts of the world. According to Haspelmath (1990) (see 4.4.3), passives can develop from causatives via reflexive causatives, and therefore the passive and the causative may have the same forms. Nichols (1992: 157) gives the following example of a reflexive causative (121) and a passive (122) in English: (121) He got himself shot. (122) He got shot. This example shows that the passive (122) and the reflexive causative (121) are close since they have the same outcome. In both (121) and (122), the outcome is ‘he is shot’. In the reflexive causative, however, the subject has control over this outcome while the subject of the passive clause (122) has no control at all. Furthermore, the difference between the reflexive causative and the passive is the agency of the verb and the presence of the reflexive pronoun. In Greenlandic, the process slightly different. The reflexive causative is expressed by causative marking on a verb which is inflected intransitively rather than transitively. As in other structures in Greenlandic, reflexives can be expressed by an intransitive inflection on an otherwise transitive verb (discussed in chapter 5). The addition of a reflexive pronoun to the reflexive clause is optional in this language. The following example from Fortescue (1984: 265) shows a reflexive causative without a reflexive pronoun: (123) Nanoq takutippoq. polar.bear.ABS.SG seePASS/CAUS3SG.INDIC ‘The polar bear let itself be seen.’ According to Fortescue, this example may also be translated as ‘the polar bear was seen’. Thus, the reflexive causative in Greenlandic can at the same time be interpreted as a passive. The difference between these two interpretations is that the reflexive causative has an agent as its subject whereas the subject of the passive interpretation is a patient, and is therefore inactive. The above example (123) becomes unequivocally reflexive when a reflexive pronoun is added to the clause, as in the following elicited example (DP): (124) Nanoq imminit aallaarsimaatunut takutippoq. polar.bear.ABS.SG REFLABL hunterALL.PL seeCAUS3SG.INDIC ‘The polar bear makes himself be seen by the people in the hunting camp.’ The subject of the reflexive causative construction usually has the semantic role of an agent, as it initiates and controls the action. In 4.4.3, I discussed the development of the getpassive in English by Keenan’s (1985) examples (6871). According to Keenan, the passive which developed from a reflexive causative can by analogy be extended to all other clauses in the language. Thus, also clauses with an inanimate subject may be passivized by the getpassive in English, as in the following example, repeated from 4.4.3: (125) The window got broken. In Greenlandic, however, the passive with tit can only be applied to clauses with animate subjects which can also be the causer of a reflexive causative construction. This can be seen in the following elicited examples, where meeraq, ‘child’ is accepted as the subject of the construction, while nasaq, ‘cap’ is not accepted (DP): (126) Meeraq igitippoq. child.ABS.SG throw.outPASS3SG.INDIC ‘The child was thrown out.’ (127) *Nasaq igitippoq. cap.ABS.SG throw.outPASS3SG.INDIC ‘The cap was thrown out.’ In the examples above, I only gave the passive translation even if these examples could also be translated as reflexive causatives. From the point of view of the passive translation, there is no structural reason why (126) is acceptable and (127) is not. However, from the point of view of the reflexive causative the difference between these two clauses can be seen. The reflexivecausative translations of these examples are (126) ‘the child made himself be thrown out’ and (127) ‘*the cap made itself be thrown out’. In (127), the reflexive causative is logically impossible since the subject is inanimate and an inanimate NP cannot do anything to itself. Thus, it seems that the use of the passive with tit is restricted to the conditions where the causative reflexive can be applied. This restriction is not found in the passive with neqar, as can be seen in the following elicited examples (DP): (128) Illu takuneqarpoq. (129) *Illu takutippoq. house.ABS.SG seePASS3SG.INDIC house.ABS.SG seePASS3SG.INDIC ‘The house was seen.’ ‘The house was seen.’ (130) Inuk takuneqarpoq. (131) Inuk takutippoq. human.ABS.SG seePASS3SG.INDIC human.ABS.SG seePASS3SG.INDIC ‘The human was seen.’ ‘The human was seen.’ The difference between NPs which can be the subject in the passive use of tit and those which cannot seems to involve animacy, but does not strictly lie in animacy. Usually, humans and ‘higher’ animates, such as dogs, cats and polar bears are accepted as subjects of this passive construction, whereas NPs lower on the animacy scale (discussed in 2.2) such as mosquitoes, fish or inanimates, such as caps and houses, cannot be the subject in a passive structure with tit. My informants described the NPs that can be subjects in this construction as individuals who can have their own free will, i.e. they are intentional and may have some control over the situation in other clauses. Even if mosquitoes are animate, they are not seen as acting by their own free will as opposed to animals close to humans, such as dogs. Furthermore, mosquitoes are not individuals but there are hundreds of them at the same time as opposed to pets which often even have a name. Thus, the restrictions on the subject of this passive construction follow the same constraints as for causers in a causative construction. Usually only intentional and controlling NPs can be causers in a causative construction, apart from forces that act unintentionally (discussed in 2.2). In children’s television programs or stories about a ‘living’ house or a very special mosquito, however, the passive forms with tit are accepted with ‘lower’ animates or inanimates. In these cases, the house or the mosquito would be something close to humans and the following example would be possible (DP, EQ): (132) Illu Piitamut sanatippoq. house.ABS.SG PeterALL.SG buildPASS3SG.INDIC ‘The house was built by Peter.’ (e.g. the house has feelings and knows that it was built) The meaning of this example is that the house knows and feels that it was built. Since the passive with tit has the implication that its subject is a living and feeling creature, inanimate and ‘lower’ animate NPs cannot be subjects in this construction, or they are interpreted as animates. I was interested in the question whether these restrictions on the subjects of the tit passive may loosen. Since my main informants were 38 and 43 years old, I questioned six younger people, all about 18 years old, using a monolingual questionnaire in order to find out which forms were possible. Unfortunately, my data are not very consistent, since I only investigated this phenomenon by relying on reported language use and I do not have spontaneous examples. However, I want to present my results of this investigation. Most of these young speakers accept lower animates as subjects of the passive construction with tit while nearly all of them reject the use of inanimate NPs. This may point to a development towards a broader use of the tit passive. However, such a development seems peculiar in the light of Top’s (1727) comments about the use of the tit passive which show that this structure has been stable for many years. Finally, I want to say a few words about the difference in meaning between the neqar passive and the tit passive. Often, these two structures have the same meaning but in some cases they may differ in meaning, as can be seen in the following examples with the tit passive (133) and the corresponding passive clause with the affix neqar (134) (DP): (133) Qimmerput toqutsippoq. dogPOSS killPASS3SG.INDIC ‘Our dog got killed.’ (e.g. because we asked someone to do it) (134) Qimmerput toqunneqarpoq. dogPOSS killPASS3SG.INDIC ‘Our dog got killed.’ (e.g. by a car) In both examples, a passive clause expresses ‘the dog got killed’. In the clause with the passive morpheme tit (133), however, the meaning is that someone planned the killing of the dog while in the clause with the affix neqar the dog was killed by chance. The explanation of this difference in the translation may be that the affix tit still may have a slightly causative meaning in passive use. However, I found such a causative meaning only in this single example, having tested it with many other clauses where the -tit- passive and the -neqar- passive did not differ in meaning. 7.1.3 saa The stative passive affix saa consists of the passive participle saq and the copula u, ‘be’. I first want to look at the passive participle, then I will discuss the stative passive. The affix saq is not really a passive affix but rather the marker for the passive participle in Greenlandic. Because one of the passive markers in this language can be formed by extending this participle, I will treat saq briefly in this section. The affix saq has the allomorphs taq and gaq  in the singular and kkat  and taq in the plural. It is a nominalizing affix, which means that it derives a nominal from a verbal stem. Other nominalizers are, according to Fortescue (1984), the abstract participle neq, the intransitive participle soq with the allomorphs toq and tsoq, and the unproductive active participle si with the allomorphs tsi, (r)ti, seq and teq. The passive participle differs from the other nominalizers in that it can only be attached to a transitive verb, referring only to the object of this verb. In this way, the participle has a passive sense. In the following examples from Brochmann (1993: 159), the active clause in (135) is derived by the passive participle in (136): (135) Nassarpaa. bring.along3SG.3SG.INDIC ‘He brings it along.’ (136) Nassataq. bring.alongPASS.PART ‘Something, which is brought along/ luggage’ In example (136), only the object ‘it’ of the active verb in (135) is referred to. The agent of (135) is implied in the passive participle, but it is not overt in (136). This demoted agent, when overt, is expressed as a personal possession marker on the passive participle, as in the following example from Fortescue (1984: 52): (137) Pisaa akivoq. In this example, the passive participle affix has the form sa, because the following possessive ending is truncating (see chapter 5). ‘address’PASS.PARTPOSS reply3SG.INDIC ‘The one he’d addressed replied.’ The literal translation of this example would be ‘his addressed one replied’, i.e. the agent of the verb derived by the passive participle is expressed as the one possessing the passive participle. Since the participle is a nominal form, this kind of nominal inflection is expected. I will now turn to the passive affix saa, which consists of the passive participle and u, ‘be’. Thus, the nominalized verb becomes verbal again. The following example from Fortescue (1984: 266) shows the use of this derived participle: (138) Asasaavoq. lovePASS.PARTbe3SG.INDIC ‘She is loved.’ From now on, I will treat this derived form (saq + u > saa) as a single passive affix. The affix saa has only stative meaning, which can be seen in its focus on the state of the undergoer of the action. The dynamic passive affixes neqar (treated in 7.1.1) and tit (discussed in 7.1.2), on the other hand, focus on the action itself. The difference between stative and dynamic passive can be seen in the following examples (HK): (139) Juullip orpia ikinneqarpoq. ChristmasPOSS treePOSS lightPASS3SG.INDIC ‘The Christmas tree was being lighted.’ (140) Juullip orpia ikitaavoq. ChristmasPOSS treePOSS lightPASS3SG.INDIC The Christmas tree was lighted. In example (139), the focus is on the action of the lighting of the Christmas tree, whereas example (140) focuses on the state of the Christmas tree. In stative passives, the agent phrase can be expressed in the same way as in dynamic passives, which can be seen in the following elicited example (HK): (141) Meeraq angutimit titartagaavoq. child.ABS.SG manABL.SG drawPASS3SG.INDIC ‘The child was drawn by the man.’ For my analysis of the stative passive in Greenlandic below, I want to recall the difference between stative and dynamic passive in the example from German from Palmer (1994: 138), from 4.1.6 (example (55) and (56)): (142) Als ich um fünf kam, when I.NOM at five come.PAST.1SG war die Tür geschlossen, … be.PAST.3SG DET.F.NOM door close.PERF.PART ‘When I came at five the door was shut, …’ (143) … aber ich weiss nicht, wann sie geschlossen wurde. but I.NOM know.1SG not when 3SG.F close.PERF.PART become.PAST. ‘… but I do not know when it was shut.’ Here sein, ‘be’ is used for the stative passive and werden, ‘become’ is used for the dynamic passive. The passive in German is an analytic construction marked by auxiliaries, whereas the passive in Greenlandic is strictly morphological. Apart from these differences, the stative passive in Greenlandic resembles the one in German: both languages use a participle and the copula element to form the stative passive. In this way, the structure of the stative passive in German and Greenlandic is formally built up as a copula clause with a nonverbal predicate. In both Greenlandic and German, this copula element is also used for nonverbal predication. Compare example (138) above with the following elicited example (DP): (144) Arnaavoq. womanbe3SG.INDIC ‘She is a woman.’ (German: ‘Sie ist eine Frau’) In both (138) and (144), the nominal element is derived by the copula u, ‘be’, followed by the inflectional ending. In this way, the stative passive has the same structure as a copula clause. This is one of the reasons why Siewierska (1984) argues that stative passives should not be analyzed as passives but as actives (discussed in 4.1.6). Nedjalkov & Jaxontov (1988) call stative passives ‘objective resultative constructions’ since the stative passive focuses on the result of a previous action from the view of the undergoer of this action. According to Fortescue (1984: 266), saa is not as frequent or productive as neqar, which I also could see in my fieldwork data since saa was not accepted in some cases. 7.1.4 sima According to Fortescue (1984), the derivational affix sima usually expresses perfective aspect, but in some cases it can have a passive meaning as well. In the concluding section of this chapter (7.1.8), I will analyze this affix as something else than a passive affix. However, I first want to look at the ‘passive’ meaning which sima has according to Fortescue. The affix can be added to both transitive and intransitive verbs and differs in this way from the passive affixes discussed above which can only be added to transitive verbs. Inherently transitive verbs can only be transitive, unless they are marked by a detransitivising affix. When sima is added to inherently transitive verbs, however, they can be used transitively or intransitively. In intransitive use, the subject is nonagentive, i.e. the underlying object of the transitive verb is used as the subject of the intransitive verb. When sima is added to an intransitive verb, the subject of this verb becomes inactive. This can be shown by taking a verb like igiC as an example. The verbal stem igiC has two different meanings, depending on transitivity. As an intransitive verb, igiC means ‘to shoot, to make a shot’, and as a transitive verb it means ‘to throw away’. In the following example, the interpretation of the verb is ‘throw away’, which means that the corresponding active clause is transitive. The affix sima (in its ‘passive’ sense) is added to this transitive active verb (HK): (145) Nasaq igissimavoq cap.ABS.SG throw.awayPERF3SG.INDIC ‘The cap was thrown away.’ The verb in (145) is passive. This clause cannot be an active clause itself since the subject nasaq, ‘cap’ is inanimate and thus cannot be the agent of the throwing. Being formally marked by the affix sima, the clause in (145) resembles other passive clauses in structure. However, sima does not always involve passivization of the clause. In the following example, the verb igiC, ‘shoot’ is used in its intransitive sense together with the affix sima which expresses perfective aspect (HK): (146) Angut appamut igissimavoq. man.ABS.SG aukALL.SG shootPERF3SG.INDIC ‘The man shot at an auk.’ In this example, the verb igiC is used intransitively, as it has the meaning ‘shoot’ and sima expresses the perfective aspect. The animal which is shot at, appa, ‘auk’, is expressed as an oblique element in this example, since the verb is strictly intransitive. Thus, example (146) cannot be understood as an antipassive. The corresponding clause to example (146) in the imperfective aspect would be angut appamut igippoq, ‘the man shoots at an auk’. In example (146), the only possible interpretation is that the man shot at an auk and not that the man is thrown away. However, ambiguities may appear in clauses where sima can be interpreted both as a marker for the perfective aspect and as a marker for the passive. This is the case in the following example of an intransitive clause with an animate subject (HK). (147) Hansi silami qinersimavoq. Hans.ABS weatherLOC.SG look.aroundPERF3SG.INDIC ‘Hans looked for something outside’ or ‘Hans was looked for outside.’ This clause is ambiguous since the subject Hansi can be interpreted as an agent who is looking around for something, or as a patient who is looked for. The function of sima as a passive affix or an affix expressing perfective aspect depends therefore largely on the context in which it is used. For an unequivocal passive meaning of the clause in (147), the speaker may prefer to add the affix neqar to the verb, as in the following example (HK): (148) Silami qinerneqarsimavoq. weatherLOC.SG look.aroundPASSPERF3SG.INDIC ‘He was looked for outside.’ In this example, the passive affix neqar passivizes the clause while the function of sima is rather to indicate perfective aspect. Thus, this clause is an example of a neqar passive instead of a passive with sima. 7.1.5 sariaqar In the same way as sima above, the derivational affix sariaqar, ‘must’ with the allomorph tariaqar can have a passive meaning when added to inherently transitive verbs. In the following examples from Fortescue (1984: 2656) sariaqar is added to a transitive verb without affecting the valence in (149) and to the same verb which is used intransitively and treated as a passive in (150): (149) Ikiortariaqarpaat. helpmust3PL.3SG.INDIC ‘They must help him’ (150) Ikiortariaqarput helpmust3PL.INDIC ‘They must be helped’ In (149), the affix sariaqar marks an active transitive verb, adding the meaning ‘must’ to the clause. In (150), the affix is added to the same verb, also with the addition of the meaning ‘must’ but in this case the verb is detransitivized and has a passive structure. As opposed to the affix sima (7.1.4), sariaqar cannot be added to intransitive verbs, entailing a passive meaning. This can be seen in the following examples with the two different stems of the verb ‘draw’: the intransitive stem is titartaa, ‘draw’ and the transitive stem is titartar, ‘draw something’ (HK): (151) Titartaasariaqarpoq. draw(intr.)must3SG.INDIC ‘He must draw.’ (152) Titartartariaqarpoq. draw(tr.)must3SG.INDIC ‘He must be drawn.’ In this way, when added to transitive verbs used intransitively, the affix sariaqar may function as a passive, keeping its meaning ‘must’, while with intransitive verbs and verbs which can be both intransitive and transitive it expresses only ‘must’. 7.1.6 si/ gi Fortescue (1984) mentions a final passivelike morpheme which is an agentive nominalization with the form si followed by gi, ‘have as’. The following example from Fortescue (1984: 266) shows this construction: (153) Ajoqersotigaara teachAGNhave.as1SG.3SG.INDIC ‘I was taught by him.’ (I had him as a teacher) This structure differs from other passives in that the ‘passive’ clause in (153) is transitive. Furthermore, the agentive nominalization with si is unproductive and in this way, the construction with sigi is limited to a few cases where a verb is nominalized by the agentive si. Since this affix is not productive and restricted to very few cases, I will not treat it as a formal marking of the passive. 