[This document can be acquired from a sub-directory coombspapers via anonymous FTP and/or COOMBSQUEST gopher on the node COOMBS.ANU.EDU.AU] The document's ftp filename and the full directory path are given in the coombspapers top level INDEX file] [This version: 4 August 1993] -------------------------------------------------------------------------- File 6. There are 9 parts of this thesis: from 0 to 8 -------------------------------------------------------------------------- A Grammar of Garadjari, Western Australia by Anna Kristina Sands 1989 Bachelor of Arts Thesis Department of Linguistics, Faculty of Arts Australian National University, Canberra ACT 0200 6 A COUPLE OF PROBLEMS 6.1 The Suffix -la The verbal suffix -la in Garadjari, briefly introduced in section 4.7.8 is the most problematic of the suffixes in Garadjari. It is a suffix that occurs on a large variety of verbs in any tense and its' distribution is therefore very much like the suffix -ra , the completive aspect described in 4.7.4. This suffix presents a problem because in Laves' translations of sentences which contain this suffix there is no clue as to what it might mean. Given the functions of suffixes on verbs in Australian languages there are basically three possibilities that this suffix could represent. It may be a discourse connective, a tense / aspect suffix or a bound pronoun cross-referencing suffix. These possibilities will be discussed below. 6.1.1 Discourse Connective Discourse connectives in Australian languages link sentences within a discourse. Capell [ 1962 ] describes, for the Waljbiri language ( later called Warlpiri ), a suffix -la , added to the verbal stem, which carries the idea of either simultaneity or immediate sequence to two actions. We also get in Yindjibarndi [ Wordick, 1982 ] a suffix -la as a conjunction on verbs. This suffix Wordick describes as the locative case being used on verbs to link sentences. Thus, in two languages of the area surrounding Garadjari, we have the suffix -la acting as a discourse connective. It is also interesting to note that we have a language on the eastern side of the continent with a similar suffix. Ngiyambaa [ Donaldson, 1980 ] has a verbal, post-tense, suffix -la alternating with -langa which indicates that the action is subsequent to the previous action. Thus, given that we get -la as a discourse connective in other languages, both closely related to Garadjari and widely divergent, and the wide variety of situations in which we get the suffix -la in Garadjari, it does appear that it is possible that -la may be some kind of sentence co-ordinator. Garadjari does, however, already have a co-ordinating suffix -gudji which links sentences such that the second action occurs subsequent to the first. This is not to say, however, that we might not have a second co-ordinator indicating, say, simultaneous action. There are, however, several factors that suggest that -la is not a discourse connective. Firstly we get both the co-ordinator -gudji and -la on the same verb. eg. (1) gaunda-ra-ya-ra-la-gudji call-CONT-3 Pl S-COMP-la-then thereupon they called out If these were both discourse co-ordinators with different meanings, ie. one indicated subsequent action and the other simultaneous action, this would represent a contradiction. So, unless -la were a connective that also indicated subsequent action this would not be possible. A second point against a discourse analysis of -la is that the two other sentence co-ordinators in Garadjari, the suffix -gudji and the particle gala are quite clearly indicated by Laves in his translations. He obviously, therefore, knew about co-ordinators in the language and it seems likely that if he picked up two co-ordinators he would know about any others. The final argument against such an analysis is in the usage of -la in the texts. There is no pattern that can be found to its usage in regard to linking sentences and in fact the only pattern that can be found is that on certain verbs it occurs all the time ( this will be discussed in 6.1.3 ) and as such appears to be related to individual sentences and not to linked clauses. Thus, there are no real arguments that would link -la to a discourse connective and there are arguments, though not conclusive ones, against it, so we conclude that -la in Garadjari is not a discourse co-ordinator. 6.1.2 Tense / Aspect Suffix The second possibility is that -la is a tense / aspect suffix in Garadjari, perhaps somewhat like the completive suffix -ra . On the verb in a sentence, in Garadjari, we may get the suffix -ra : eg. (2) bargara ya-rnu-ra turkey go-PAST-COMP a turkey came along or the suffix -la ; eg. (3) bu-ngu-bilinja-la cut-PAST-3 Du O-la he cut them two or both the suffixes -ra and -la in combination. eg. (4) gurga dja-lgu-warngungku-ra-la rise(AUX)-FUT-REFL D-COMP-la he got himself up Now the verbs that may occur in sentences like (2) and (3) here are in complementary distribution. There is no verb, in the data, which occurs with only -ra in one sentence and only -la in another. This fact is highly suspicious and suggestive that -la and -ra are actually allomorphs of the same morpheme. However, if we look at the verbs with which we get -ra and compare this to those with which we get -la there is no regularity with regard to conjugation nor transitivity. Look at the following : -la -ra l-conjugation gama-rnu 'taste' Tr yinda-rnu 'tell' Tr 0-conjugation djai-nju 'come' Intr garli-nju 'dig' Tr ng-conjugation bara-ngu 'run' Intr Nja-ngu 'see' Intr Here we see that we get both -la and -ra on verbs of each conjugation and on both transitive and intransitive verbs. We note that there is no phonologcal conditioning for the difference, as we may get either suffix in combination with any other suffix and we would not expect phonological conditioning by the verb root because of the great distance that may occur between the verb root and the suffix ( note that up to four or five suffixes may intervene ). We also note that in no other segment of the language do we have an alternation, either conditioned or in free variation, between a rhotic and a lateral. Now, in a language in which all allomorphic variation is phonologically or conjugationally conditioned, we would not expect an alternation -ra /-la to be neither phonologically conditioned nor conditioned by any other overt factor such as conjugation or transitivity, but to be based purely on the lexicon. We can thus conclude that, despite the fact that we get -la and -ra on a different set of verbs, this is purely co-incidental, and we do not have allomorphic variation between these two. We are, thus, left with the question of whether -la could then represent another tense / aspect. If we have a completive suffix -ra appearing after the cross-reference suffixes, why not another suffix -la ? The problem remains as to what aspect -la could represent. We already have a completive and a continuous aspect and Laves does not give us any clue in his translations as to what it may represent. 