28 March 2009

Causative Makeovers in New Guinea Oceanic Languages, 3

In contrast to Austronesian languages almost everywhere else, the Oceanic languages on the north coast of the Papua New Guinea mainland show an unusual disinclination to make use of the morphological causative inherited from Proto-Oceanic and Proto-Austronesian. Innovative causatives derived from causative serial constructions appear to have supplanted to varying degrees the inherited prefix *pa(ka)-. Part 1 summarizes the dethroning of the inherited prefix. Part 2 outlines the replacement pattern of serial causatives. Part 3 (here) suggests reasons for preferring the serial causatives.

Ambiguous syntax preferred

What accounts for the decline of the inherited morphological causative and the rise of the serial causatives in the Oceanic languages along the north coast New Guinea? From a strictly structural point of view, the path of causative serialization seems an unnecessarily complicated way to move from SVO to SOV. The most grammatically economical way would be to add one rule inverting every VO to OV. But what is grammatically economical in this case seems sociolinguistically implausible.

I assume that speakers of neighboring Austronesian and Papuan (non-Austronesian) languages were often familiar with each other’s languages, at least to the extent that each could understand when spoken to in the other language(s) and could be assured of being understood when using their own language. Even when fluent in the other language(s), speakers may have wished to speak their own language for emblematic reasons, to signal their identity as members of a particular linguistic group. Or they may have had to speak to monolinguals some of the time. In such circumstances, speakers may have tended to mitigate the differences between the various languages they used. They may increasingly have favored structures that were originally highly marked or syntactically ambiguous in their own language(s) precisely because such structures resembled the patterns of their interlocutors’ language(s). Some of these structures could have been analyzed according to the patterns of either language type.

Causative serial constructions permit just this sort of multiple analysis. They contain no clause boundary markers and thus permit speakers and hearers to parse the constructions to suit their own preconceptions about clause structure and word order. At the same time, serial causatives run little risk of being misunderstood. They describe cause-and-result events in an order matching the unfolding of those events in the real world. The first verb denotes the manner in which the Agent behaved; the second describes the effect of that behavior on the Patient.

(NOTE: Slobin [1982] presents evidence that children have a harder time learning periphrastic causatives than they do learning morphological causatives. I claim only that serial causatives, not periphrastic causatives, are semantically transparent. The periphrastic causative verb [‘do’, ‘make’, ‘cause’, etc.] carries more grammatical information than it does information about the extralinguistic world. The serial causative verb [‘chop’, ‘hit’, ‘grab’, ‘twist’, etc.], on the other hand, carries more real-world than grammatical information. The periphrastic causative indicates a causative relationship between one event or entity and another event. The entity responsible for the causing event may be described but the specific nature of the causing event must be inferred from context. The serial causative describes both causing event and resulting event. It leaves the causative relationship to be inferred from the construction—from the order in which the two events are described and from the semantic and grammatical bonds between the two descriptions. The periphrastic causative involves embedding. The result verb is grammatically subordinate to the causative verb. The serial causative involves no embedding. The cause and result verbs share one or more arguments but neither is subordinate to the other.)

Serial causatives can be parsed in various ways depending upon where the parser decides to insert a clause boundary and whether the construction involves a same subject or switch subject relationship between the verbs. A further option is not to insist on parsing the structure into two separate clauses.

If one decides the construction contains two clauses, one can insist on perceiving a clause boundary in either of two likely locations: either before or after the Patient NP. A clause boundary after the Patient NP will produce an initial SVO clause and thus make the structure compatible with SVO word order. In the switch-subject type, the second clause will be intransitive, with no overt subject NP; while in the same-subject type, the second clause will be transitive but contain neither a subject nor an object NP. As an illustration, take the serial construction, ‘I chopped the tree toppled’. If one inserts a clause boundary between ‘the tree’ and ‘toppled’, then one has an SVO clause ‘I chopped the tree’, and a second clause with just the verb ‘toppled’. A switch-subject reading would produce ‘I chopped the tree and it toppled’ while a same-subject reading would yield ‘I chopped the tree and toppled it’.

But suppose one felt that a clause boundary before the Patient was more in line with one’s preconceptions. In this case, the first clause would consist of a subject and a verb with no object NP, while the second clause would contain either a Patient NP acting as subject of an intransitive verb, or a Patient NP acting as object of a transitive verb. Again take the example ‘I chopped the tree toppled’. A switch subject version could be parsed into ‘I chopped and the tree toppled’, while a same subject version could be parsed to read ‘I chopped and toppled the tree’. Both versions with the clause boundary before the Patient are thus compatible with SOV word order. The switch subject version contains no overt object NP and the same subject version contains an object NP in front of the final verb.