7.1.7 Concluding section: the formal marking of the passive In the preceding sections, I have described different formal markings of the passive according to their use and their functions. Some of the affixes discussed above mark only the passive while the majority of the affixes mark other structures along with the passive. The affixes that only mark the passive are neqar (7.1.1) and saa (7.1.3). The stative passive, which is represented by the affix saa in Greenlandic, however, is not always analyzed as a type of passive, as I discussed in 4.1.6 and 7.1.3. Thus, perhaps this form of the passive should not be included as a strictly passive marker after all. Therefore, only one affix, neqar, has strictly passive use. The remaining affixes which mark other structures along with the passive, are tit (7.1.2), sima (7.1.4), and sariaqar (7.1.5). In the present section, I want to analyze these affixes, to find out whether they are passive markers or not. The affix tit is a productive causative marker, but can also function as a passive. This affix is commonly used to mark the passive in the spoken language alongside the affix neqar. However, there are restrictions in the use of tit, which are not found with neqar. According to Haspelmath (1990) and Keenan (1985) (see 4.4.3), a causative affix can develop into a passive affix with reflexive passives in the intermediate stage. In Greenlandic, this development from a reflexive causative to a passive appears to have taken place. However, the structure is restricted to clauses with subjects which can function as the causer in a reflexive causative. Furthermore, this development from a causative to a passive has been going on for at least 270 years since Top (1727) first mentioned the use of tit in a passive in Greenlandic (discussed in 7.1.2). There are two ways to deal with the tit element: the first possibility is to classify passives with tit as reflexive causatives which are close in meaning to passives. The second possibility is to call this structure a passive. I claim that this construction is a passive structure and can be analyzed as distinct from the reflexive causative. All passive constructions with tit in my material can be described as directly derived from a reflexive causative clause. Furthermore, these passive constructions have the same outcome as the causative clause. However, the tit passive is frequently used in the spoken language, displacing other passive affixes, such as neqar. Furthermore, this affix is understood as a passive rather than a causative reflexive marker in metaphoric expressions, as I have shown in example (132) in (7.1.2). The translation of this example is not ‘the house made itself built by Peter’, but rather ‘the house was built  feeling it itself’. Thus, the restrictions on the subjects in this passive structure may have developed out of the restrictions on the causative reflexive. Nevertheless, the passive construction has taken over the restrictions for its own use, since they do not have anything to do with the reflexive causative in all cases. The derivational affixes sima, marking the perfective aspect, and sariaqar, ‘must’ are normally active derivational affixes but they can express a passive meaning in some cases. Usually, inherently transitive verbs cannot be used intransitively without being marked by a detransitivizing affix. However, when sariaqar or sima are added to such an inherently transitive verb, the verb may be used transitively or intransitively together with the meaning of the derivational affix. So they have a detransitivizing or passivizing function when used intransitively and the subject is nonagentive, i.e. it is coreferent with the object of the transitive verb. In this way, verbs derived by sariaqar or sima resemble non-agentive verbs, i.e. verbs which can be transitive and intransitive, having the object of their transitive form as the subject of their intransitive form (see chapter 5). Because of the nonagentive use, this structure resembles a passive. These affixes, however, are not restricted to intransitive use since they can also be used with transitive verbs transitively without any such change. Thus, the affixes sima and sariaqar are affixes which form non-agentive verbs, which can be used transitively and intransitively, out of inherently transitive verbs. They are not really passive affixes. Furthermore, both sima and sariaqar can occur together with passive affixes (example (148) in 7.1.4) which other passive affixes cannot. The affix sima may also be added to intransitive verbs. This is not a typical trait of the passive since usually only impersonal passives are used with intransitive verbs. The affix sima is not an impersonal passive since it keeps the core argument from the underlying intransitive clause. Rather, sima can derive nonagentive intransitive verbs from agentive intransitive verbs. Nevertheless, sima may be added to agentive intransitive verbs without making these nonagentive. The function of sima together with intransitive verbs thus largely depends on the context. This is another reason why sima is not a passive affix, but an affix which can make nonagentive verbs out of inherently transitive verbs. Having analyzed the affixes given by Fortescue (1984), I conclude that only three of these actually function as passives. These affixes are neqar, tit, and saa, and not -sima-, -sariaqar- and -si-gi-. I will now look at the way these passive affixes are attached to the verb. In the beginning of this chapter I claimed that all formal markers of the passive in Greenlandic are additional stem affixes in the sense of Haspelmath (1990) (discussed in 4.1.2). This means that the affix is added directly to the verbal stem before any tense, aspect, and person marking. In the following elicited examples, I show that the passive affixes are additional affixes attached to the stem in Greenlandic. In (154), the passive marker neqar occurs together with aspect and person marking (HK) and in (155) the passive affix tit occurs together with tense and person marking (EQ): (154) Piitaq politimit ujaneqarsimavoq. Peter.ABS policeABL.SG search.forPASSPERF3SG.INDIC ‘Peter was wanted by the police.’ (155) (Aqagu) Ole misissortinniarpoq. (tomorrow) Ole.ABS examinePASSFUT3SG.INDIC ‘Ole will be examined (tomorrow).’ These examples show that the passive affix precedes tense, aspect, and person marking in Greenlandic. In example (154), the passive marker neqar is followed by the marker for perfective aspect sima and the fused person, number, and mood marker voq. In example (155), the passive marker tit is followed by one of the markers for the future tense niar and the person, number, and mood marker poq. 7.2 The promoted element Languages differ in which arguments can be in subject position in a passive clause (treated in 4.1.3). Usually, the object of an active transitive clause becomes the subject in the corresponding passive clause. In some languages, however, also oblique elements may become the subject of the passive clause. Either these elements are directly promoted, or they are brought into object position by the applicative voice before they are promoted to the subject of a passive construction. The latter is the case in Greenlandic. In Greenlandic, only objects in the active clause can be promoted to subject position in a passive clause. Thus, an oblique element first has to be brought into object position. This is carried out by the applicative (discussed in 6.1.5), as in the following example from Fortescue (1984: 91) with the transitive verb nassiC, ‘send’. The verb nassiC, ‘send’ has the semantic roles of the agent in subject position and the recipient in object position. The patient, i.e. the thing being sent, is expressed by an oblique element in the clause, here not overtly present. In (156), the passive is formed from this verb and thus, the recipient object is in subject position. In (157), the subject of the passive clause is the patient which is brought into object position by the applicative (ss)ut(i) (un), ‘have as means/ time of’, in which way it is made accessible for passivization (see 6.3. for the same example (112)): (156) Nassinneqarpoq. sendPASS3SG.INDIC ‘He was sent (something).’ (157) Nassiunneqarpoq. sendAPPLPASS3SG.INDIC ‘It was sent.’ 7.3 The demoted element The subject of an active transitive clause is demoted in the corresponding passive clause. The demotion either results in the omission of the former subject, or the subject becomes an oblique element in the passive clause. This oblique element is usually called ‘agent phrase’ (discussed in 4.1.4). As I discussed in 3.1.7, Comrie (1989) points out that the agent phrases of passive constructions often resemble agent phrases of causatives. In Greenlandic, the causative affix tit can function as a passive and also the two other causative affixes may have a passive meaning in some cases, as will be discussed in 7.5 below. For these reasons, I will combine the description of the agent phrases in passive and causative structures in the present section. According to Fortescue (1984), the passive construction formed by the affix neqar has its agent phrase in the ablative case while the agent phrases of causatives usually are in the allative case. Fortescue mentions, however, that in some dialects the agent phrases of causatives may be in the instrumental or ablative case. I investigated the case markings of the agent phrases in different types of passives and causatives by interviewing informants of different ages and dialectal groups. The following table shows the different passive and causative morphemes in relation to the three cases found in the agent phrases: passive causative neqar saa tit (intr) tit, qqu, tsaali (intr) tit, qqu and tsaali (tr) allative x x x x ablative x x x x instrumental (x) (x) (x) (x) Table 21: the cases of agent phrases in different passive and causative constructions In my fieldwork data, the agent phrase in the passive construction with neqar is not always in the ablative case but it may also be in the instrumental. This seems to be a conservative variant and depends on the speaker. One of my informants from