1 Thus anything we said about -la would be pure guesswork and not based on known facts. We can also gain no help from nearby langauges as I can find no tense / aspect suffix -la or similar. It therefore seems unlikely that -la is a tense / aspect suffix. 6.1.3 Cross-Referencing Marker In its known usages -la is the cross-reference marker for the third person singular dative. eg. (5) djanba-gu ya-rnu-ra-la djanba-DAT go-PAST-COMP-3 Sg D he went along tracks of the Danba man It therefore cross-references an oblique case as grammatical relation to the verb. Is it not possible, therefore, that it may cross-reference more than the dative case and be extended to cover other oblique cases ? The first thing to note with reference to -la is that there are certain verbs which always co-occur with the suffix. ie. these verbs never appear without it. These verbs are : bu- = to cut gama- = to taste bira- = to cover over djanima- = to cover There are also many verbs which never apppear with -la , despite being very common verbs. Now of the above verbs bu- appears some twenty times, gama- ten times and bira- and djanima- some four or five times each. It is thus, particularly for the first two, not an insubstantial point that they always appear with -la . Now where you have verbs that always appear with a particular suffix this suffix is likely to be a cross-referencing suffix for an argument that is specified by the verbs valency. Note that, for example, a transitive verb has two slots in its valency, a subject and an object slot, and these are always cross-referenced with subject and object cross-referencing markers. Thus, we could say that the above four verbs have associated with them a grammatical relation, not subject or object, that is cross-referenced by -la . We do, however, have a problem with this in that these verbs also have a subject and an object. If we look at the following two sentences with bu- : (6) bu-ngu-bilinja-la cut-PAST-3 Du O-la he cut them two (7) bu-ngu-bula-la cut-PAST-3 Du S-la they two cut it ( Note that as the third person singular subject and object cross-reference markers are zero they do not appear overtly here. ) We see here that we have an object in (6) and a subject in (7) cross-referenced by the regular suffixes. Thus we would have to say that -la cross-referenced a third grammatical relation and that this never appears overtly in the sentence as there are no sentences with bu- that have a non subject / object noun phrase that could be claimed to be cross-referenced. With the verb djanima -, however, we have one sentence which may give us a clue. ie. (8) djanima-ri-la djura-ngka cover-CONT-la fire-LOC he covered it in the fire Here we have a locative noun phrase that is clearly not the subject or object. For djanima- it seems conceivable that -la here cross-references the locative noun phrase. We also have another sentence in the data where we get the locative co-occuring with -la . ie. (9) djayi-njari-la warlibi-ngka come-CONT-la camp-LOC he came into camp We can compare (8) and (9) with sentences with the locative which do not have the suffix -la . eg. (10) mirling bala-rnu banji-ngka warayin-dju diyar spear pick up-PAST hand-LOC one-ERG diar the other diar picked up a spear in his hand (11) guyi nja-ngku-bula djigur-dja animal see-FUT-2 Du S treehole-LOC they saw an animal in the treehole The difference between the first set of sentences and the second is in the semantic cohesion of the locative to the verb. In the first set it is very closely linked to the verb. If we were to remove the locative phrase from either (8) or (9) we would ask a natural question ; 'with what did he cover it ?' or 'where did he come to ?'. However, in the second set, the locative noun phrase could be easily removed and we would not think to ask, for example, 'with what did he pick it up ?'. We might like to say, therefore, that the dative -la also functions as cross-reference marker to a locative relation. To test this theory it would be nice to find a plural locative relation cross-referenced with the dative plural marker. However, I have found no examples of a plural locative relation. Thus far, it seems that -la cross-references the third person singular dative, possibly a locative grammatical relation and some unknown grammatical relation(s). If we look at languages of the surrounding area we can gain some extra evidence for -la as a cross-reference marker for the locative. In Nyangumarta [ Geytenbeek, 1988] , a Marrngu language along with Garadjari, the bound pronoun -lu ( harmoninsing to -la or -li according to the preceding vowel ) cross-references both the third person singular dative and locative functions. Thus, given that Nyangumarta is Garadjari's closest relation, it would not be unexpected that in Garadjari too the suffix would cross-reference a locative. We also find that in Gurindji [ McConvell, 1980] , a Ngumbin language, there is a suffix -rla ( note that this is a post-alveolar lateral as compared to the alveolar lateral in Garadjari ). This cross-references the third person singular dative whereas the dative of all other person number combinations are cross-referenced by the normal object clitics. We also find that the object clitic in combination with -rla cross-references a locative, allative or elative with animate noun referents. Thus, in Gurindji the suffix -rla has no equivalent in other person and numbers. It is not a part of a paradigm set. Now if we had a similar situation in Garadjari with -la not as a part of a paradigm set ( where it cross-references the locative and possibly other ( unknown ) oblique grammatical relations ), but simply an extra suffix that is tacked onto the object cross-reference suffix this would help to explain why -la is the only suffix that presents a problem. If -la were a part of a paradigm set we would expect