Äîêóìåíò âçÿò èç êýøà ïîèñêîâîé ìàøèíû. Àäðåñ îðèãèíàëüíîãî äîêóìåíòà : http://www.philol.msu.ru/~otipl/new/main/people/kibrik-aa/files/Case-Head%202006%20%23%23%23A.pdf Äàòà èçìåíåíèÿ: Thu Feb 7 14:45:16 2008 Äàòà èíäåêñèðîâàíèÿ: Sat Sep 6 04:19:38 2008 Êîäèðîâêà: Ïîèñêîâûå ñëîâà: southern cross

WHAT'S IN THE HEAD OF HEAD-MARKING LANGUAGES? or IS THERE CASE IN HEAD-MARKING LANGUAGES1?
Andrej A. Kibrik (Institute of Linguistics, Russian Academy of Sciences) kibrik@comtv.ru
Abstract Marking of clause participants' semantic roles is an important concern in human languages. Dependent-marking languages, prevailing in the ENCA area, mark semantic roles by means of nominal case affixes. This article explores the expression of roles in head-marking languages, using Athabaskan languages of North America as the main source of evidence. In headmarking languages clause participants are represented by personal pronominal affixes on the verbs. Linear morphological positions in which personal affixes appear are functionally equivalent to nominal case affixes, while the construal of positions in terms of grammatical is misguided. Athabaskan verb involves the following positions: nominative, accusative, dative, and oblique. A typology of role marking in other types of languages is also proposed.

1. Introduction
This article is a contrubution to "basic linguistic theory", to use the term of Dixon (1997), clearly expounded by Matthew Dryer as follows:
The expression "basic linguistic theory" (following R. M. W. Dixon) refers to the theoretical framework that is most widely employed in language description, particularly grammatical descriptions of entire languages. It is also the framework assumed by most work in linguistic typology. (http://wings.buffalo.edu/soc-

sci/linguistics/people/faculty/dryer/dryer/blt; also see Dryer in press). In this article I look at the phenomenon of head-marking of semantic roles in the clause that is very widely spread in the languages of the world and also found in a number of LENCA. It seems that this phenomenon is not adequately represented in basic linguistic theory, as conceived by many typologists and descriptive linguists. Specifically, I claim that the the head-marking technique of role marking is functionally equivalent to nominal cases ­ nominative, accusative, ergative, etc., rather than to grammatical relations ­ subject, direct object, etc. I also make some suggestions on terminology used to describe the head-marking technique. In the previous years I wrote a number of papers in which I already used this approach and the corresponding terminology (Kibrik 1990, 1992, 1996a, 2002). This approach seemed so obvious and straightforward to me that it hardly called for detailed argumentation and explanation. However, I keep noticing that descriptive linguists and typologists keep using the system that I find flawed. In addition, in several oral presentations I experienced difficulties in making my approach clear to the audience without a very detailed explanation. In this article of an argumentative nature I attempt to convince the reader of how a fragment of basic linguistic theory must look.

2. Marking of semantic roles
Supposedly, the concern for clause arguments' semantic roles is a next to universal feature of speaking humans. In particular, if there is an event involving more than one participant, the speaker and the hearer care to know who is the agent and who is the patient, or who is the experiencer and who is the stumulus. There seem to be some languages that, despite this obvious concern, do not mark semantic roles in any consistent way (Gil 1999 about Riau Indonesian). But the vast majority of languages actually do mark semantic roles, or at least differences between two or more roles in one and the same clause. The most familiar type of role marking is by means of case markers added to nouns or NPs; this is characteristic of conservative Indo-European languages, such as Latin or Russian, as well as many other languages. According to certain rules, semantic roles map onto nominal cases. For example, in a language with the accusative role alignment such as Russian, in a transitive clause the agent-type argument is systematically marked by the nominative case and the patient-type argument by the accusative case (1). (1) Russian Pap-a dad-Nom.Sg
1

umy-l wash-Past(MSg)

dochk-u daughter-Acc.Sg

`Dad washed the daughter'

Research underlying this article was supported by grant #05-04-04240 from the Russian Foundation for the Humanities.