the town Maniitsoq on the west coast of Greenland used the instrumental in all the cases where it was possible while others did not accept the marking of agent phrases by this case at all. Some of the latter informed me that the instrumental is only used by older people. However, the informant who used the instrumental all the time was not old. He may thus speak a conservative dialect. With causatives, however, the agent phrase is always in the allative case. Eight informants of very different ages, speaking different dialects all allow only allatives and no other case markings for the agent phrase in causatives. Thus, I have no confirmation for Fortescue’s (ibid.) claim that causatives may have agent phrases in the instrumental or ablative case. The marking of agent phrases with stative passives and intransitively inflected causatives varies. Both the ablative, allative, and instrumental case are possible. The use of the instrumental case seems again to depend on the speech variant of the informant since only few informants use this case to mark the agent phrase, while most of my informants do not accept the use of the instrumental case. The latter informants use the ablative and allative case with roughly the same frequency. Most speakers prefer one form, but they accept the use of the other cases as well. There are no indications that the use of one form as opposed to the other has to do with dialectal or age differences in the investigation with my informants. It seems rather to be the choice of the speaker. Thus, the case of the agent phrase in Greenlandic seems not to indicate the affectedness or control of the demoted element, as reported from some other languages (treated in 3.1.7). The reason for the use of different cases in agent phrases may be found in the semantics of these cases. The ablative case corresponds to a preposition, such as ‘from’ in English. The passive with neqar often has an agent phrase in the ablative case. In this way, the ablative case may indicate that the action comes from the element in the agent phrase. The allative case, on the other hand, corresponds in meaning to the English preposition ‘to’. This case is always used to mark the causee in the agent phrase of a causative construction. Since the causee is the one receiving a causation from the causer, the action goes to the causee. Therefore the use of the allative case to mark the agent phrase is not surprising. As mentioned, the case marking of the causative affixes used intransitively varies. In the case of the affix tit which has both a reflexive causative and a passive use when inflected intransitively, the case of the agent phrase seems neither to reflect the use of tit as a reflexive causative or as a passive. This can be seen in the form of the reflexive pronoun, which is parallel in structure to the agent phrases discussed here (DP, compare with 7.1.2): (158) Nanoq imminut takutippoq. polar.bear.ABS.SG REFLALL seeCAUS3SG.INDIC ‘The polar bear made itself be seen.’ (159) Nanoq imminit takutippoq. polar.bear.ABS.SG REFLABL seeCAUS3SG.INDIC ‘The polar bear made itself be seen.’ (160) Nanoq imminik takutippoq. polar.bear.ABS.SG REFLINSTR seeCAUS3SG.INDIC ‘The polar bear made itself be seen.’ The reflexive pronoun may be in all three cases in the same way as the agent phrase of a passive with tit may be in all three cases. This use may be explained by the meaning of the allative and ablative case. To say ‘he makes himself be seen to the people’ or ‘from the people’ has the same meaning and therefore both cases are possible in agent phrases of reflexives. In the passive use of tit, however, one would expect the use of the ablative case as with the other passive affixes. Nevertheless, the agent phrases can be in all three cases, as with causative reflexives. 7.4 Impersonal passives In my fieldwork, I encountered an affix that can be described as an impersonal passive. It is the affix toqar with the allomorph soqar, which is built up by the intransitive participle toq and the verbalizing affix qar, ‘to have’. The intransitive participle can only be added to intransitive or intransitivized verbs and means ‘the one who is doing … (the meaning of the verb)’ (to be discussed in 7.6). This can be seen in the following elicited example (HK): (161) Sinittarfimmi sinittoqarpoq. sleepuse.toplaceLOC.SG sleepIMP.PASS3SG.INDIC ‘There is sleeping in the sleeping room.’ (rough approximation) In (161) the impersonal passive has a corresponding active clause which is intransitive. This clause could be sinittarfimmi sinippoq, ‘he sleeps in the bedroom’. In the impersonal passive, the core argument ‘he’ is suppressed. Even if the verb has an intransitive crossreference ending, referring to a ‘subject’, this subject is unspecified and can be compared to the dummy subject in German impersonal passives (discussed in 4.1.5). In my next example, the impersonal passive is again formed by an intransitive verb. However, this verb is semantically transitive, but what would have been its object, siku, ‘ice’ is incorporated into the verb in the active clause and therefore the verb is intransitive. In this way, the clause from which the impersonal passive is formed is an intransitive clause (HK): (162) Italiami sikutortoqarpoq. ItalyLOC.SG iceeatIMP.PASS3SG.INDIC ‘In Italy there is ice-eating.’ (rough approximation) In the following example, the impersonal passive is added to a transitive verb which is intransitivized by the antipassive affix si (EQ): (163) Perlukunik igitsisoqarpoq. rubbishINSTR.PL throw.outANTIIMP.PASS3SG.INDIC ‘There was throwing out of the rubbish.’ (rough approximation) Without the application of the antipassive, the transitive verb igiC, ‘throw out’ could not be derived by the impersonal passive. Thus, Greenlandic impersonal passives can only be formed from intransitive clauses. 7.5 Causatives functioning as passives? In 7.1.2, I discussed the causative affix tit, ‘cause, let’, which can function as a causative or a passive when it is used intransitively. In the present section, I want to analyze the use of the other causative affixes, qqu, ‘ask, want, tell to’ and tsaali, ‘prevent from’, to see whether these might have passive uses in the same way as tit. I became interested in the analysis of the other causative affixes when reading Berge (1997). In a footnote, Berge (1997: 273, footnote 25) claims that: “the use of certain morphemes, such as tit, tariaqaq Berge (1997) presents the affix, which I call sariaqar as tariaqaq., qqu, and others with intransitive inflectional morphology on a transitive stem automatically entails passive semantics (P. Langgård, personal communication)”. The ‘passive semantics’ entailed by sariaqar, ‘must’ which are mentioned in Berge’s footnote, were discussed in this study in 7.1.5 and 7.1.7. In the present section, I will investigate the intransitive use of the causatives qqu, ‘ask, want, tell to’ and tsaali, ‘prevent from’. When qqu is added to a verb with intransitive morphology, a reflexive sense arises, as can be seen in the following elicited example (HK): (164) Meerap ataataminut illu titartaqquaa. childERG.SG fatherPOSSALL.SG house.ABS.SG drawask.to3SG.3SG.INDIC ‘The child asks his father to draw a house.’ (165) Meeraq ataataminut titartaqquvoq. child.ABS.SG fatherPOSSALL.SG drawask.to3SG.INDIC ‘The child asks his father to draw him’ (i.e. the child wants to get drawn by his father). The verb in example (164) has a causative suffix and a transitive inflectional ending and expresses causation. In (165), the same causative affix is used in a verb with an intransitive inflectional ending, and the translation of the clause is reflexive. Thus, meeraq, ‘child’ which is subject in (164) and (165), is coreferent with the object in (165). In both clauses, the causee ataatami, ‘his father’ is in the allative case. The reflexive causative clause (165) has only a reflexive meaning and cannot be interpreted as a passive such as ‘the child is drawn’. In the same way as qqu, tsaali expresses a reflexive rather than a passive in an intransitively inflected verb, as in the following elicited example (DP): (166) Nanoq Piitamut titartartaalivoq. polar.bear.ABS.SG PeterALL drawprevent.from3SG.INDIC ‘The polar bear prevents himself from being drawn by Peter.’ (e.g. by running around, so Peter cannot see him) However, I have one example in my field data, which could possibly contain the causative affix tsaali used intransitively as a passive (HK): (167) Ukkusissamik qiperugaq igitsaalivoq. soapstoneINSTR.SG sculpture.ABS.SG throw.awayprevent from3SG.INDIC ‘The soap stone sculpture is prevented from being thrown away.’ In the intransitive use of qqu and tsaali, the same restrictions on animacy are apparent as with tit (discussed in 7.1.2). In the above example, however, the subject is inanimate. Nevertheless, the informant accepted the example in the interpretation that the soapstone figure was made by the speaker’s father which made the figure very important for the speaker. In this way, the soapstone sculpture could be associated as being alive or at least as very close to the speaker (discussed in 7.1.2). According to my informant, the translation of example (167) is that the speaker prevents the soapstone sculpture from being thrown away, rather than the sculpture does it itself. This points to a passive analysis of the example, since the subject ukkusissamik qiperugaq, ‘soap stone sculpture’ has the semantic role of the patient while the agent is not expressed in the clause. However, when asking other informants in Greenland and Denmark the above example is not accepted. According to my consultant in Denmark, the interpretation of the example is reflexive and cannot be passive. Furthermore, he categorizes the affix tsaali as a form used by older people which is not used by young people. The informant who accepted example (167) is middle aged (43 years), the other informants were all younger. Whatever its exact status, even if this construction is correct for some speakers, it remains marginal at best. Thus, qqu and tsaali in intransitive use are mostly interpreted as reflexive causatives and not passives. Only one of my informants accepted in a single case a passive interpretation. Thus, when a causative clause with qqu or tsaali is passivized, a passive affix is added to the clause as in the following elicited example, in which the agent can be expressed with an ablative (DP): (168) Jeanette sulitsaalineqarpoq Emiliamiit. Jeanette.ABS workprevent.fromPASS3SG.INDIC EmiliaABL2.SG ‘Jeanette is prevented from her work by Emilia.’ Another question to be discussed here is the following: why has only the causative marked by tit, ‘cause, let’ developed into a passive, while the other two causative structures qqu, ‘ask to’ and tsaali, ‘prevent from’ have not. The explanation I want to offer has to do with the semantic content of these affixes. The affix tit is causative in meaning and thus a ‘usual’ causative while qqu and tsaali are not inherently causative in their meaning although they have the same structure as causatives (discussed in 6.1.4). The differences in meaning of the affixes result in distinct outcomes of the action. The affix tit expresses causation which usually leads directly to an effect, i.e. the outcome results from what is expressed by the verbal stem. Thus, a clause with this type of causative, such as ‘the child makes his father draw him’, results in that the father paints the child. With the causative affix qqu, ‘ask, want, tell to’, on the other hand, the outcome of the action is unknown. It is only expressed that some effect is desired. Thus, ‘the child wants his father to draw him’ may result in that the father draws the child, but the outcome may also be that the child is not drawn. The final causative affix, tsaali, ‘prevent from’ leads to an outcome that is the opposite of the one expressed by the verbal stem. Thus, ‘the child prevents his father from drawing him’ has the outcome that the child is not drawn. In this way, the three different markers for the causative differ in the outcome of the situation. The outcome of the affix tit, ‘cause, let’ is the same as that of the passive: ‘the child makes his father draw him’ and ‘the child is drawn’ both result in a picture of the child. The outcome of the other two affixes, however, differ in that of the passive. This is probably why these affixes have no passive use with an intransitive inflection and why only the causative affix tit developed into a passive. 7.6 The development of the passive in Greenlandic To describe the development of the passive in Greenlandic is the aim of this section. According to Haspelmath (1990), there are several sources for passive morphology, most of which involve grammaticization (treated in 4.4). In the present section, I want to discuss how the different forms of the passive in Greenlandic may have developed in terms of Haspelmath’s theory. Haspelmath (1990: 48), has Greenlandic as one language in his sample. He describes the Greenlandic affix tit as a passive affix which developed from a reflexive causative. I would conclude of the basis of my data that the use of this affix is not extended by analogy. Rather, only the first step in the development is taken towards a use as a unrestricted, productive passive marker (discussed in 7.1.2). On the other hand, the passive use of the affix tit developed from a reflexive causative hundreds of years ago since this affix is represented already in Top (1727) with both functions, which means that this structure has marked both the causative and the passive for a long time. A reason for this continuity may be that there already is a passive structure, which covers all clauses in the language. In this way, it is not necessary for the language to develop another passive form of this kind. Rather, tit is used to express the passivization affecting only a human or higher animate participant while neqar may express all kinds of passivization. According to Haspelmath (ibid.), the affixes saa and neqar have developed from inactive auxiliary suffixes. I would add the impersonal passive affix toqar here since it has developed in the same way. These ‘auxiliaries’ in Greenlandic are not analytic forms, but the derivational affixes u, ‘be’ and qar, ‘have’. Both affixes have an existential meaning. As in analytic passives, the auxiliaries stand together with the participle forms of the verb. These participles are formed with the passive participle suffix saq, the abstract participle suffix neq, and the intransitive participle suffix toq. The stative passive affix saa is built up by the passive participle and the copula u. The stative character arises via the use of the auxiliary u, ‘be’ which can also be used as the copula in a nonverbal predicative construction. The passive sense arises via the use of the passive participle saq. This form of the passive was discussed comprehensively in 7.1.3. The impersonal passive toqar and the passive affix neqar are both built up by a participle suffix and the verbalizing derivational affix qar, ‘have’. I want to investigate how the passive meaning of neqar and toqar could arise, and for that purpose I will look at the participles toq and neq respectively in more detail. The impersonal passive toqar involves the intransitive participle toq. Van der Voort (1991: 189) defines this intransitive participle in the following way: The intransitive participle, soq, toq, tsoq, refers to the one who carries out the ‘action’ indicated by a formally intransitive verb. (…) When applied to a transitive verb, a halftransitive affix i.e. antipassive affix, J.S. has to precede the participle (…) unless it is intended as reflexive. Thus, the intransitive participle soq is only added to intransitivized or intransitive verbs. In many European languages, the derivation of an intransitive verb by the passive results in an impersonal passive. When the passive affix toqar is added to intransitive verbs in Greenlandic, it may also result in an impersonal passive. It therefore seems to have been a natural development. I want to discuss the development from an intransitive participle to an impersonal passive by means of some examples. When the intransitive verb siniC, ‘sleep’ is derived by the intransitive participle, it becomes sinittoq, ‘sleeper, sleeping person’. This participle, formally a nominal, can be fused with the verbalizing affix qar, ‘have’, becoming sinittoqar, literally ‘sleeping person have’, i.e. ‘there is a sleeping person’. In this way, qar, ‘have’ may have developed an existential meaning, known from many other languages, e.g. Spanish hay and French il y a. Finally, the inflectional ending following qar, contains a third person singular which does not refer to the person sleeping since this is incorporated by the verb. However, there is only one participant in an intransitive verb. Thus, the third person inflection in the verb does not refer to a specific participant, but is unspecified and can be compared to a dummy subject. Thus, sinittoqarpoq may be translated as ‘there is sleeping’, i.e. ‘someone is/ people are sleeping’. The passive morpheme neqar contains the abstract participle neq. Van der Voort (1991: 1920) defines this participle in the following way: The abstract participle, neq, is used for nonfinite complement clauses (…). This suffix refers to the ‘action’ itself that is indicated by the verb. On intransitive verbs, the possessor corresponds to the semantic subject of the verbal stem. In rare cases where it is attached to a transitive verb, the possessor will correspond to an object (…), and the verbal stem can get reflexive or passive sense. Since the participle neq is nominalizing, the relation to the arguments of the former verb has to be expressed by other means than by verbal inflection. According to van der Voort, these arguments may be expressed as possessors of the nominalized verb. In this way, the subject of an intransitive verb derived by neq becomes the possessor while in transitive verbs derived by neq, the object is expressed as the possessor. Fortescue (1984: 46) has the following example of a transitive verb derived by neq where the object of the underlying verb is the possessor: (169) Angutip toqunnera. manPOSS killAB.PARTPOSS ‘The killing of the man.’ The passive affix neqar is only added to transitive verbs while the participle neq is mostly added to intransitive verbs, as can be seen from the definition above. When the nominalization with neq is incorporated by the verbalizing affix qar, the possessor becomes the subject of the clause. Thus, the clause gets a passive meaning and the above example becomes angut toquneqarpoq, ‘the man was killed.’ To sum up, the affix toqar is only added to intransitive verbs and results in the impersonalization of the situation. The affix neqar is only added to transitive verbs where it emphasizes the object of that verb and thus results in the passivization of the situation. An explanation why neqar is only added to transitive verbs may be that another structure, toqar, already functions as an impersonal passive for intransitives. Economically, there is no reason for a language to have two different impersonal passives and thus, neqar is restricted to transitive verbs, where it expresses a dynamic passive. In this way, these two passive affixes divide different functions between each other. In the following section, I want to discuss the functions of the different passive affixes in Greenlandic. 