that we would get, in the data, at least one example in a person-number other than the third person singular. However, we do not. Now, this would also help in the analysis of those verbs like bu- which always appear with -la . I have noted that bu- appears with the object cross-referencing suffixes. If, as I propose is possible, -la is tacked onto this object suffix to indicate an oblique grammatical relation then bu- has only two arguments cross-referenced for - not the three previously suggested - and the fact that we never get a third noun phrase in the sentence is no longer a problem. Unfortunately, there are no examples in the texts of the verb bu- with an overt noun phrase 'object' or patient, ie. the thing that is cut. There is, therefore, no example that can tell me whether the case ( rather than the cross-referencing marker ) of the thing that is cut is absolutive ( ie. the object case ) or some other case. However, we could say that bu- is not a transitive verb but is a semi-transitive or "middle" verb. It takes an absolutive subject 2 and another argument ( perhaps in the dative or locative case ). Now we see that in Nyanguamarta we get such semi-transitive verbs as well ; verbs that have two core arguments though one of these is in the dative or locative case with dative / locative cross-referencing on the verb, ie. they are cross-referenced with the Nyangumarda equivalent of the suffix -la . Thus it would come as no suprise to find a similar situation in Garadjari. Thus far the suffix -la has been associated with the dative, the locative and possibly some other case when these cases appear on core nouns as opposed to peripheral nouns to the verb, ie. we have the dative indirect object, the locative that is semantically coherent with the verb and the second argument specified in the valency of verbs like bu- . However, we are still left with a large number of occurences of -la which cannot be so explained. For example, there are two sentences in the data, with -la , which appear, on the surface, to have no subject. (12) njim bira-ri-la bundju fog cover-CONT-la grass fog covered the grass (13) ngaba yilba-nju-la djirbal water throw-PAST-la cloud it rained These sentences appear to be transitive, in that they have two arguments, however, they have no subject in the ergative case. (12) includes the verb bira- , already described above as a verb that always takes the suffix -la and is problably a semi-transitive verb. We see that the non-transitivity of the verb is backed up here by the lack of the ergative suffix. (13), however, includes a verb yilba- 'to throw' which is very clearly a transitive verb in other situations. ie. (14) mirin-dju yilba-nju mirling ngura-ngka Mirin-ERG throw-PAST spear ground-LOC Mirin threw the spear on the ground Thus, we have a transitve verb that normally appears with an ergative subject without -la , appearing here with an absolutve subject and the suffix -la . We might like to say, therefore, that the suffix -la is some kind of de-transitivizer demoting a transtive subject ( A ) to an intransitive subject ( S ) . It would therefore behave like an antipassive. There are two problems with this. First is that in an antipassive construction you find that the object is demoted to an oblique case, thus avoiding the possible ambiguity arising from two noun-phrases in the absolutive. Here, however, we find that no such demotion takes place. We also find that in languages of the surrounding area we have no anti-passive construciton and with the possible exception of (12) there are no other examples in the data. If we therefore assume that this is not an example of an anti-passive we are left with the question of what is happening. It is interesting to note that sentences (12) and (13) both represent natural phenomena. We can add another sentence to this set. ie. (15) barda ngalba-nju-ra-la sun set-PAST-COMP-la the sun set This sentence is intransitive as compared to (12) and (13) which are transitive but these are all actions that occur quite naturally with no form of volition or control on the part of the "subject". They are also the only examples of natural phenomena in the data. If the ergative is the case of the Agent, an actor who does something, then perhaps the ergative case is omitted in (12) and (13) becasue the subject does not really do anything but is simply following its natural path. We are still left with the question, though, of what is the function of -la here. Is -la the cross-reference marker for the non-ergative subject? for the object? or something else entirely ? To help answer this we can look at other sentences within Garadjari and at other languages of the area. The following two sentences that contain the suffix -la appear in the data : (16) marala djinka-rnu-ya-ra-la-ngala Marala kill-PAST-3 Pl S-COMP-la-might they must have killed Marala (17) gala wayini bara-ngu-la then frightened run-PAST-la he ran away frightened These are two verbs that are very common in the texts but normally appear without the suffix -la . Now, here, we do not get any change in the sentence accompanying the suffix, unlike (13) where the subject is de-transitivied. 3 These two sentences, however, have something in common and that is that they both assume a cause or a reason, for the action. To have killed someone there must have been some reason for the killing and to run away frightened there must be something that caused the fright. We could say that there is a third, unstated but implied, noun phrase to the sentences and perhaps -la is the cross-reference marker for this hidden noun phrase. 