This type of role marking is generally very typical of LENCA. Among the most characteristic LENCA are Turkic languages; they can be considered, so to speak, standard average LENCA. Consider a natural discourse example from a story told in Karachay-Balkar, a Turkic language of Northern Caucasus: (2) Karachay-Balkar ol sirnik-le-ni zandir-ip, he:Nom match-Pl-Acc light-Conv semantic role Agent Patient case Nom Acc `He was striking matches and throwing them bÆrÝ-le-ge at-a wolf-Pl-Dat throw-Conv Goal/recipient Dat at (lit. to) the wolves' e-di Cop-Past(3Sg)

Two or three decades ago the standard view of clause structure, based primarily on typical LENCA such as Latin, Russian or Karachay-Balkar, used to involve the following theses: i. ii. iii. Participants (both arguments and adjuncts) bear semantic roles Participants are coded by nominals Semantic roles are marked/coded by the inflectional category known as case that modifies nominals

Thesis (i) is still valid while theses (ii) and (iii) were widely recognized as non-universal due to two important trends of research. I will consider these two trends one after another. One trend is associated with the idea of "pronominal arguments" dating back to Boas (1911) and even earlier, but emphasized recently (in different terms) by a number of authors (Kumaxov 1974, Van Valin 1977, Jelinek 1984, Mithun1986, Kibrik 1992 among others). Consider the following example from Abkhaz, a language of AbkhazAdyghean (or North-West-Caucasian) language family: (3) Abkhaz (Kibrik 1991:136) a. axac'a apH°s asalamsq' i-l-z-i-j°it' mani womanj letterk 3N.Nomk-3SgF.Oblj-for-3SgM.Ergi-wrote `The man wrote the letter to the woman' i-l-z-i-j°it' 3N.Nom-3SgF.Obl-for-3SgM.Erg-wrote `He wrote it to her'

b.

Each participant of the clause in (3a) is represented by a corresponding personal2 affix in the inflected verb. Against the background of LENCA stereotypes, this can be understood as agreement of the verb with multiple NPs. However, this interpretation misses the point. Personal affixes in languages like Abkhaz are functionally analogous to pronouns of more familiar languages ­ the example in (3b) makes this clear. Often such personal affixes are termed pronominal affixes, and I will use these terms interchangeably below. The agreement interpretation is particularly inept for languages like Abkhaz because most of the time there is nothing to agree with. If an argument is a locutor (first or second person) then the only place where it appears in the clause would be a verbal personal affix. If an argument is a non-locutor (= third person) still very frequently there would be no full NP. Incorporated, or bound, pronouns are genuine clause arguments, while the function of full NPs in a clause like (3b) is to referentially specify the pronominal argument. Nominals are not an inherent part of the clause that consists primarily of the inflected verb. This is the essence of the concept of pronominal argument languages that renders thesis (ii) non-universal. I have just presented a somewhat radical formulation of this approach that encounters certain problems, but these are not central to this paper. In any case, it is clear that pronominal affixes, or bound pronouns, have many argumental properties that are held by full NPs in more familiar languages. Mithun (2003) demonstrated that bound pronouns function much like free pronouns of English and other similar languages. The issue of pronominal arguments is an important background but not the central concern of this article. Of primary importance is the second trend of research that originated in the work of Nichols (1986). How are semantic
I occasionally use the expression "personal affixes" in this paper. Note that this expression is a simplification; on some occasions these are not just personal affixes but also gender affixes, and sometimes this may be more important than person. However, for the sake of simplicity I still generalize this phenomenon as "personal affixes".
2


