Language Typology

Possible generalization of Comrie's hypothesis about the syntactic position of the causee

Valery Solovyev

Kazan State University,

causee in causative constructions, focusing on transitive constructions in nominative-accusative languages, with some comments on ergative languages. The generalization of Comrie's hypothesis about the causee's position is also given in terms of this metalanguage. Several statements pretending to be called linguistic universals are formulated.

1. In analyzing causative constructions we are concerned with the problem of describing the causee's position. Though Comrie's hypothesis that the causee takes the first available place in the accessibility hierarchy SUB>DO>IO>OBL is well known, some examples, noted already by Comrie [1], contradict it. Taking this fact into account, Comrie's hypothesis would rather be characterized as a tendency than an absolute universal. The first group of instances concerns omitting some positions. The following pair of sentences in the Kannada language is discussed in [1].

(1) Avanu nanage bisketannu tinnisidanu
     he(nom) I(dat) biscuit eat(causat)
     'He fed me a biscuit'
(2) Avanu nanninda bisketannu tinnisidanu
     he(nom) I(instr) biscuit eat(causat)
     'He got me to eat the biscuit'

The permissibility of sentence (2) contradicts Comrie's hypothesis from the viewpoint stated above (to be an absolute universal). To explain such a situation it is proposed in [1] to take into account some additional factors of a semantic nature. To be more exact, sentences (1) and (2) differ in the degree of event control exercised by the causee. In (2) this degree is higher. In the Tatar language the equivalent of example (2) is shown in example (3):

(3) ul minnen biskvitny ashattyrdy
     he I(abl) biscuit(acc) eat(causat,past)

The causee is marked by the ablative affix, which is used for oblique objects. The verb ashattyrdy also permits an indirect object. In Tatar the indirect object position is marked by the dative.

(4) ul minnen ulyma biskvitny ashattyrdy
     he I(abl) son(1-poss,dat) biscuit(acc) eat(causat,past)
     'He got me to feed my son a biscuit'.
However, even if this position is free the causee cannot occupy it.
(5) * ul mina biskvitny ashattyrdy
     he I(dat) biscuit(acc) eat(causat,past)

Causee marking by means of the ablative does not depend on whether that case has beenencountered in the clause or not yet.

(6) ul minnen oemnan biskvitny kitertterde
     he I(abl) house(abl) biscuit(acc) bring (causat,past)
     'He got me to bring a biscuit from the house'

None of the semantic factors defines the causee marking in Tatar and none of these factors can help us explain such a violation of Comrie's hypothesis.

2. A new explanation of choosing the causee's position.
In the most typical - active - clause the subject plays the role of agent. The semantics of the agent has two main components: the initiator and the executor of the action. For usual causative constructions it is right to think that the semantics of the agent splits: the causer keeps the semantics of initiator while the causee takes the semantics of executor. The following question seems pertinent: how in general can the executor be encoded in this language? The executor of an action (not the agent) can exist also in noncausative constructions.

For example, in Lak [2] ergative markers are replaced by oblique markers when the property "to be the initiator" is lost.

(7) Gwana-l cwurku iik'undi
     he(erg) thief(nom) kill(past)
     'he killed the thief'

(8) Gwana-sha cwurku iik'undi
     he(abl) thief(nom) kill(past)
     'he killed accidentally the thief'.

In Tatar causative constructions the executor acts in the following way: it occupies the object position if no object is present, otherwise it takes puts on special markers (different from the agent's markers in passive constructions).

Let us write a sequence of positions that the executor can occupy (starting with the subject position): SUB-DO-Spc, where Spc means special mark of the executor. We shall call such sequences "chains". We note that all members of the chains, except Spc, are syntactic positions in the ordinary sense. Spc is not a syntactic position, but only a marker of NP. It is not enough data to distinguish Spc as a special syntactic position.

3. General hypothesis about marking the executor.
For each type of regular transformation (causative, etc.) the language has a chain of positions (markers), and the executor occupies the first available slot in this chain. The last slot of the chain is available for the executor even if it is already filled by a noun phrase.