4 If we look to other languages of the region we find that in Warlpiri [ Hale, 1982 ] , a Ngarga language, we have a suffix -rla which is an adjunct dative cross-reference marker that cross-references the benefactive, possessive, or causal noun phrase. eg. (18) maliki-rli ka-rla kurdu-ku kuyu marlaja-nga-rni dog-ERG PRES-rla child-DAT meat causal-eat-NPAST the dog is eating meat because of the child Note the fact that, despite the fact that this suffix is post-alveolar, it is cognate with the Garadjari alveolar. A similar change, l --> rl , has occured in the ergative suffix. Now in Walpiri the suffix -rla cross-references a causal noun when in combination with the causal particle marlaja . This is not to say, however, that in Garadjari we might not get -la cross-referencing a causal that has no other causal marking. Djaru [ Tsunoda, 1981] , a Ngumbin language, has a suffix -la which cross-references the indirect object of a sentence. One may also add it to any sentence to indicate the purpose, beneficiary or possessor. It may, thus, also be used to indicate the purpose, or cause, of an action. Thus, in at least two other languages of the region, languages that are related to Garadjari, we have -la , or a cognate of -la , as a suffix cross-referencing the cause of an action. For neither of these languages is it mentioned that you may get the suffix without an overt noun phrase that it cross-references, but in Garadjari the deletion of noun phrases is very common and so if -la appears without a noun phrase it can be no more surprising than if, say, -djinja ( third person plural object marker ) appears without a noun phrase and this is very common. We can now return to sentences (12), (13) and (14) where we have a natural phenomenon. To say that -la here cross-references a causal noun to the sentence would obviously be incorrect, as if there is no volition on the part of the subject there is not likely to be a cause. But we see that in both Djaru and Warlpiri -la may cross-reference a beneficiary as well as a causal. And, in fact, we may also link -la to the beneficiary in Garadjari. eg. (19) yira ma-rnu-djanagura-la name(AUX)-PAST-3 Pl B-la he named it for them Here the beneficiary function is marked by -djanagura alone, ie. the -la is not an essential ocmponent. However, this does show that -la quite happily co-occurs with the benefactive. We also note that Capell [ 1962 ] describes the dative and benefactive as interchangeable. Now in a story where the action described revolves around a central character, or characters, it is to be assumed that anything described in that story will be in some way associated with that character. For example, if the sun sets, it is so that the hero may go to sleep, or some such thing. We may like to say, therefore, that in this group of sentences -la is cross-referencing a beneficiary, or an affected noun. In assuming this we remove the problem of trying to associate the suffix -la with the non-ergative subject and thereby introducing a seperate, non cross-referencing syntax to the suffix. In summary we have -la appearing in the following situations : a) as third person singular dative cross-reference marker for a dative grammatical relation. b) in combination with the object cross-reference marker as cross-reference marker for a locative grammatical relation. c) in combination with the object cross-reference marker as cross-reference marker for a non-subject / non-object grammatical relation to verbs such as bu- which always appear with -la . d) on verbs that describe natural phenomena to cross-reference or indicate a benefactive or affected human noun. e) on ( any ? ) verb to cross-reference or indicate a cause or reason for the action. In all of the above the suffix -la may, and most often does, appear without an overt noun phrase in the sentence which it cross-references. For this reason it is very difficult to pinpoint the actual function(s) of the suffix -la and the above analysis is only put forward as a possible analysis given the data and information from surrounding languages. The above situations, a) to e), may be divided in two manners. In a), b) and c) we have -la cross-referencing an oblique grammatical relation ( ie. a core argument to the verb that is not A, S or O ) that is essential to the verb, while in d) and e) -la functions as cross-reference marker or an inidicator 5 of a non-core or peripheral adjunct. The second manner in which we may divide the usage of -la is in whether it occurs alone, as in a), or in combination with the object suffix as in b) and c). For d) and e) we cannot determine whether -la appears alone or in combination because as the third person singular object suffix is zero this does not show up overtly on the verb. If all of the hidden referents in d) and e) are third person singular, as is likely, -la could be appearing to appear alone when it is actually in combination with the object suffix. From this we can divide the suffix -la into possibly three different suffixes : 1) As third person singular dative cross-referencing suffix within the dative paradigm. 2) An indicator of a non-core grammatical relation that is added to a verb with an object cross-reference marker. 3) Added to any verb ( whether or not with the object suffix ) to indicate a cause or affected noun. These suffixes all appear after the completive aspect -ra but before the suffixes -ngala and -gudji . eg. (20) marala djinka-rnu-ya-ra-la-ngala Marala kill-PAST-3 Pl S-COMP-la-might Marala must / might have killed them 6.1.4 Origins of -la There is one further question that can be discussed in relation to the suffix -la and this is from where did it originate? As the dative third person singular cross-reference marker it differs from all other cross-reference suffixes in that it appears after the completive aspect -ra rather than before it, ie. compare sentence (20) above with the sentences in section 4.6. Suffixes 2) and 3) above differ from other cross-reference suffixes not only in the position on the verb but also in that they do not from a part of a paradigmatic set and they appear in combination with another cross-reference suffix, the object suffix. -La thus differs grammatically quite markedly from cross-reference markers though semantically it seems to behave like one. From this the question is raised as to how and from where has this suffix come from ? As we get a similar suffix with similar behaviour in surrounding languages the origins of the suffixes -la are not within the Garadjari language but at some stage in its history when it was one with the languages that are now its closest relations. However, when it arose is not the quesiton of interest. Now, the suffix -la appears on the verb after all tense and cross-reference markers and so does the purposive suffix -gu ( note, however that the purposive suffix appears after the suffixes -gudji and -ngala whereas -la appears before them ). This is a suffix that has been related both in form and meaning to the dative suffix on nouns. This similarity has been noted by Dixon [ 1980 ] . He states, for example, that in Djirbal, a language from Queensland, the suffix -gu may be added to a noun phrase in the sentence or to the verb of a complement clause with very similar semantic effect, ie. 