If one exercises the option of not inserting any clause boundary in a serial causative, then one has a clause that is both verb-medial and verb-final. An interpretation under which the manner verb is the main verb will be compatible with SVO prejudices, while one in which the result verb is the main verb will conform to SOV prejudices. Semantics will not easily decide. There seems no more reason to suppose that the manner-of-action is ancillary to the result-of-action than there is to presume that the result is subordinate to the manner. The manner verb describes the role of the Agent while the result verb describes the effect on the Patient. Neither the Agent nor Patient is a dispensable member of the Agent-Patient transitive clause. These parsing options are summarized in Table 1.

Table 1: Parsing Options for SVOV Serial Causatives

One clause
Parsed as SVO: S Vmain O Vsub
Parsed as SOV: S Vsub O Vmain
Two clauses
with switched subject
Parsed as SVO: S V O ## Vi
Parsed as SV: S V ## S Vi
with same subject
Parsed as SVO: S V O ## Vt
Parsed as SOV: S V ## O Vt

Reflexes of both switch-subject and same-subject serial causatives are found throughout NGO languages. Very often, reflexes of both types are found in the same languages, as in Manam. However, it appears that same-subject serial causatives are more common. In Manam, for instance, the only intransitive result verb that appears in homologs (inherited similarities) of the serial causative is -mate ‘to die, be dead’ (Lichtenberk 1983).

Table 2: Reflexes of Two Types of Serial Causatives in Manam

Switch subject: *S s-V-o O=S s-V ==> S O s-V-V-o
boro i-mate ‘the pig died/is dead’ (Vi)
pig 3s-die
(di) boro di-rau-mate-i ‘they killed the pig’ (Vt + Vi)
(3p) pig 3p-hit-die-3s
Same subject: *S s-V-o O s-V-o ==> S O s-V-V-o
(ngai) ‘ai i-sere‘-i ‘he split the wood’ (Vt)
(3s) wood 3s-split-3s
(ngai) ‘ai i-zan-sere‘-i ‘he split the stick lengthwise’ (Vt + Vt)
(3s) wood 3s-punch-split-3s

There is also evidence that speakers of Oceanic languages in New Guinea adopted the one-clause analysis of causative serial constructions. In almost every language, either the manner or the result verb has lost its verbal status. Prevailing word-order patterns seem to have been the sole determinant of which of the two verbs became grammaticalized (reduced to a smaller, more restricted grammatical class). Many of the languages that have not made the full shift to SOV (those in Morobe Province) have grammaticalized the clause-final result verbs. The reflexes of those verbs now form a class of resultative particles. In contrast, the nonfinal manner verbs have grammaticalized in the fully SOV languages. They have yielded a set of prefixes classifying the manner of action by which various results are achieved. Some of these prefixes have degenerated to the point where their meanings are indeterminable on solely language-internal evidence. These contrasting patterns of grammaticalization are summarized in Table 3.

Table 3: Different Resolutions for Two Verbs in One Clause

The VO Solution, adopted by Numbami and other VO languages:
S Vmanner O Vresult ==> S Vmain O Result
The OV Solution, adopted by Manam and other OV languages:
S Vmanner O Vresult ==> S O Manner-Vmain

One way to get the verb from one position to another within a clause, then, is to render the information of that clause in such a way that verbs fall in both positions in a construction that contains no boundary markers and thus permits multiple analyses. The evidence suggests that Oceanic languages on the north coast of New Guinea adopted this strategy in changing from SVO to SOV. The availability of this strategy suggests that serialization need not arise from either coordinate or subordinate relationships between two separate clauses. In fact, there is no evidence that I am aware of—at least in the Oceanic languages on the New Guinea mainland—that causative serial constructions were ever two fully separate clauses. The languages are adequately supplied with conjunctions, conjunctions that are often ubiquitous in other constructions. But no conjunctions show up between the constituents of serial causatives, not even as morphological remnants. Serial causatives in these languages were apparently from their very inception structures containing one clause worth of event-construction semantics and two or more clauses worth of verbs, without any intonational or morphological indication of a clause boundary.


Lichtenberk, Frantisek. 1983. A grammar of Manam. Oceanic Linguistics Special Publication No. 18. Honolulu, University of Hawaii Press.
Slobin, Dan I. 1982. Universal and particular in the acquisition of language. In: Eric Wanner and Lila R. Gleitman, eds., Language acquisition: The state of the art. Cambridge, Cambridge University Press.

1 comment:

John Cowan said...

It's interesting that Cham, Bengali, and Mandarin (a) use serial verb constructions and (b) have a long history of being adopted by populations that formerly spoke another language. In Cham in particular, although the morphological elements tend to be old — even PAA — in age, they are applied in a rigorously transparent way that makes the language systematic and easy to learn. See Graham Thurgood's excellent paper "Learnability and direction of convergence in Cham: the effects of long-term contact and linguistic structures".