7.7 The functions of the passive in Greenlandic Many theories of the passive describe the functions of this structure in terms of foregrounding and backgrounding of certain elements in the clause (discussed in 4.6). Furthermore, the primary function of the passive is sometimes said to be the fulfillment of pivot conditions. Languages without pivot restrictions, such as languages with switch reference, are then said to lack a passive, or they may have a passive but one with very restricted functions. Dixon (1994: 154) expresses this in the following way (repeated from 3.3): Languages with switch reference typically lack passive and antipassive or  if they do include these  they do not have any pivotfeeding function. (…) It can be argued that there is no syntactic need of passive or antipassive, since it is not necessary to bring an underlying O or A NP into derived S function, to facilitate NP omission without any chance of ambiguity. However, Greenlandic goes against Dixon’s expectation that languages with switch reference typically lack a passive since this language has an elaborate passive and switch reference at the same time. Thus, Greenlandic is an exception to Dixon’s claim. However, I want to argue that the fulfillment of pivot conditions is not the primary function of the passive in general, and therefore, also languages with switch reference may have a passive. According to Haspelmath (1990), the passive is a verbal category and its primary function is the inactivization of the situation (discussed in 4.6). The foregrounding and backgrounding of certain elements are the logical consequences of this function. The reason why passive structures are found in languages with pivot restrictions is that the passive is essential in the grammar of these languages: by means of foregrounding and backgrounding of constituents, elements can be brought to pivot position and in this way, the structural requirements can be fulfilled. The passive has developed the function to fulfill pivot conditions in languages with pivot restrictions, but it is not the primary function of the passive. Languages with pivot restrictions usually have a passive while languages with switch reference may or may not have to have a passive. However, I have the hypothesis that passive actually is used in Greenlandic and other languages to fulfil certain conditions in the switch reference system (see also Sakel 1998a). Imagining the sentence ‘The professional killer murdered John, because the Mafia leader hated John’, the Greenlandic verb would be marked by ‘different reference’, since the subjects of these adjacent clauses are not coreferent. However, the objects of these two clauses are coreferent, but this cannot be shown in the switch reference marking. I have the hypothesis that there is a pragmatic tendency towards the expression of coreference when possible. In the example above, coreference can be shown by passivizing both clauses, marking the verb for ‘same reference’. However, I have no empirical data for this claim, and it would need further research to find out whether this claim is true. The functions of the passive in Greenlandic may thus be described in terms of inactivization of the situation, leading to foregrounding and backgrounding of certain elements. I want to discuss these functions, using the following elicited examples with the passive affix neqar in (171) and the corresponding active clause in (170) (DP): (170) Inuit meeraq takuaat. humanERG.PL child.ABS.SG see3PL.3SG.INDIC ‘The people saw the child.’ (171) Meeraq (inunniit) takuneqarpoq. child.ABS.SG (humanABL2.PL) seePASS3SG.INDIC ‘The child was seen (by the people).’ In the active clause (170), the situation described by the verb is active since the performer of the action is the subject of the clause. In (171), the situation described by the verb is passive since the undergoer of the action expressed by the verb is in subject position. In this clause, the performer of the action is demoted and may be omitted from the clause which is indicated by the parentheses. Thus, the subject of the active clause is backgrounded in the passive clause. At the same time, the object of the active clause is foregrounded in the passive clause where it is in subject position. However, for a complete analysis of the functions of the passive in Greenlandic, one has to investigate the passive in its use, such as in spoken texts. An investigation of the functions of the passive would require a broad corpus of spoken texts which is not available. My fieldwork in Greenland primarily involved elicitation and passives are very rare in my textual material. Therefore, I cannot give an appropriate analysis of the functions of the passive in Greenlandic discourse. This is left open for future research. 7.8 Summary In the present chapter, I have investigated the forms and the distribution of the different passives in Greenlandic. The Greenlandic passives are formally marked by the morphemes neqar, tit, saa, and toqar. These are all marked as additional stem affixes in the sense of Haspelmath (1990), i.e. they are added to the verbal stem before any tense, aspect, and person marking. The passive affix sigi, given by Fortescue (1984), is not productive in the language and therefore I did not include it in this study. The other two affixes sima and sariaqar which were analyzed as having a passive function by Fortescue, turned out not to be passives but rather affixes deriving non-agentive verbs, i.e. verbs which can be used transitively and intransitively, having the object of the transitive form as the subject of the intransitive one (discussed in 7.1.7). Apart from tit, all passive affixes are built up by a participle and a verbalizing derivational affix. The affix tit developed from a productive causative. In this way, the formal marking for the tit passive is at the same time a marking for the causative. All other passive affixes are strictly used for passive purposes. The affixes -tit- and -neqar- have overlapping meanings in that they both express the dynamic passive. However, tit is more restricted in its distribution as it is only used with human and ‘higher animate’, i.e. intentional, subjects. The other two passive affixes have different meanings. The affix saa expresses the stative passive and toqar the impersonal passive. The different forms of the passive in Greenlandic, their uses and their historical sources are summarized in the following table: use base verb historically derived from -neqar- dynamic passive transitive abstract participle + ‘have’ -tit- dynamic passive (restricted to intentional human or ‘higher’ animate subjects) transitive causative -saa- stative passive transitive passive participle + ‘be’ -toqar- impersonal passive intransitive intransitive participle + ‘have’ Table 22: the different passive affixes As mentioned, the functions of the passive in Greenlandic can be described in terms of inactivizing the situation which leads to the foregrounding and backgrounding of certain elements. In the foregrounding, or promotion of elements, prominent objects become subjects in the passive clause. Only objects can be promoted in Greenlandic. Thus, when promoting an oblique element to the subject of the clause, this element has first to be derived by an applicative to enable it to come into object position. Then, the passive can be applied and the element comes into subject position. Since Greenlandic is morphologically ergative, the promoted object is in the absolutive case both in the active and in the passive clause. The backgrounded, i.e. demoted, element is coded as an oblique in the agent phrase of the passive clause. Both passives and causatives have agent phrases which resemble each other apart from their case marking. The agent phrases of the passive affix neqar are marked by the ablative or instrumental case while the agent phrases of causatives and the stative passive are in the allative case. In this way, the passive affix tit and the stative passive affix saa differ from neqar since they can have an agent phrase in the ablative, allative or instrumental case. 8. Conclusion In the present study, I gave a description of the category voice with special emphasis on passive in Greenlandic from a typological perspective. I concentrated on the different forms of the passive and their uses. There are four different voice structures in Greenlandic. These structures can be roughly described in terms of valence, even if they do not in all cases involve a valence change: the passive and the antipassive decrease the valence of a verb, apart from impersonal passives which involve intransitive active and passive clauses. The causative and applicative are voice structures which increase the valence of intransitive clauses. When added to transitive verbs, however, they do not result in a change of the valence. All of these voice structures are marked morphologically on the verb and they are represented by at least three affixes. There may be more than one voice structure in the same clause, and most of these voice structures can be freely combined, resulting in different nuances of meaning. My fieldwork revealed four different productive passive affixes. These are the dynamic passives neqar and tit, the stative passive saa, and the impersonal passive -toqar-. It appeared that the passive with the affix tit, which is a productive causative marker at the same time, has restrictions on the passive subjects. Only NPs which can be causers in a causative construction (i.e. animates) can be subjects in this type of passive. In this way, the passive construction with the affix tit formally has the same structure as a reflexive causative. However, this structure can be identified as a passive. First, the passive affix tit is regularly used by the speakers of Greenlandic instead of the passive affix neqar, expressing roughly the same meaning. Second, the affix -tit- is identified as a passive in metaphorical expressions when used with an inanimate subject. In these cases, the inanimate subject is either thought of as a ‘living thing’ or the passive affix neqar is used instead of tit. Furthermore, only the causative affix tit can be used as a passive when inflected intransitively, the other causatives cannot. However, being identified as a passive, tit differs from the other passive affixes in many ways. While the other passive affixes have their respective uses, tit seems to be a dynamic passive in the same way as neqar. Often tit and neqar appear interchangeably and only in very few cases is there a minor difference in meaning between these affixes. In addition to that, the causative construction marked by tit is productive in Greenlandic and thus, tit is the only passive affix that marks both the passive and another