roles marked in a language like Abkhaz that lacks nominal cases (thesis (iii))? (This question is particularly relevant if one bears in mind that pronominal affixes on the verb rather than full NPs have many argumental properties.) The phenomenon of role marking by means of nominal inflectional morphology has long been taken for granted in linguistics. Even after Fillmore (1968) reinterpreted the traditional term `case' in a semantic vein linguists keep using the word combination "case marking" referring exclusively to the technique as in (1) and (2). Nichols (1986) drew general attention to the fact that the locus of role marking in a clause does not need to be restricted to nominal morphology. There exists a world-wide technique that is an alternative to the one of (1) and (2). Arguments' (and adjuncts') roles can be marked on the verb. Nichols observed this contrast very clearly when she compared languages of two North Caucasian language families: Nakh-Daghestanian and Abkhaz-Adyghean that are strikingly different in this respect. While the former use almost exclusively nominal cases to mark roles (Nichols dubbed this technique dependent-marking), the latter sticks generally to marking roles on the verb (head-marking). Compare the strongly dependent-marking Nakh-Daghestanian language Godoberi (4) with the head-marking Abkhaz (3): (4) Godoberi (Kibrik 1996b: 116) il-u-di wasa w-ali mother-Ext-Erg son(Nom) M-call.Past `Mother called up the son' The opposition between head- and dependent-marking techniques3 is a really fundamental typological parameter. This feature permeates the whole grammars of languages. As has been demonstrated by Nichols (1992), the locus of marking strongly affects other morphosyntactic characteristics of languages. Note that the locus of marking is totally independent of alignment type. Like Russian and Karachay-Balkar, Godoberi in (4) displays consistent dependent-marking4, typical of LENCA, but it differs from those two languages in using the ergative alignment. Abkhaz in (3) displays head-marking and ergative alignment; below I will demonstrate examples of head-marking combined with the accusative alignment. In Godoberi the fact that it is the mother (rather than the son or anybody else) who did the calling and is thus the agent of the event is established by the fact that the noun meaning `mother' bears the ergative case affix. This is what the dependent role marking is about. Let us see the working of the technique employed by Abkhaz. In Abkhaz, nominals are bare and contain no role marking altogether. All role marking resides in the verb. How do language speakers tell apart different referents when they encounter a structure such as in (3)? Consider the marking of the agent role pertaining to the referent "the man" in (3). right in front of the stem that indicates that a masculine singular referent plays the NP `the man' in the clause, as in (3a), any speaker of the language knows that it is the referent "the man" is known from the context (3b), the calculation is much the There is the prefix i- on the verb role of the agent. If there is a full the man who did the writing. If same.

The two trends of thinking address essentially the same circle of facts. Pronominal argument languages and headmarking languages coincide to a significant extent. However, these two lines of researchers largely ignore each other's results. Proponents of the argumental and referring status of personal affixes pay little or no attention to the fundamental characteristic of marking locus, while students of the typology of locus (see e.g. Nichols and Bickel 2005) talk about Abkhaz-type structure in terms of mere agreement. Any expression denoting a clause participants, be it a nominal or a pronominal, does two important things: first, it refers; second, it carries a certain semantic role with respect to the predication. These two functions cannot be reduced to one another, and none of them can be neglected. My point is that these two aspects must certainly be combined, and only a combination can yield a realistic picture of the phenomena in question.

The contrast between head- and dependent-marking can be observed in a number of various constutuents (Nichols 1986). In this article, however, I only discuss the locus of marking in the clause, that is, the marking of relations between the predicate and clause participants. For example, the expression "head-marking language" is to be understood below as `language that uses the head-marking technique to mark roles in a clause'. 4 All the three languages used for illustrating dependent-marking so far ­ Russian, Karachay-Balkar and Godoberi ­ are not pure examples of dependent marking. As the given examples demonstrate, these languages possess an element of headmarking as the verb `agrees' with one of the arguments in number, person, and/or gender. An example of a purer system is found in Japanese that marks roles exclusively on nominals; Japanese verb is devoid of any elements of head-marking.

3


In dependent-marking, nominal argument languages (the majority of LENCA) the referring and the role-marking functions are clearly distinct as they are fulfilled by different elements: referring primarily by nominal stems, rolemarking primarily by case desinences. In head-marking, pronominal argument languages it seems that personal affixes on the verb do both jobs. (Note that this combination is reflected in the structure of glosses: in (3) a gloss of a personal affix consists of two parts, the reference part and the role part, separated by a dot; this will be detailed below.) I believe that these two functions must be teased apart very clearly, in order to properly understand how these languages work. This article is about the head-based technique of role marking. In the rest of this paper I am going to discuss the following questions: · · What ensures role marking in such languages? How this technique of role marking should be understood in comparison to the dependent-marking technique?

My main claim is that the linear positions of personal affixes on the verb have essentially the same function as nominal cases. For this reason, I propose to understand and, accordingly, gloss the linear positions in the verb in terms of cases. Analogously to dependent-marking languages, head-marking languages can have the accusative, ergative, and agentive/patientive alignment (cf. Siewierska 2005). In accordance with the alignment type, headmarking languages have nominative, accusative, ergative, and other case linear positions in their verb templates.