Comrie's hypothesis appears to be a special case with the chain completely coinciding with the accessibility hierarchy. Though chains are coordinated with the hierarchy, some of their positions (belonging to the hierarchy) may be omitted.

To say some positions are omittable is not enough to describe all cases of causee marking.

Sentences (1) and (2) show that the causee's position can be chosen in different ways, i.e. one language may have several chains regulating the executor's disposition in causative constructions. In Kannada chains look like this: SUB-DO-IO-Spc and SUB-DO-Spc. Sentence (1) is constructed accordingly to the former strategy while sentence (2) is constructed according to the latter strategy.

To return to our theory, if causative constructions are managed by two (or more) chains, they are competitors from the viewpoint that in constructing a clause only one chain is needed. In Kannada the motivation for this choice lies in the semantics. And such motivation agrees well agreed with the explanation proposed in this paper. Sentence (2) leaves greater action control in the hands of the causee (meaning it keeps the executor's properties). So, the former agent (executor) has a special marker. Usage of sentence (1) depends on low retention of control. In this situation the causee must be interpreted rather like a passive participant of the action than the executor.

The next interpretation (of course, further investigation is necessary) can be proposed. In sentence (1) avanu 'he' is an initiator and an executor simultaneously (a full agent); accordingly nanage 'I' takes a position, not included in the chains for executors. The fact, that the object of "feed" can not be an executor is confirmed by the example (4). In accordance with this interpretation, only one chain - SUB-DO-Spc - must be postulated in Kannada.

In Turkish we observe the same chains, but no semantic motivation for choosing between them [1]. One may say that there is an indeterminate choice of the chain.

The stated approach offers clear demarcation between syntactic and semantic aspects of the causative constructions. The set of language chains is the syntactic object, but the semantics defines the method of choosing one of the competing chains.

4. The second class of instances contradicting Comrie's hypothesis consists of cases where the causee is about to occupy a position that is already filled. This possibility can be simply expressed in terms of the chains. Up to this moment all given chain instances have ended with the special mark Spc that may be doubled if needed (i.e. it may encode two noun groups in a clause). Doubling at any other position means that chains may end with that position. According to [1], in Sanskrit the causee may be marked as the object even if this position is already filled. It means that one of the chains in Sanskrit looks like SUB-DO (there is also another possibility: SUB-DO-Spc).

We formulate several properties characterizing the structure of a chain (for nominative-accusative languages).

Property 1. Every succeeding element of a chain is situated in the accessibility hierarchy to the right side of the previous element.

Property 2. Spc takes the oblique case marking.

As a consequence of properties 1 and 2, Spc can fill only the last slot in the chain.

Property 3. Chains belonging to the same class of construction (e.g. causative) may have only one Spc marker. Spc may vary from one class to another.

Property 4. Only the last slot of the chain may be doubled.

We note that a language can permit no chains. In Songhai [1] the causee is simply omitted because it cannot occupy any position at all.

Are these properties universal? The present writer does not know of any violation of these properties of causative constructions.

The concept of chains can be applied not only to causative constructions. Let's consider the so-called "mutually-joint" voice (abbr. "mj") in Tatar. It is marked by the affix -sh. This voice has two meanings: 1) "to help the executor do something"; 2) "to do something together". Let us consider executor marking in the mutually-joint constructions.

(9) ul mina biskvitny ashashty
     he I(dat) biscuit(acc) eat(mj,past)
     'He helped me to eat the biscuit'

(10) ul mina biskvitny iptesheme zhibereshte
     he I(dat) biscuit(acc) friend(1-poss,dat) send(mj,past)
     'He helped me to send the biscuit to my friend'.

On the basis of these instances it is easy to hypothesize that for this class of constructions the executor moves through the chain SUB-DO-IO. Surprisingly, from

(11) min baram
     I go
    we cannot get

(12) * ul mine baryshty
     he I(acc) go(mj,past)
     'He helped me to go'.