structure. All other passive affixes are restricted to marking the passive. The construction with tit has the same formal structure as reflexive causatives, having agent phrases in three cases and restrictions on the subjects. The passive affixes have developed from inactive auxiliaries and are built up by verbalized participles, apart from the affix tit, which has developed from a causative. It has still the same formal structure as a reflexive causative. This has been the case for at least 270 years, as the passive with the affix tit is mentioned by Top (1727). How can it be that this affix has never been extended to other subjects than just those with a human or ‘higher’ animate subject as in causative clause? The answer to this question can be that other affixes, such as neqar, already functioned as passive affixes with animate and inanimate subjects. Therefore, there was no reason for the passive construction with the affix tit to build analogy to other clauses. However, the question arises why tit developed in the first place when the passive construction with the affix neqar already had the functions of a dynamic passive. To answer this question, one would have to investigate the diachronic development of the passive in Greenlandic much more closely than is possible in the present study. I have the hypothesis that tit has developed independently from the other passive affixes by reinterpretation of a reflexive causative, without there being use for another passive affix. This happened because the reflexive causative is semantically very close to the passive, both having the same outcome. Thus, the affix tit got a passive interpretation. However, it did not develop into a productive passive marker for all clauses in the language because other affixes already had this function. Nevertheless, the affix tit kept its passive interpretation and is used frequently in the spoken language instead of or alongside neqar. From a typological perspective, Greenlandic is an exception to Dixon’s (1994: 154) claim that languages with switch reference typically lack voice structures such as passive and antipassive: Greenlandic has a passive, an antipassive, and switch reference at the same time. Furthermore, Dixon (1994: 147) claims that ergative languages typically have an antipassive, while accusative languages typically have a passive. Greenlandic, being morphologically ergative, has both voice structures. Nevertheless, Greenlandic is not exceptional, since Nichols (1992) found that the passive appears in both morphologically ergative and accusative languages. On the other hand, only languages with a special focus on their objects, such as morphologically ergative languages, have an antipassive, according to Nichols. Apart from the passive and antipassive, Greenlandic has the voice structures causative and applicative. All voice structures are represented by at least three affixes which usually differ functionally. Thus, the passives in Greenlandic have their respective uses except that tit and neqar function as dynamic passives: the affix saa denotes a stative passive and the affix toqar marks the impersonal passive. The causative has three different affixes, expressing ‘cause, let’, ‘ask, want, tell to’, and ‘prevent from’ (discussed in 6.1.4). The applicative has five affixes, meaning ‘do together with’, ‘do with/ for’, ‘have as means/ time of’, ‘have as reason for’ and ‘have as place/ time, etc. for’ (discussed in 6.1.5). Finally, the five markers of the antipassive seem to differ in aspect, according to Bittner (1987) (discussed in 6.1.3). However, Bittner is not very clear in her description of the different uses of these affixes. Thus, apart from the titpassive, the different voice affixes in Greenlandic seem to differ in meaning, i.e. they have different functions. The different functions of the voice affixes may explain why there are several different affixes in each voice structure in Greenlandic. Nevertheless, the question arises why Greenlandic has so many voice structures with different functions, as opposed to other languages such as English with far fewer voice structures. I have the hypothesis that the existence of so many voice structures has to do with the structure of Greenlandic. As a polysynthetic language, Greenlandic expresses many concepts morphologically in the verb, as opposed to other languages, such as English, which express many structures by analytic means or by a change in the word order. I want to discuss this difference between Greenlandic and English by means of the following example from Fortescue (1984: 270) (repeated from 6.3): (172) Ilinnik oqaloqatiginnissutigaara. thouINSTR.PL speakAPPLANTIAPPL1SG.3SG.INDIC ‘I wanted to speak to you about it.’ (literally: ‘I speak to you because of it’) In this example, the Greenlandic clause consists of an oblique instrumental element and a verb containing three voice affixes. In the English translation, on the other hand, the concepts which are expressed by voice affixes in Greenlandic are expressed by separate words. The Greenlandic verb in example (172) is built up by the intransitive verbal stem oqalu, ‘speak’, followed by qatigi, ‘do together with’, resulting in the transitive clause in Greenlandic ‘I speak with you’. The object ‘you’ of this transitive clause is demoted to the oblique instrumental element ilinnik, ‘to you’ by the application of the antipassive, marked by the affix nnig. This results in the intransitive clause ‘I speak to you’. By the application of the applicative ssutigi, ‘have as reason for’, the pronoun it, i.e. the thing talked about, is added as the object of the clause and the clause is retransitivized. In this way, the clause becomes ‘I speak to you because of it’. In the Greenlandic clause, ‘I’ is the subject and ‘it’ is the object, as can be seen in the crossreference markers. In English, the same concepts are expressed by separate words in the clause: the beneficiary is expressed by the preposition ‘to’ in ‘to you’. This is expressed by the applicative suffix qatigi, ‘do together with’ and an antipassive in Greenlandic. In English, ‘about it’ consists of the preposition ‘about’ and the pronoun ‘it’, i.e. the thing talked about. In Greenlandic, however, another applicative is used for expressing this concept. Thus, two different applicatives are used in the Greenlandic example, in the same way as two different prepositions are used in the English clause to express different concepts. All in all, I claim that Greenlandic has so many different voice structures, because of its morphological structure. English expresses many concepts by separate words, while Greenlandic expresses these concepts primarily morphologically in the verb since it is a polysynthetic language. Thus, concepts which involve a change of the situation in the clause are expressed by voice structures in Greenlandic, while in other languages they may be expressed by voice structures, by prepositions or in other ways (such as word order). In Greenlandic, these are more conspicuous because of the explicit presence of morphological markers, which can be explained by the fact that most grammatical structures in Greenlandic are marked morphologically. In English, similar constructions exist, but they are expressed by analytical means, which is less visible. In this way, other languages may express as many voice concepts as Greenlandic. One can argue that the use of prepositions in English can be described as a voice structure as well. In the same way, so-called word order inverses (discussed in 3.1.5) are called voice structures, since they are structurally parallel to morphologically marked inverse structures. However, word order inverses are not marked but they result only in a change of the word order. In my definition of voice (given in 4.7), I do not treat these structures as voice structures, even if they may be structurally similar to voice structures. Cree (Algonquian), another polysynthetic language, has many different voice structures, which are represented by a number of different affixes (Peter Bakker, personal communication). It could be that polysynthetic languages typically have many morphologically marked voice structures, since languages of this type often mark structures which are expressed by analytic means in other languages morphologically in the verb. However, this discussion is outside the scope of this study. Additional research will show if my hypothesis is true. The present study is partly built on fieldwork data. My fieldwork involved mostly elicitation which, according to Payne (1997), is not appropriate for an analysis of voice phenomena. However, considering the limited time I had for my fieldwork, elicitation was most appropriate for my investigation of the forms and uses of the passive. In future research about passive and voice in Greenlandic, these results should be tested in spontaneous spoken Greenlandic in order to find out whether the passive is used the way I recorded by elicitation and furthermore to find out about the functions of the passive in connected discourse in this language. 