3. Athabaskan
My argument relies primarily on the data from languages of the Athabaskan family. Athabaskan are of course outside of the ENCA area, but they are generally thought to descend from the last wave of immigration from Siberia into America (not counting the Eskimo-Aleut languages that still reside on the both sides of the Bering Strait). Athabaskan constitute the core of the greater Na-Dene family (that also comprises two languages of Southern Alaska, Eyak and Tlingit). Athabaskan are among the largest language families of North America; there are some 40 languages in this family. The Athabaskan family comprises three areal groups: Northern (Alaska and Western Canada), Southern, or Apachean (Southwest of the U.S.), and Pacific (U.S. Pacific coast). In some ways, Athabaskan languages are typical of North America ­ in particular, they are highly polysynthetic, head-marking, and arguably have pronominal arguments (see below). Nearly all grammatical meanings in Athabaskan are expressed by verbal morphology rather than by nominal morphology or function words. The roles of clause arguments are expressed exclusively by pronominal affixes on the verb. More peripheral roles can be expressed either on the verb or by means of postpositional phrases. In this respect, Athabaskan are quite representative of the North American type in general. In some other ways, Athabaskan are very unusual, both at the American scene and more generally. The most obvious peculiarity of Athabaskan is that they are almost exclusively prefixal. There are very few languages of comparable morphological complexity, if any, that would be so highly prefixal; see Kibrik 2002 Athabaskan examples in this article are taken exclusively from Navajo, a language of the Southern (Apachean) group spoken primarily in the states of Arizona and New Mexico. Navajo is representative of Athabaskan in general, and many other languages could provide comparable examples. In some respects Navajo makes a better example of consistent head-marking than more northern Athabaskan languages; this concerns the head-marking of peripheral clause participants, see section 7. Navajo examples have been obtained through personal communication with Bernice Casaus, New Mexico, unless otherwise specified. Athabaskan languages have very complex verb morphology. Usually the structure of the Athabaskan verb is described with the help of the notion of template (Kari 1989, Kibrik 1995; but cf. Rice 2000) ­ a sequence of linearly arranged slots, or positions, dedicated to certain classes of grammatical or lexical meanings and the corresponding morphemes. As Athabaskan are overwhelmingly prefixing, it is reasonable to count positions from right to left. The verb stem has number 0, and depending on language and theoretical approach the number of prefixal positions varies from about 10 to about 20. One frequently sees numerous prefixes in actual verb forms. Table 1 summarizes the linear order or prefixal morphological position in the Navajo verb. This table can be used for reference when particular linear positions are mentioned below. Note that some positions can hold more than one prefix in a verb form, so the template presented in Table 1 is a minimal one.


Non-locutor Nominative

Various derivational

Distributive

Locutor Nominative

Iterative/ repetitive

Qualifier

Oblique

Preverb

Mode

Position #

11B

11A

10

9

8

7

6

5

4

3

2

1

Table 1: Basic prefixal morphological positions in the Navajo verb

4. Core arguments and marking of their roles in Athabaskan
If a core argument is a locutor (first or second person) under normal circumstances no independent pronouns would be used in Athabaskan. The locutor arguments are expressed by pronominal prefixes on the verb. There are two morphological positions dedicated to marking locutor arguments in Navajo: positions #2 and #6. Navajo, as Athabaskan in general, has accusative alignment: single arguments of one-place verbs and agentive arguments of two-place (transitive) verbs are marked by pronominal prefixes in position #2: (5) One-place verb with an agentive argument dahnishjÀÀd `I jump' dahnishMorph. position # 3 2 upward- Impf1Sg.Nom-

jÀÀd jump

(6)

One-place verb with a patientive argument õinishgai `I am white, I whitened' õinishgai 2 Morph. position # Pref1Sg.Nomwhite Two-place verb, agentive nishteeh ÷Morph. position # 3.Accmonitored `I carry him (here)' nishõ- teeh 3 2 Impf1Sg.NomTI- handle.AnO

(7)

In examples (5) through (7) one can observe the first person singular morpheme sh- appearing in position 2 and translating as English `I'. In contrast to that, patientive locutor arguments of transitive verbs appear in position #6: (8) Two-place verb, patientive monitored [OEÒkééOE] shÜõteeh `He carries me (over there)' (e.g. an invalid speaking) shi÷niõ- teeh Morph. position # 6 3 [to.there] 1Sg.Acc3.NomImpfTI- handle.AnO

Not surprisingly, the first person singular morpheme appearing in position #6 and translating into English by means of the accusative pronoun me looks similar to position #2 sh- discussed above. The shape of this position #6 morpheme in example (8) is sh(i)-, where i represents a shwa epenthesizing between two consonants. Thus the two morphemes, sh- and sh(i)-, can be viewed as fundamentally identical. The same concerns the second person singular morpheme n(i)- in positions #2 and #6, although morphophonemics is much more complex. In the tradition of Athabaskan linguistics such morphemes are viewed as incorporated pronouns rather than agreement affixes (Sapir and Hoijer 1967, Young and Morgan 1987, etc.). There is significant evidence demonstrating that this traditional approach is correct (Jelinek 1984, Kibrik 1992, Mithun 2003). At issue is

