agent/patient

[Spike Gildea]

Wally's example from Caddo calls to mind some languages illustrated in
Givón's (1991, Syntax: A Functional-Typological Introduction, Vol. 2)
chapter on passives.  He shows a number of languages where a transitive
verb inflected for third person plural subject is interpreted as having an
indefinite subject; he calls such a construction a "nonpromotional
passive".  Such a construction is even attested in relatively well-known
languages, such as Colombian Spanish and Brazilian Portuguese, where third
person plural forms of the verb are replacing the analytical passive in
informal speech.  (Just because I can't resist, one example: a woman is
writing notes over her data, leaves the room for a few minutes, and a young
man enters, picks up her pen, and begins playing with it while conversing
with me.  She re-enters the room, sees that her pen is missing, spots it in
his hands, and says, in a miffed tone of voice, 'Roubarem a minha caneta!'
-- lit. 'They stole my pen!')

The interesting thing is that in Kimbundu, you can actually see the further
evolution of this construction into a more protoypical passive in that when
the patient is topicalized (etymologically left dislocated, later
re-incorporated into the intonation contour of the main clause), an oblique
agent-phrase (of any person or number) can co-occur with the erstwhile
"indefinite subject" marking on the verb.  Thus, you can see where an
obligatory morphological slot on the verb is filled by a morpheme which
etymologically coded a referential third person plural agent, but which
begins to be used in situations where the agent is not specified (even
though it might be identifiable -- i.e. a passive) and which then continues
evolving semantically until the addition of a new agent phrase confirms
that the morpheme does not (at least in this construction) refer to the
agent at all.  Apparently a similar process happened in the Caddo data
Wally refers to.

To bring this back to Suzette's question: is reference to the agent a part
of such a construction (i.e. is it a counterexample to the apparent
universal she is so uncomfortable with)?  Like many typological questions,
the universality of a generalization depends on your definitions, and how
far you are willing to extend categories.  For this case to violate the
putative universal, the key definition that requires stretching is
'reference': If by 'agent' you mean simply a particular participant which
is inherent in the schema of a particular event, and if by 'reference' you
mean simply alluding to the existence of such a participant, then perhaps
such 'indefinite' morphemes serve as place-holders to indicate that such a
participant does exist semantically (even though it may not be
pragmatically feasible or desirable to refer to it explicitly).  However, I
would hesitate to use the term 'refer' for such an indefinite morpheme,
especially since (in this unique case) the person and number of the agent
does not have to match the person and number indexed by the morpheme (i.e.,
agents might be neither third person nor plural).  By the time such a
morpheme arrives at the passive stage seen in Kimbundu and Caddo, I would
consider 'reference' only to be an etymological function of a passive
morpheme.

This example highlights one of my concerns with typologies based solely on
morphosyntax, especially when they do not explicitly attend to functional
shift and incipient grammaticalization.  Given the constant tension between
the semantic roles schematized with each verb and the pragmatic needs of
each utterance which must refer to a specific event, I would be amazed if
you could find a language in which there was no way to move the agent
"off-stage"; in this sense, I am quite comfortable with the functional
usefulness of the universal you posit.  In some languages the means of
moving the agent off-stage might utilize morphology usually restricted to
transitive sentences, even morphology that elsewhere refers to on-stage
agents of transitive events.  While this would not violate the functional
universal, it might violate the morphosyntactic universal.

Spike


>Dear Suzette,
>
>You wonder about an agent being necessary "as only some sort of
>indefinite." I believe there are a number of languages (I am most familiar
>with Caddo), where there is no passive as such, but a similar function is
>performed by means of a so-called indefinite (I've called it in Caddo a
>"defocusing") marker.  So the Caddo equivalent to "the oil was spilled" would
>be "one spilled the oil".  The pronominal prefix translated "one" is in
>the agent form.  Not at all unusual.  Interestingly, however, if the agent
>is specifically identified, as in "the oil was spilled by the mechanic",
>"the mechanic" would be added to the above construction as an oblique, in
>a prepositional phrase, just as in English.
>
>Wally Chafe