9. Appendix The following tables show the forms of the inflectional endings in the intransitive and transitive inflection with mood marking in the indicative and conditional mood. These two moods represent independent moods (indicative) and subordinate moods (conditional) which differ in their inflectional endings. The initial consonants of the suffixes can have a different form due to differences in the verbal stems discussed in chapter 5. For more information about the forms of the inflection in Greenlandic, see Fortescue (1984). Greenlandic case markers are listed in chapter 5. Indicative mood, intransitive inflection: 1.sg vunga 1. pl vugut 2. sg vutit 2. pl vusi 3. sg voq 3. pl pput Indicative mood, transitive inflection: object subject 1. sg 2. sg 3. sg 1. pl 2. pl 3. pl 1. sg - -vakkit -vara - -vassi -vakka 2. sg -varma - -vat -vatsigut - -vatit 3. sg -vaanga -vaatit -vaa -vaatigut -vaasi -vai 1. pl - -vatsigit -varput - -vassi -vavut 2. pl -vassinga - -varsi -vassigut - -vasi 3. pl -vaannga -vaatsit -vaat -vaatigut -vaasi -vaat Conditional mood, intransitive inflection: 1. sg guma 1. pl gutta 2. sg guit 2. pl gussi 3. sg ppat 3. pl ppata 4. sg guni 4. pl gunik Conditional mood, transitive inflection: subject object 1sg 2sg 3sg 4sg 1pl 2pl 3pl 4pl 1sg gukkit gukku gunni gussi gukkit gutsik 2sg gumma gukku gunni gutsigut gukkit gutsik 3sg ppanga ppatit ppagu ppani ppatigut ppasi ppagit ppatik 4sg guninga gunisit guniuk gunisigut gunisi gunigit 1pl gutsigit gutsigu gutsinni gussi gutsigik gutsik 2pl gussinnga gussiuk gussinni gutsigut gussigik gutsik 3pl ppannga ppatsit ppassuk ppanni ppatigut ppasi ppatigik ppatik 4pl guninnga gunitsit gunikku gunisigut gunisi gunikkik 10. References Ayres, Glenn (1983) ‘The antipassive voice in Ixil’ International Journal of American Linguistics 49(1): 2045 Barber, Elizabeth (1975) ‘Voice  beyond the passive’ Berkeley Linguistics Society 1: 1624 Berge, Anna M.S. (1997) Topic and discourse structure in West Greenlandic agreement constructions dissertation at the University og California at Berkeley Bergsland, Knut & Jørgen Rischel (eds.) (1986) Pioneers of Eskimo Grammar Copenhagen: The Linguistic Circle of Copenhagen Bittner, Maria (1987) ‘On the Semantics of the Greenlandic Antipassive and Related Constructions’ International Journal of American Linguistics 54 (2): 194231 Brochmann, Helene (1993) Qanoq 1, Opgave og grammatikbog unpublished manuscript Bussmann, Hadumod (1990) Lexikon der Sprachwissenschaft Stuttgart: Alfred Kröner Verlag Bybee, Joan (1985) Morphology: A study of the relation between meaning and form Amsterdam: John Benjamins Bybee, Joan, Revere Perkins & William Pagliuca (1994) The evolution of grammar: tense, aspect and modality in the languages of the world Chicago: University of Chicago Press Chomsky, Noam (1957) Syntactic Structures The Hague: Mouton Chomsky, Noam (1965) Aspects of the Theory of Syntax Cambridge Mass.: MIT Press Cole, P. & Sadock, J.M. (eds.) (1977) Grammatical Relations New York: Academic Press Comrie, Bernard (1977) ‘In defense of spontaneous demotion: The impersonal passive’ in P. Cole & J. Sadock (eds.) (1977) Grammatical Relations New York: Academic Press, pp. 4758 Comrie, Bernard (1988) ‘Passive and voice’ in M. Shibatani (ed.) (1988) Passive and Voice Amsterdam: John Benjamins, pp.924 Comrie, Bernard (1989) Language Universals and Linguistic Typology 2nd edition (1st edition: 1981), Chicago: University of Chicago Press Cooreman, Ann (1994) ‘A functional typology of antipassives’ in B. Fox & P.J. Hopper (ed.) (1994) Voice  Form and Function Amsterdam: John Benjamins, pp.4988 Croft, William (1990) Typology and Universals Cambridge: Cambridge University Press Dahlstrom, Amy (1991) Plains Cree Morphosyntax New York and London: Garland Dalgish, G.M. (1976) ‘Passivizing locatives in Olutsootso’ Studies in the Linguistic Sciences 6(1): 5786 De Wolf, Charles M. (1988) ‘Voice in Austronesian languages of Philippine type: Passive, ergative or neither?’ in Shibatani (ed.) (1988) Passive and voice Amsterdam: John Benjamins, pp.14393 Dik, Simon C. (1997) The Theory of Functional Grammar (Part I: The Structure of the Clause) Berlin: Mouton de Gruyter Dixon (ed.) (1976) Grammatical Categories in Australian Languages Canberra: Australian Institute of Aboriginal Studies Dixon, R.M.W. (1979) ‘Ergativity’ Language 55: 59138 Dixon, R.M.W. (1994) Ergativity Cambridge: Cambridge University Press Egede, Hans (1722) ‘Nogle grønlandske vocabula’ reprinted in Bergsland & Rischel (1986) Pioneers of Eskimo Grammar Copenhagen: The Linguistic Circle of Copenhagen, pp.4952 Egede, Hans (1725) ‘Dictionarium, Formula conjugandi, etc.’ reprinted in Bergsland & Rischel (1986) Pioneers of Eskimo Grammar Copenhagen: The Linguistic Circle of Copenhagen, pp.5376 Foley, W.A. & R.D. van Valin, JR (1984) Functional Syntax and Universal Grammar Cambridge: Cambridge University Press Fortescue, Michael D. (1980) ‘Affix ordering in West Greenlandic derivational processes’ International Journal of American Linguistics 46(4): 259278 Fortescue, Michael D. (1984) West Greenlandic London: Croom Helm Fortescue, Michael D. (1991) ‘Switch reference anomalies and ‘topic’ in West Greenlandic: A case of pragmatics over syntax’ J. Verschueren (ed.) (1991) Levels of Linguistic Adaption Amsterdam: John Benjamins, pp.5380 Fortescue, Michael D. (1992) ‘The development of morphophonemic complexity in Eskimo languages’ Acta Linguistica Hafniensia 25: 527 Fortescue, Michael (1993) ‘Eskimo word order variation and its contactinduced perturbation’ Journal of Linguistics 29(2): 267289 Fortescue, Michael (1997) ‘Eskimo influence on the formation of the Chukotkan ergative clause’ Studies in Language 21(2): 369409 Fox, B. & P.J. Hopper (ed.) (1994) Voice  Form and Function Amsterdam: John Benjamins Givón, Talmy (1979) On understanding grammar New York: Academic Press Givón, Talmy (1981) ‘Typology and functional domains’ Studies in Language 5: 163193 Givón, Talmy (ed.) (1994) Voice and Inversion Amsterdam: John Benjamins Greenberg, Joseph H. (1966) ‘Language universals, with special reference to feature hierarchies’ Janua Linguarum, Series Minor 59, The Hague: Mouton Haiman, J. & P. Munro (eds.) (1983) Switch reference and universal grammar Amsterdam: John Benjamins Harris, Zellig S. (1951) Methods in Structural Linguistics Chicago: University of Chicago Press Haspelmath, Martin (1990) ‘Grammaticization of passive morphology’ Studies in Language 14(1): 2572 Heath, J. (1976) ‘Antipassivization, a functional typology’ Proceedings of the 2nd Annual Meeting of the Berkeley Linguistic Society: 202211 Hopper, P.J. & S.A. Thompson (1980) ‘Transitivity in Grammar and Discourse’ Language 56(2): 251299 Hopper, P.J. & S.A. Thompson (ed.) (1982) Studies in Transitivity New York: Academic Press Jacobsen, William H. Jr. (1961) ‘Switch-reference: a Hokan-Coahuiltecan syntactic device’ Papers presented at the 36th Annual Meeting of the Linguistic Society of America Janussen, Estrid (1987) Håndbog i grønlandsk grammatik Nuuk: Pilersuiffik Jespersen, Otto (1924) The Philosophy of Grammar London: Allen & Unwin Kalmár, I. (1979) ‘The antipassive and grammatical relations in Eskimo’ in F. Plank (ed.) (1979) Ergativity London: Academic Press, pp.117143 Keenan, E.L. (1985) ‘Passive in the world’s languages’ in T. Shopen (ed.) (1985) Language Typology and Syntactic Description Cambridge: Cambridge University Press, pp.243281 Kemmer, Suzanne (1993) The Middle Voice Amsterdam: John Benjamins Klaiman, M.H. (1991) Grammatical Voice Cambridge: Cambridge University Press Kleinschmidt, Samuel (1851) Grammatik der grönländischen sprache mit theilweisem einschluss des Labradordialects Berlin: G.Reimer Lacey, H.M. (1979) ‘Control, perceived control, and the methodological role of cognitive constructs’ in Perlmuter & Monty (eds.) (1979) Choice and perceived control Hillsdale, NJ: Lawrence Erlbaum Associates, pp. 516 Meillet, A. (1912a) Linguistique historique et linguistique générale 2nd edition (1921), Paris Meillet, A. (1912b) ‘L’évolution des formes grammaticales’ in Meillet (ed.) (1912a) Linguistique historique et linguistique générale Paris, pp.130148 Mithun, Marianne (1988) ‘Systemdefining Structural Properties in Polysynthetic Languages’ Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikationsforschung 41(4): 442451 Nedjalkov, Vladimir P. (ed.) (1988) Typology of resultative constructions Amsterdam: John Benjamins Nedjalkov, V. & S. Jaxontov (1988) ‘The typology of resultative constructions’ in V. Nedjalkov (ed.) (1988) Typology of resultative constructions Amsterdam: John Benjamins, pp.362 Nichols, Johanna (1986) ‘Headmarking and dependentmarking in grammar’ Language 62(1): 56119 Nichols, Johanna (1992) Linguistic Diversity in Space and Time Chicago: University of Chicago Press Olsen, L.L. & B. Hertling (1988) Grønlandsk tilhængsliste Nuuk: Pilersuiffik Palmer, F.R. (1994) Grammatical Roles and Relations Cambridge: Cambridge University Press Payne, Thomas E. (1997) Describing morphosyntax Cambridge: Cambridge University Press Perlmuter, L.C. & R.A. Monty (eds.) (1979) Choice and perceived control Hillsdale, NJ: Lawrence Erlbaum Associates Perlmutter, David M. & Paul Postal (1977) ‘Toward a universal characterization of passivization Proceedings of the 3rd Annual Meeting of the Berkeley Linguistic Society: 394417 Plank, Frans (ed.) (1979) Ergativity: towards a theory of grammatical relations London: Academic Press Rasmussen, Chr. (1888) Grønlandsk Sproglære Kjøbenhavn: G.E.C. Gad Sakel, Jeanette (1998a) Switch reference i grønlandsk unpublished paper at the Institut for Lingvistik: Aarhus Universitet Sakel, Jeanette (1998b) Ergativitet i grønlandsk unpublished paper at the Institut for Lingvistik: Aarhus Universitet Sakel, Jeanette (1998c) Fieldwork on passive in Greenlandic unpublished paper at the Institut for Lingvistik: Aarhus Universitet SchultzLorentzen, C.W. (1945) A Grammar of the West Greenlandic Language Copenhagen: Meddelelser om Grønland 129(3) Shibatani, Masayoshi (ed.) (1988) Passive and Voice Amsterdam: John Benjamins Shopen, Timothy (ed.) (1985) Language typology and syntactic description I, clause structure Cambridge: Cambridge University Press Siewierska, Anna (1984) The Passive: A Comparative Linguistic Analysis London: Croom Helm Silverstein, Michael (1976) ‘Hierarchy of features and ergativity’ in B. Dixon (ed.) (1976) Grammatical Categories in Australian Languages Canberra: Australian Institute of Aboriginal Studies, pp.112171 Tesnière, Lucien (1959) Éléments de syntaxe structurale Paris: Librairie C. Klincksieck Top, Albert (1727) ‘Relation’ reprinted in Bergsland & Rischel (1986) Pioneers of Eskimo Grammar Copenhagen: The Linguistic Circle of Copenhagen, pp.76154 Verschueren, J. (ed.) (1991) Levels of Linguistic Adaption Amsterdam: John Benjamins Vilkuna, Maria (1996) Suomen lauseopin perusteet An introduction to Finnish syntax Helsinki: Edita Voort, Hein v.d. (1991) Relative Clauses in West Greenlandic  A descriptive analysis in terms of Functional Grammar unpublished MA thesis at Universiteit van Amsterdam PAGE 10