Root 0

Morphological position

Transitivity indicator

Accusative

Reflexive accusative


another question: how are roles of the arguments marked in the clause? The tradition does not provide us a direct answer to this question. Apparently, the fact that the speaker referent is the agent in (7) and the patient in (8) is expressed plainly by the linear position in the morphological verb structure taken by the morpheme sh(i)-: position #2 and position #6, respectively. If one does not like the account of morpheme locations in terms of theoretical position numbers one can say that in (7) and (8) the first person singular morpheme appears on different sides of the imperfective prefix ni-. Functionally, position #2 is equivalent to the nominative case marker in languages like Russian or KarachayBalkar, and position #6 to the accusative case marker. If such functional equivalence is in place, my suggestion is that these two positions should be understood as the nominative and the accusative positions, and termed accordingly. This terminology is reflected in the glosses employed in examples (5) ­ (8) (as well as in the glosses of the Abkhaz example (3), mutatis mutandis). The gloss for each personal affix consists of two parts separated by a dot: in the first part, the morpheme's referential properties are indicated, in the second its role, in terms of cases. But it must be understood very clearly that role marking is done not by the personal affixes themselves but by the linear slots wherein they are inserted. The difference between the dependent-marking technique of Russian and Karachay-Balkar on the one hand, and the head-marking technique of Navajo, on the other, is that in the former roles are marked by material markers, such as case affixes, while in the latter by the linear order of morphemes inside the verb. In some instances there is a way to distinguish between the nominative and the accusative morphemes not just in terms of order but also in terms of material. For example, Navajo plural locutor pronouns differ both in position and in material shape. (9) Morph. position # (10) Morph. position # 3.Acc(11) Morph. position # niilk'eOE `We are ÷ni4 3.Acc- Prefnoõk'eOE ÷cooling him down' iidõ- k'eOE 2 1Pl.Nom- TI- cool

`You guys are cooling him down' niohõ- k'eOE 4 2 2Pl.Nom- TI- cool Prefis cooling us / you guys down' niõ- k'eOE 4 PrefTI- cool

nihiniõk'eOE `He nihi÷6 12Pl.Acc- 3.Nom-

The plural locutor pronouns occupy the already familiar positions #2 and #6, in accordance with the referent's role. In the nominative position, the first and second person pronouns have the underlying forms iid- and oh-, respectively. However, in the accusative position they have an entirely different form nihi-. This phenomenon of different material form is analogous to suppletivism of pronouns in more familiar languages, such as we and us in English. Note that the accusative plural locutor pronoun is semantically underspecified as it is does not distinguish the first and second persons. Taking all facts outlined so far into consideration it is obvious that the technique of differentiating roles by morpheme form is supplementary, while positioning in different linear slots is the main one. It is easy to imagine a head-marking language in which personal affixes would never depend on the role and only the linear position in which the affix appears would matter. It seems less likely to find a head-marking language in which linear position would be entirely irelevant, and only the shape of personal affixes would mark roles. Of course, relative contribution of these two role marking techniques must be a subject of an empirical typological research. In this article I rely on the assumption that the linear position technique is a much less idiosyncratic and more robust way to differentiate roles. In the discussion of Athabaskan evidence below I do not concentrate on differences in morpheme shape and restrict my argument to linear positions as the major role marking device. Now let us turn to marking of non-locutor (= third person) arguments' roles in Navajo clause. Unlike locutors, third person arguments in certain instances are not expressed on the verb. In particular, third person nominative is always


zero, see examples (8), (11) above, and third person accusative is zero when the nominative argument is a non-third person ­ see (7), (9), (10). A third person argument is expressed on the verb by a non-zero morpheme in the accusative position, but only if the nominative is a third person, too: (12) Morph. position # yiniõk'eOE `He is cooling her down' yi÷niõ- k'eOE 6 4 33.Acc- 3.Nom- PrefTI- cool