'I'm going for fish' versus 'I'm going to spear fish'. This is exactly the type of similarity that I have noted between the dative and purposive in Garadjari in section 4.7.7. It is therefore, likely that the purposive is derived from the dative suffix being used on nominalised verbs that form the second clause, the 'in order to' clause of a sentence. In Garadjari this has been extended and we now get the purposive suffix on the single ( and therefore verbal ) verb of a sentence. Its position on the verb, following all inflectional endings is consistent with the idea that the purposive arose from the dative case being used on verbs. So it is with -la . Could not -la , also, have arisen from a nominal case being extended to the verbs ? Though the locative case in Garadjari is -ngka and thus nothing like the -la found on verbs, it is the common pattern in Australian languages to have the locative differ from the ergative only in a difference of a final [ a] in contrast to a [ u] on the ergative. If this were the case in Garadjari we would expect the locative case to be -la after a vowel and this is exactly the form of suffix that we are looking for. Now, if we look to other languages of the area, we find that -la as an allomorph of the locative is very common. In Badimaya [ Dunn, 1988 ] , Luritja [ Heffernan, 1984 ] and Yankunytjatjara [ Goddard, 1985 ] we get an allomorph -la of the locative appearing on proper names, kin terms or pronouns, ie. -la appears as the locative case with human referents, while -ngka appears on other nouns. In Djaru [Tsunoda, 1981 ] we get -la on trisyllabic or longer vowel final nouns and in Yulbaridja [ Burridge, ? ] we get -la on vowel final stems alternating with -tya on consonant final stems. This is just to name a few. We would, therefore, expect that the form -la was definitely a locative case suffix in the proto-language of the area. Now the fact that the conditioning environment for -la is different for these languages presents no problems. If -la , in the proto-language when -la apppeared on verbs, were conditioned by the human reference of the noun, we note that where cross-referencing occurs it would usually cross-reference a human referent. All verbal inflections end in a vowel and thus, if vowel final is the conditioning environment for allomorph -la , ( as it was in proto-Australian ), then we would get the -la allomorph used on verbs. Thus, phonologically, we have a match between the locative case on nouns and the suffix that is causing the problems on the verbs. However the similarity between the locative and the suffix -la becomes even more apparent when we look at the semantics of the locative case in the surrounding languages. In Badimaya, Djaru and Yankunytjatjara at least, the locative case covers the semantic range of a cause. For example in Yankunytjatjara we have the following : (21) ngayu-lu wari-ngka tjitininga-nyi 1 Sg (NOM) cold-LOC shiver-PRES I'm shivering in / because of the cold Thus, not only does the locative case match phonologically, but semantically, with the suffix -la found on verbs. It would, therefore, be quite reasonable to assume that the suffix -la were originally the locative case suffix which somehow came to be expanded in its usage and was added to a verb. The evidence for this lies in the similarity between the locative case and the suffix -la , at least in the non-dative usage, ( note that the suffix -la covers the locative grammatical relation and the causal ), the position of -la on the verb, after all inflections, and the fact that -la is the only suffix of its kind and does not form a part of a paradigm as do all other cross-referencing suffixes. Whether the dative cross-reference suffix -la has its origins in the locative, expanding to cover the dative, is unknown. It does, however, seem likely that the dative -la had independent origins and its unusual position on the verb, in Garadjari, after the suffix -ra could be attributed to analogy with the other suffix -la . There is one further point to note in relation to the above discussion and this is that the above assumption, that the locative case could be expanded to occur on verbs is not pure assumption, and this very thing does in fact occur in modern Garadjari. Two sentences in the data appear to take the locative suffix on the verb. ie. (22) yaga-nba-li maba djari-ngka bidi-ngka leave-REC PAST-1 Du S lost(AUX)-LOC hole-LOC we must have lost it in the hole (23) ga-ngku-djanagura-ngka ngaba binka-ngka carry-FUT-3 Pl B-LOC water conch-LOC he brought them water in a conch shell What is actually happening in these sentences is unknown from such limited data but it appears that the locative on the verb seems to have something to do with the locative noun that appears in the sentence. It is through just such a situation that the verbal suffix -la may have arisen. 6.2 The Ergative Construction or Double Case Marking It has been mentioned in chapter 3 that the ergative suffix may be used in conjunction with other cases. This construction is interpreted by Laves as a 'while' clause, ie. a simultaneous subordinate clause where the case utilised with the ergative is the allative or perlative. eg. (24) wulguru-gardi-lu nga-njari ngaba-ngka shade-ALL-ERG eat-CONT water-LOC while going to the shade they drank at the well (25) warlibi-nguru-lu gardaga camp-PERL-ERG sleep while they were in camp they slept We also get the verbal particle marndi in combination with the ergative case translated with a 'while'. eg. (26) marndi-lu djungku-ngka yaga-rnu-bula going-ERG fire-LOC leave-PAST-3 Du S (while) going away they left it on the fire However, where the case is the locative or ablative Laves translates these with a relative clause. eg. (27) gulja-ngu-lu ngaidju yandja south-ABL-ERG I north the man from the south / the southerner said ' I am going north' (28) gulja-ngka-lu ngana nadja south-LOC-ERG what here he in the south said 'what's here ?' Now, as these constructions are so similar, we would perhaps like to attribute them to a single construction. This will be discussed further, but first we must look to the functions of case. 6.2.1 Functions of Case [ Dench and Evans, 1988 ] The possiblity of having double case marking comes about because case may have different functions within a language. The classical function of case is the inflectional use to relate a noun phrase to the verb. If this, however, were the only function that case covered we would never get double case marking on a single noun. A second function that case may cover in Australian languages is an adnominal role relating two noun phrases within a noun phrase. The genitive case is the classic and common example of this as it relates one noun phrase as possessor to another noun phrase. A third function of case is a referential function where a noun phrase or adverb is marked in agreement with another noun phrase in the same clause. It is common with this function of case to have subject oriented referential function where another noun phrase of the clause is marked in agreement with the subject. One further function that case covers is in a complement clause where the case establishes a relationship between the complement and the matrix clause. 