Moving along the grammaticalisation path: locative and allative marking of non-finite clauses and secondary predications in australian languages

3.2 Double case–marking in Australia

Double case marking or Suffixaufnahme (Planck ed. 1995) refers mainly to addition of a genitive marker following another case marker, but it is extended to include other types of addition of the case of a controlling NP to other items such as NPs with local case marking. In some Australian languages, this kind of double case-marking (Dench and Evans 1988) is used to specify more precisely the referent. For example, if an Object has Dative or Accusative case, adding Dative or Accusative case to a local-case-marked phrase indicates that the phrase expresses the spatial position of the object rather than the event. An example follows from Ngarluma, a Pama-Nyungan language spoken on the coast of Western Australia.⁷

Ngarluma (Kenneth Hale's fieldnotes (Hale 1960a, Simpson 1981)
ACCUSATIVE on locational modifier of ACCUSATIVE-marked OBJECT
(10) Wanji-la-ku nyinta marrparnta-nha.

where-LOC-ACC you.NOM find-PAST

‘Where did you find it?’
(11) Ngayi marrparnta-nha yanthan-ta-ku thurnu-ngka-ku.

I.NOM find-PAST swag-LOC-ACC under-LOC-ACC

‘I found it under the swag’
The double-case-marked phrase here is referred to as a secondary predication with the understood subject of the secondary predication being ‘it’.
Double case marking is found in some of the languages below such as Warlpiri. Since double case has parallel functions in marking control to AN/LS/AS systems one might attempt to trace a historical connection between them but this line of enquiry has not been productive.⁸

3.3 Switch reference in Australia

Some of the phenomena in Central Australia have been noted before by Austin (1981 cf. Blake 1993) and Dixon (2002:239) and subsumed under the heading of ‘switch reference’. In a northern area which corresponds approximately to the area examined in this paper, Austin claims that the LOC signals ‘same subject’, i.e. the subject of the subordinate clause is the same as the subject of the main clause. Over most of this northern area the LOC is complemented by the ALL which is used to signal different subject. Austin and Blake cite the languages Wanyi, Garrwa and Wakaya as exhibiting the common pattern which is for a marker originally denoting a locative meaning of origin to indicate same subject (SS) while a marker of originally denoting an allative meaning origin indicates different subject (DS).

The use of the ALL on the object of the ALL-marked subordinate clause follows a common areal property whereby the object of a non-finite verb bears the same case as the verb. Relevant to the AN use is the fact that some of these languages allow case-marked nominals to appear in clauses without a corresponding nominalised verb, but with an understood action 'involved with, doing something with' e.g. 'I've been watching the white fellow involved with a bullock.'
Switch reference marking typically signals the identity or nonidentity of the referent of the subject of one clause, with the referent of the subject of another clause. The definition may be extended to include prominent argument. It is often seen as an information structure property - signalling a change of topic. But the ALL construction AS in these Central and Northern Australian languages appears to be doing something slightly different: marking the identity of the referent of the object of a main clause with the referent of the subject of a subordinate clause, (or, in the case of Gurindji, the identity of the referent of a non-subject argument of a main clause with the subject of a subordinate clause). In Gurindji too, LOC is used to mark clauses where there is no identity of NPs between the clauses, so ‘same subject’ is not a correct characterization of the function of LOC in this case (see 4.1 below).
Austin (1981: 325) makes clear that the Warlpiri and Warlmanpa system of cross-clausal control marking is not of the classic ‘switch reference’ type. The Warlpiri system does not involve only a simple binary contrast (SS vs.DS); it is based upon the binary contrast of whether the subordinate subject is coreferential or non-coreferential with a grammatical argument of the main clause verb, with additional subdivisions in the former category.
Austin (1981: 330) canvasses the idea that switch-reference in Australia is a case of indirect (structural) diffusion, and this idea is taken up and extended by Dixon (2002).

But the nature of the Ngumpin-Yapa and probably other systems in this area casts into doubt the claim that the areal pattern of what we will be discussing as ‘object control’ marked by ALL (AS) is caused by diffusion of a ‘switch reference’ pattern, in an abstract sense. Rather in many cases the relevant object control systems are mainly inherited, and merged together when they co-occur. Particular uses of the distinction between LOC, and in contrast ALL for object control, arising from such a merger seem to have diffused to some other neighbouring languages.