There are several reasons to assume, on systemic grounds, that third person arguments (even when marked by zero), like the locutor arguments, are expressed in the verb, and that roles are marked there rather than anywhere else. First, when arguments are represented by full NPs they do not bear any role marking. Second, there cannot be more than one zero-marked argument in a clause, and therefore the role of the zero-marked referent can always be calculated, relying on the explicitly marked argument(s) and the lexical meaning of the verb. Third, besides the plain third person illustrated above there are special subtypes of the third person that are expressed by non-zero morphemes ­ the so-called fourth person, the indefinite, and the areal. Consider an example of the fourth person (essentially a deferential third person) functioning as the nominative: (13) Morph. position # dzizghas ÷6 3.Acc`She (deferential) scrathed him' jiz- ghas 5 3 4.Nom- Pf- scratch

Note that the fourth person, when it appears as the nominative, turns out in a different position than the locutor arguments: not in position #2 but in #5. All variations of the third person (that is, non-locutors) that have explicit nominative marking appear in position #5. These two positions are in complementary distribution with respect to person and the two of them in combination constitute the nominative position; see Table 1 above. The presence of two complementary positions does not undermine but rather supports the analogy with nominal cases as dependentmarking languages frequently have several complementary affixes for the same case; cf. declension-dependent allomorphs of case markers in Latin. Thus I conclude that the roles of arguments are marked in the Navajo verb by means of linear positions of pronominal morphemes. Of course, this system system may be less that 100% efficient. One can imagine a situation when a pronominal affix remains ambiguous for role ­ for example, if the other argument is marked by zero and there is no affix in between the nominative and accusative positions. It is true that such instances may happen. But this does not mean that such system cannot function. Note that nominal case marking is also far from being totally efficient. For example, in Russian only a subset of nouns distinguishes the nominative and the accusative forms, and instances of ambiguity happen.

5. Terminology based on grammatical relations is misguided
The understanding of the linear positions' function as role markers, as outlined above, is not traditional in Athabaskan linguistics. Traditionally, the linear positions within the verb form are dubbed "the subject position" and "the object position" (Sapir and Hoijer 1967, Young and Morgan 1987), and this terminology is replicated in dozens of recent publications. In a broader contexts of the study of head-marking languages, the case-based interpretation of verbal positions occasionally occurs (e.g. Rude 1997 for Sahaptian), but it is rare. The majority of typologists use the conceptual system based on grammatical relations (subject etc.); see, for example, Mithun 1999, GivÑn 2001, Dryer 2005. This approach and the terminology reflecting it are typical of most descriptive and typological accounts of head-marking languages. To be sure, the question of the function of verbal positions or, vice versa, of the way how roles are marked on the verb, is rarely discussed explicitly. Typically, one can judge of the way how the given writer views these issues only on the basis of interlinear glosses used. If one sees a gloss like "1Sg.Obj" this is an indication that the writer believes that an adequate way to convey the function of the referent in the clause is in terms of grammatical relations (henceforth: GRs).


Examples below are taken from the very rich monograph by Marianne Mithun (1999) devoted to the native languages of North America. As North American languages are particularly inclined towards head-marking (see Nichols 1992: 69ff.), using the work by Mithun is particularly convenient. However, many other publications could be used to illustrate the point. In her analysis of Salishan and Chumashan data, Mithun (1999: 52; 208-9) discusses the verbal marking of roles in terms of grammatical relations. The glosses used are "3SUBJECT", "1PL.OBJECT" and the like. I argue against understanding personal affix positions on the verb in terms of grammatical relations "subject" and "object". There are serious reasons underlying this point of view. As has been demonstrated above, linear positions arguments' semantic roles. By analogy with more type devices, and not by grammatical relations. If grammatical theory they cannot serve one and the of personal morphemes on the verb serve to identify the familiar dependent-marking languages, roles are marked by caseone wants to distinguish cases and grammatical relations in same function.

To make this line of reasoning clearer, consider the situation in a familiar dependent-marking language. Let us take a Latin example: (14) Puer-Ü boy-Nom puell-am girl-Acc amat loves `The boy loves the girl'

In Latin, the distinction between role marking and grammatical relations is quite obvious. Roles are marked by means of case desinences, and case desinences serve exactly role marking. It would be odd to suggest that Latin case desinencs are identical with GRs. No one proposes to call Latin cases in (14) "the subject case" and "the object case", thus getting rid of case notions. Two sets of notions ­ cases (nominative, accusative, etc.) and GRs (subject, direct object, etc.) have elolved in basic linguistic theory, and the necessity to keep them apart seems obvious in application to a dependent-marking language such as Latin. Unlike the notion of case, the notions of grammatical relations serve to capture behavioral properties of arguments (or nominals), expressing both is clauseinternal