6.2.2 Double Case Marking As each noun phrase in a sentence is generally marked for case relating it to the verb, in an Australian language, where case has two functions in a language you will generally find that you may have the possiblity of double case marking on a noun. For example in Kunjen the ablative case may have an adnominal meaning and you get double marking. eg. (29) abm onalkal-ngand-iy egalng arir adhun person island-ABL-ERG head hit me an islander hit me Where you get double case marking the major relational function always follows any secondary function of case and thus in (29) above the ergative suffix relates the noun phrase to the verb 'hit' as the subject, and not the ablative suffix. 6.2.3 The Situation in Garadjari The situation that we have in Garadjari, of double case marking, is thus not unique in Australian languages and the clue to the interpretation of this construction must lie in determining what other function(s) case may cover in Garadjari. If we look to Laves' two different interpretations of the double case marking construction we see that his interpretation of the allative and perlative assumes a noun phrase referential function, ie. the double case marked noun is marked as referential to the subject, while his interpretation of the locative and ablative assumes an adnominal function where one noun phrase is related to another within the noun phrase. Note, however, that in the data the second noun is absent and we have a headless relative clause. Thus, Laves appears to divide the cases to cover different secondary functions. However, if we look at the situation more closely we see that perhaps this is not so. We can compare, for example, the following two English sentences ( possible translations for (28) ) : a) the man ( who is ) in the south . . . b) the man, while being in the south . . . We see that the difference is not so great. In a) we have a relative clause where we specify more information about the subject whereas in b) we specify the temporal setting of the event [ Hale, 1976 ] . They are both relative clause type constructions. Thus, it would not come as a complete surprise to find Laves interpreting the same construction in the two above manners. The question remains as to what function, or functions, does case cover in Garadjari, and thus exactly what is the construction that we have. 6.2.4 Adnominal Function of Case In Laves' relative clause interpretation of the ablative or locative plus ergative Laves assumes an adnominal function of the ablative and locative cases. The adnominal use of case marks one noun as subordinate to another and we have a relative clause that gives further information about the head noun. Garadjari is not unique in the languages of the area in assuming such a funciton of case. Walmatjari, a Ngumpin language, has a similar type of construction with the ergative suffix being used in conjunction with other cases and here Hudson [ 1978 ] interprets this as a relative clause type construction regardless of the case that is associated with the ergative. We also find that in Martuthunira [ Dench, 1987 ] case has several functions, one of them adnominal, and in Warlpiri [ Simpson, 1983 ] too. Thus from the fact that it does occur elsewhere in related languages we might like to say that Garadjari has an adnominal function of case and try to extend a relative clause interpretation of the 'ergative' construction to the perlative and allative cases. There are, however, problems associated with such an interpretation. The first argument against a relative clause like interpretation is that we get this construction in places where we would certainly not expect a relative clause, ie. it occurs where we get subjects already known to the listener and it does not serve to distinguish the topic from amongst a possible group. eg. (30) yungu-njari-bula-ra guyi , nga-njari wulguru-nguru-lu give(AUG)-CONT-3 Du S-COMP animal, eat-CONT shade-PERL-ERG they gave him meat, he ate it while in the shade (31) njirani-djangka-lu bina gari-nju-ya behind-LOC-ERG hear(AUX-CONT)-PAST-3 Pl S while being behind they heard a sound Sentence (30) is self explanatory but in (31) we are talking about a single group of people, ie. we are not distinguishing 'they who are behind' from another group of people, and thus this does not look like a clause that gives more information about the subject. The second argument against stating that this construction is a relative clause is that we only get this construction in conjunction with the ergative suffix. We never get gulja-ngka etc. interpreted as a relative clause unless this is followed by the ergative suffix. Thus, unless this is an accident of the data, the ergative case is an essential component of the supposed relative clause construction. If this were truly an example of relative clause formation we would expect to get it in all cases and find at least one example of an intransitive subject or an object in the absolutive case which is a relative clause. We note that in the data there are some twenty examples of the relative clause plus the ergative but not others. 