Only the construction with an indirect object is possible:

(13) ul mina (dat) baryshty.

So, marking the executor reduces the chain to SUB-IO, with omitted DO. The mutually-joint voice in Tatar has still another interesting property. As mentioned above, two values of the voice are present, so it makes sense to study the behavior of the executor in a clause, separately for each value. Here an interesting effect is revealed. In examples (10) and (11) the verb retains the first meaning - "to help". On the other hand, (13) turns the verb to the second value only: "he go with me"; the first meaning "to help" must be arrived at by other means: the verb baryshty cannot be used here.

Concerning mutually-joint constructions with the "help" meaning, to describe the executor's displacement by means of the chain SUB-IO seems to be impossible; for though the executor can occupy the position IO (instances (9) and (10)), it can only do so when the position DO is filled.

In order to account for such a situation, let us introduce the term "forbidden position".

Definition. Syntactic position in the chain is called "forbidden" if the executor cannot occupy it even if the position is not filled. In that case the executor does not move to the next available slot and the phrase cannot be correctly constructed in any way at all.

Forbidden positions will be marked by the sign "#". Thus, in Tatar mutually-joint constructions with the "help" meaning describe the executor's displacement as: SUB>DO#>IO.

The next property is expected:

Property 5. The first and the last positions in a chain cannot be "forbidden".

For the set of properties determining some classes of constructions, the problem of universality ought to be supplemented by the problem of completeness. In the current case completeness implies that any set of chains agreeing with properties 1-5 has a realization in some language. For example, the set {SUB>DO, SUB>DO>IO, SUB>DO>IO>Spc} satisfies the mentioned properties. Is there a language in which the choice of the causee's position is determined by this set of chains? The existence of such a language would confirm the possibility of doubling for both direct and indirect objects, and also the existence of a special marker for the causee. If no appropriate language exists, the proposed set of properties is incomplete and must be expanded.

5. Now let us outline, briefly, the application of the proposed approach to ergative languages. Of course, it demands much more detailed, special investigation

Instances from Chamalin (Daghestan language family) [2] are:

(14) Oss-i vog'a un
     he(erg) tree(nom) push(past)
     'He pushed the tree'

(15) De oss-uchI vog'a un-al
     I(erg) he(loc) tree(nom) push(caus,past)
     'I got him to push the tree'.

In this example, similarly to (2), the causee is marked by the locative (oblique marker) even though the slot of the direct object is free.

The chain for Chamalin may be represented as: ERG-Spc. To compare chains in ergative languages with the accessibility hierarchy, it is possible to interpret the hierarchy using the following scheme: NOM-ERG-IO-OBL. Such an interpretation is offered in [3].

One curious aspect of ergative chain structure is the presence of chains with a single element - Nom. Such chains appear when the initial clause has no object and, thus, the agent is marked with the nominative. In the causative clause the causer appears in the position of the ergative while the causee remains in the nominative (i.e. the executor's marking does not change). For ergative languages properties 1-5 must be reformulated appropriately.

We have introduced the term chain of syntactic positions to indicate the sequence of positions that the executor (agent) can occupy in causative constructions and constructions standing near them. The hypothesis about choosing the syntactic position of the causee is Comrie's hypothetical generalization given in terms of chains. Its status as a universal depends on the universality of the basic properties 1-5. The problem of completeness of the proposed description is addressed in section 5 above.

In the investigation of Tatar mutually-joint constructions an interesting effect is observed: the presence of available syntactic positions which nonetheless cannot be occupied by the causee.


1. Comrie B. Language Universals and Linguistic Typology. Chicago: The University of Chicago Press, 1981.
2. Kibrik A.E. Ocherki po obtshim i prikladnim voprosam iazikoznanyia. Moscow: Moscow State University, 1992.
3. Johnson D. On the role of grammatical relations in linguistic theory. In: La Galy, 1974, p. 269-283.

Back to FCCL HomePage
Back to FCCL Information Bulletin