4. Related systems in Northern and Central Australia

In this section we review the control constructions which use local case-marking for several languages, in terms of the typology of AN, LS and AS introduced above. There are additional features which occasionally have a bearing on whether LOC or ALL case is used. For the subordinate clauses, relative time of the main and subordinate clause is relevant, at least in Warlpiri. Further, the distinction between whether the subject is in the same location as the object or not can have a bearing on the choice of ALL or LOC in AN constructions, with ALL used in some languages when subject is not in the same location as object or where the emphasis in secondary predication is not on the whole event but just the object (cf. Austin 1981: 332). Although it may be thought that the semantics of the verb (e.g. whether motion is involved) could affect choices in the AN construction, there is no solid evidence for this in the languages being considered.

We have not found the AN and AS patterns outside the language area described here although LS is found in some other Pama-Nyungan languages. In the Non-Pama-Nyungan languages of the Top End of the NT and Arnhem Land, e.g. Ngalakgan and Mangarrayi, there is a neutralisation of the distinction between Locative and Allative in their basic semantic function anyway, so no grammaticalisation of the distinction would be expected.
We consider here twelve languages, in four groups:

1. 
   Languages of the Ngumpin-Yapa sub group of Pama-Nyungan in the Tanami Desert and southern Victoria River District (VRD) of the Northern Territory, and the southern Kimberley region of Western Australia. These include Gurindji (4.1); a closely related Ngumpin language to its west, Jaru (4.2); and Warlpiri to the south (4.3)
   
2. 
   Languages of several other Pama-Nyungan sub-groups to the east and south of Ngumpin-Yapa: Warumungu (4.4); Wakaya, a Warluwarric language of the Barkly Tableland (4.5); and Kaytetye, an Arandic language to the south, which is also compared to Alyawarre another more southerly Arandic language (4.6). These languages, in contrast with Ngumpin-Yapa, share a property of not not distinguishing between LOC and ERG (i.e. the case-suffixes denoting ergative and locative meanings) the LOC and case suffixes. The reasons for this and the implications for the history of the constructions in focus here are discussed below in this section.
   
3. 
   The Yolngu subgroup of languages in North-East Arnhem Land, which are an outlier of Pama-Nyungan separated from the rest of the family by some hundreds of kilometres (4.7).
   
4. 
   Several Non-Pama-Nyungan languages neighbouring the Pama-Nyungan languages in A & B sets, which also have aspects of the AN, LS and/or AS constructions: Garrwa-Waanyi (4.8) two closely related languages around the Northern Territory- Queensland border north-east of Wakaya; to the west, Wambaya (4.9) and Jingulu (4.10) two languages of the Mirndi family in the Barkly Tablelands (Harvey 2008); Wardaman, north of the eastern Ngumpin languages (4.11); and Gooniyandi in the Kimberley of Western Australia (4.12).
   

One of the allomorphs of proto-Pama-Nyungan LOC was *-la, and in Ngumpin-Yapa this (or the form with the retroflex lateral –rla), remains the form for vowel-final nominals of more than two morae. The languages of set B above appear to have retained this LOC suffix but it has been subject to sound changes and morphological augmentation. Another consequence of the sound changes was the loss of the formal distinction between suffixes denoting locative and ergative meanings in these languages since the proto-Pama-Nyungan form of the ERG suffix for multimoraic vowel-final nominals was *-lu, differing only in the vowel from the LOC. The relevant changes are as follows:

1. 
   the reduction of final vowels in Arandic to a central vowel /e/, neutralising the distinction between –la (LOC) and –lu (ERG) and yielding -le;
   
2. 
   the loss of final vowels in Wakaya, (and possibly Warumungu), yielding –(r)l respectively for the LOC/ERG
   
3. 
   progressive vowel assimilation in Warumungu, neutralising final vowels, yielding -ngkV, -jjV and -CV;
   
4. 
   the augmentation of the –l by a preceding ligature –kV- or –ngV- These ligatures are also present including with LOC and ERG in Ngumpin-Yapa, e.g. Gurindji -kula LOC, -kulu ERG following non-coronal consonants.