6.2.5 Referential Function of Case The second way that Laves translates double case marked constructions in Garadjari is as a 'while' construction. This can be seen to be actually subject agreement as it is always the subject that does something 'while' doing something else, and thus we have a referential usage of case with the 'while' clause related to the subject. As we only get the double case marking where the second case is ergative we would have to say that if Garadjari has a referential function of case this is triggered by the subject and is not a general phenomenon. This subject triggered reference function of case also occurs in Yankunytjatjara [ Goddard, 1985 ] and Pintupi [ Hansen, 1975 ] . In Pintupi the second case of a sequence is called the same participant marker and is the ergative suffix -lu where the main clause is transitive and the absolutive suffix zero where the main clause is intransitive. We have the same situation in Yankunytjatjara and if we look at an example in this language we see that it behaves semantically very much like Laves' 'while'. eg. (32) wati-ngku kungka tali-wanu-ngku nja-ngu man-ERG woman(ACC) sandhill-PERL-ERG see-PAST the man saw the woman ( by looking ) across the sandhill We could almost say 'the man saw the woman 'while' being at the sandhill' . Now if Garadjari shows actor agreement of a kind as found in Pintupi and Yankunytjatjara we would expect that the ergative case would be used only with transitive clauses and the absolutive would be used with the intransitive, though as the absoutive suffix is zero, this does not show up overtly. We do, however, get the ergative suffix -lu used with both transitive and intransitvies in Garadjari. eg. Transitive : (33) nga-njari ngaba wulguru-gardi-lu eat-CONT water shade-ALL-ERG he drank water while going to the shade Intransitive : (34) warlibi-nguru-lu gardaga camp-PERL-ERG sleep he slept while in camp We thus have a problem with trying to state that the ergative case here is showing actor agreement with the subject of the main clause as the subject of the main clause in (34) is in the absolutive case. 6.2.6 A Solution We are thus left with a problem. The ergative construction in Garadjari does not seem to conform to an adnominal usage of case nor a referential funciton and the only other remaining function of case in languages of Australia, as a complementizer, appears to have no relationship or similarity to the Garadjari data. Thus we are left with two options. 1) to assume that the suffix -lu appearing after another case suffix is not an example of the ergative case but is a quite unrelated suffix. 2) to solve the problems with either the adnominal or referential function of case. Now option 1) above seems highly unlikely as the suffix -lu may function exactly as the ergative. We can compare the following : (35) Gargidja-lu ngana nadja Gargidja-ERG what here Gargidja said 'what's here' (36) gulja-ngu-lu ngayidju yandja south-ABL-ERG I north the southerner said 'I am going north' Here we see that -lu as the ergative in (35) and as the second case in a sequence in (36) both imply that the missisng verb is understood to be 'to say / tell'. We are thus left with the second option of finding a solution to the adnominal or referential function of case. As the double case construcion appears only with the ergative case, ie. a subject case, it seems more likely that a referential with actor reference is the solution rather than an adnominal solution. We thus have to explain how a sentence such as (34) is possible with an intransitive main clause triggering ergative ( or transitive ) agreement on the noun. Now the first thing to note is that not all examples with the allative, case etc. do show the suffix -lu . We have examples where we could possibly imagine the invisible absolutive suffix in agreement with an intransitive subject. eg. (37) gagara-ngu-0 warara gurga dja-rnu-ra east-ABL-ABS old man arise(AUX)-PAST-COMP the old man from the east arose (38) widi ngaba-nguru-0 widi play water-PERL-ABS play they played while at the well The translations given here are in accordance with the translations that I imagine that Laves would have given, had there been overt marking of a second case. Thus, it appears that sentences such as (34) may be an exception and in fact we do possibly have absolutive agreement as well as ergative agreement. A second point to make is that in one other segment of the language do we have the ergative case being used where it should not on intransitive subjects and this is in the pronouns. It has been noted in section 3.3.1 that mix-ups occur with pronoun subjects and you may get either the ergative or absolutive case on transitive and intransitive subjects alike. As, in the data, the ergative construction appears with no overt noun that it is co-referential with, this missing noun, or subject of the main clause, could be pronominal. Note that in the examples we are talking about the 'hero' of the story, a person who is thus referred to continuously by a pronoun in the texts. Thus, if this pronominal subject may take either case, then the thing that shows agreement with it should, in theory, take either case. 6 Thus, example (34) need not present such a problem to the theory and we can assume that case in Garadjari, as well as the relational function, functions as a co-referential marker in agreement with the subject on certain nouns in one of the four cases; ablative, locative, perlative or allative. It is examples of the above cases that describe an action of being or going that is simultaneous with the main action that take actor agreement, and not examples where the case describes a relational function to the verb. We can compare : (39) ngaba nga-njari wulguru-gardi-lu water eat-CONT shade-ALL-ERG he drank water while going to the shade (40) bara ma-rnu-gudji marla-gardi run(AUX)-PAST-then anthill-ALL then he ran to the anthill Here, in (40) we have no actor agreement because the allative case is here in the relational function of 'towards'. In (39) , however, you do not 'eat towards' as you do 'run towards' and the allative is interpreted as an action of 'going to the shade' and we get actor agreement. With the locative case, which may be used to describe the location of the action or of the subject, we would only get actor agreeement where the subject is being described as in the location, and not necessarily the action. Compare : (41) njirani-djangka-lu bina gari-nju-ya behind-LOC-ERG hear(AUX-CONT)-PAST-3 Pl S while being behind they heard a sound (42) yangka-rnu-bula njangamarda-ngka ngura hunt-PAST-3 Du S Njangamarda-LOC land they hunted in the Njangamarda country In (42) the action, the hunting, occurs in the Njangamarda land and the fact that the actors themselves are also in the land is purely related to where the action takes place. In (41) however, the noise they heard occurs in a different spot to where they, the