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

5.2 Origins of the AN construction

One of the striking things about AN is that it is found in Pama-Nyungan languages across the VRD and the Barkly Tablelands and in the Yolngu group far to the north, with as far as we know, no occurrence in between. This could suggest inheritance of this pattern from fairly deep within the Pama-Nyungan family. However the feature is also absent in the western Ngumpin-Yapa languages and is not present in any Non-Pama-Nyungan languages. This suggests that this is perhaps an areal feature, whose origin may be in Tablelands-Gulf Pama-Nyungan, perhaps Warluwarric.¹⁹ The areal spread of the feature must have developed at a time when Yolngu was still in touch with this area: this scenario avoids the unlikely hypothesis of independent invention in two areas.

Apart from its distribution we need to look at how it may have evolved. As noted, this is a rare type of grammaticalisation of a semantic category meaning ‘motion towards’. One conjecture could be that initially this began in contexts which combined motion and transitive objecthood.
Perhaps the bridging context is verbs of putting. Many of the languages use an ALL case-marker on the locative goal of verbs of 'putting'. For Alyawarre, Yallop (1981:80) has some interesting examples where he interpolated a "and put "
"I got it (and put it) into a pannikin"

"we dug yams (and put them) into a piece of bark"

"We were cutting bush bananas (and putting them) into a baby carrier”
The situation in Jaru where Tsunoda reports the parallel subordinative use of ALL only in contexts of motion is also suggestive, but the evidence is rather sparse.
Another feature which is found in a number of languages with AN, either categorically or as a tendency, is the requirement that the subject and object be in a different location to trigger a change to AN marking. Alternatively LOC can be seen as marking scope of location as being over the whole clause and ALL functioning to mark restricted scope of the location to the object NP. This may well have been important in the semantic matrix leading to the grammaticalisation, even though it is less clear-cut or even absent in some contemporary languages. The action by the agent on the patient may be conceptualised as motion-like since they are in two locations. This seems to relate typologically to the ‘directional location’/’fictive motion’ phenomenon in Finnic languages, and this is further commented on in the Conclusions below.
This construction is absent in other western and central Pama-Nyungan languages but there are other constructions which may have played a role in its development, such as the use of Dative in Alyawarre described in 4.6.

5.3 Origins of LS

LOC case marking is used not only for spatial location but also for temporal location. It is thus not surprising that a LOC case can attach to a non-finite verb to produce a subordinate clause with a temporal relation to the main clause. Exactly what the temporal relation is (simultaneous or preceding) varies from language to language, probably as a result of pressure from other affixes. The control of the subject of the locative clause is either specifically the subject of the main clause, or can be more general, as in Gurindji. Forms such as the Warlpiri -rlarni which is built on the LOC case form may reflect earlier stages with other types of marking specifying controllers.

LS may be more widespread than it at first appears. Forms which play a role like LS elements but are not LOC in present-day languages like –kil in Warumungu may be related to more common Pama-Nyungan locatives and specifically forms which have a kV or ngV ligative preceding the old LOC suffix. In Pama-Nyungan more generally the common subordinating suffix –inyja may be composed of a verbal suffix with the LOC allomorph –ja, historically.
Dench & Evans (1988:30) observe that in their survey the LOC suffix used to mark same subject only occurs in languages in which the LOC suffix also has denotes or has had denoted an ergative functionmeaning. Nordlinger (1998:213) points out that this is not borne out in all languages with LOC subordination, like Wambaya. Gurindji and Jaru discussed here, and other Ngumpin-Yapa languages are also counterexamples, although LOC is not strictly a ‘same subject’ marker in these languages. Pensalfini (2003:128) however follows the general line of argument of Dench & Evans.
Wambaya has a split pattern of marking of subordinate non-finite clauses, but the object control suffix -barda~-warda is not the ALL form in the current language. One possibility though is that this is related to the Wakaya ALL form –ert .

5.4 Origins of AS and its relationship to AN

The use of the ALL as marker of the location of the object is almost certainly linked to the use of the ALL as a marker of the control of a non-finite clause by the object of the clause. One possibility for the bridging semantics is perhaps the extension of locational meaning of the Allative "OBJECT is at the fire" to a more general associative meaning "OBJECT is involved with the fire", and from then on to an explicit verbal meaning "OBJECT is doing something with the fire", which can be represented explicitly by a non-finite verb. Some evidence for this comes from the fact that in Gurindji, Warlpiri, Warumungu and Wakaya at least the object of the non-finite verb can, or must, take the same ALL case-marking.

It is noteworthy that all of these languages also have the AN construction, so transfer of object ALL marking to the verb when it has a nominal non-finite form is not a great leap. If both LS and AN exist in a language it might be simply a regular natural process for the ALL marking on nominals to be extended to nominalised non-finite forms of verbs as well as their objects, under object control.
Yolngu stands out as the language with AN but no local marking of non-finite subordination (LS or AS). This indicates that AN was a separate matter from subordination at some early stage across a putative linguistic area including Warluwarric, Ngumpin-Yapa, pre-Warumungu and Yolngu. It is worthy of note that most of these languages also show the semantic split whereby LOC signals co-location of subject and object and ALL spatial separation of them, fully or to some degree. It seems likely that this was an old feature of AN, declining in the non-core area – Ngumpin-Yapa and Warumungu. In Warumungu this feature is entirely absent.
The distribution of the AS pattern is generally within the LS area except in one case, Warumungu (and even in this case LS may have existed in the past, as we have noted). The AN construction area intersects with LS in Warluwarric and eastern Ngumpin-Yapa so it looks likely that the object-control feature inherent in the alternation to ALL in secondary predications on object NPs became transferred to the subordinate non-finite clauses. This yielded the hybrid AS pattern, which is found across the area of intersection of AN and LS.