subject, are located and it is where they, the subject, are that is being described and this takes actor agreement. Thus, we have established that for at least some examples in the data of the four cases, ablative, allative, perlative and locative we have a referential function of case with the subordinate, or second, noun phrase marked in agreement with the actor. We have not, however,excluded the possibility that we also have an adnominal function of case. If we look to Nyangumarta [ Geytenbeek, 1988 ] , Garadjari's closest relation, we see that case has an adnominal function, eg. (43) ngani ji-ninyi-0 pala-nga-lu what VB-PRES-ERG that-LOC-ERG what is that person over there doing ? as well as a referential function. eg. (44) wararr-ju 7 nga-rna-yirni-0 mayi-0 standing-ERG eat-NF-1pl.exclERG-3sgACC we ate food standing up Therefore, it would not be surprising to find that in Garadjari, too, we had an adnominal function of case and there are examples in the data that fit in with this quite nicely. eg. (45) gulja-ngu-lu yandja-ngu-lu gagara-ngu-lu nja-ngu-ya-ra south-ABL-ERG north-ABL-ERG east-ABL-ERG see-PAST-3 Pl S-COMP they can see ( us ) from the north, south and east. ie. they who are in the north, south and east can see us We are still left with the problem that we only get the relative clause with a transitive subject. However, if we look at section 5.3.5 we see that the only other two examples of possible relative clasues ( not of the 'ergative' type ) are an object and indirect object of the sentence. Thus, we do have non-subject relative clauses and these are of a different construction. Perhaps the lack of non-subject relative clauses using case marking in the data is co-incidental, but perhaps it indicates that this relative clause type is in fact used only with transitive subjects and we get a different relative clause type in other slots of the sentence. Perhaps the original situation was that we had a referential function of case only in the language, and thus got the ergative in conjunction with other cases, and then, due to the similarities between the adnominal and referential functions of case ( as stated in 6.2.3 ), the referential use of case before the ergative extended to the adnominal function. Of course this is only postulation to explain the facts, but it does seem that we have both an adnominal and a referential function of case in Garadjari which occurs, if not exclusively, at least most commonly, with the ergative case relation. 1 Note that we similarly get very little help from Laves in his translations as to what the completive suffix -ra is. He does, here, however give us one clue with the verb ya- 'to go', ie. he translates ya-rnu as 'he went' whereas ya-rnu-ra as 'he went/came up' 2 I have, unfortunately, in the data no examples of the verb bu- wtih an overt noun phrase subject, so I cannot tell whether this would be in the ergative case ( ie. a transitive subject ) or in the absolutive ( ie. an intransitive subject ). I do, however, have two examples of a compound verb gadi bu- with an overt subject. This verb is related to the cutting or chiselling of sacred objects and is formed with bu- as an auxiliary. This compound verb does not take the suffix -la . The overt subject of this verb is in the absolutive in one example and in the ergative in the other. ie. mirin gadi bu-ngu garliguru Mirin(ABS) sacred cut-PAST bullroarer Mirin chiselled the bullroarer gargidja-lu gadi bu-ngu birimala Gargidja-ERG sacred cut-PAST birimala Gargidja chiselled the birimala object Now the first example here is the only example in the texts of a verb where we would expect the ergative suffix and we do not get it, ie. the informant(s) clearly knew how to use the ergative case and did not easily confuse it ( except in the pronouns, but see section 3.10.1 about this ). Thus, why would we get a sentence such as the one above without the ergative suffix? If, as I say, bu-, as a monosyllabic verb, took an absolutive subject ( ie. it were not a transitive verb ) then speakers might be excused for getting slightly confused when using a verb with bu- as an auxiliary and, even though this compound verb were transitive, treating it as an intransitive or semi-transitive and giving it an absolutive subject. 3 Note, however, that in (16) the subject does not appear as a noun phrase and thus we cannot tell the case that it appears in. It is quite possible that a subject would appear in the absolutive case and not the ergative as expected. However, with no evidence that this is so, I will assume that the normal situation would hold true and that sentence (16) has an ergative subject. 4 It is interesting to note tha tin Nyangumata [ Geytenbeek, 1988 ] we have posibbly a similar situation. Geytenbeek admits that there are certain examples of the third person singular dative / locative bound pronoun ( ie. equivalent to -la ) for which she is completely at a loss to account fro why it is used and so suspends her judgement. She gives the following example. partanykarrangu-lu malya-rni-yi ngalayika-lu mungka-0 kaju-jartiny-ju children-ERG chop-NF-3plERG-1du.exclDAT-3sgDAT tree-ACC axe-COM-INST the children chopped wood for us with an axe She examines the possibility that here the third person singular dative may cross-reference the instrumental but rejects the idea and leaves the question open. Might not the third person singular dative be cross-referencing some reason that the children are chopping wood, ie. maybe somebody told them to chop it ? 5 Note that as we never get the suffix -la in combination with an overt noun which we could say is its benefactive or causal referent, we might prefer to call -la an indicator of a cause etc. rather than a cross-reference marker. 6 An alternative to this assumption would be to say that -lu is coming to be extended from simply a transitive same participant marker to cover transitive and intransitive subject. However, given that change is occuring in the pronominal system away from this nominative / accusative system we would not expect simultaneous change in the language towards a nominative / accusative system. 7 Note that Geytenbeek describes this as the intrumental suffix, hwoever, with the knowledge form Garadjari and other languages it seems likely that it is actually an ergative marker in agreeement with the subject. -------------------------------------------------------------------------- end of file