[Avery Andrews 970206]
I've fnally managed to read a (very short & light) Conversation Analysis
book, so here are some impressions.
The book is `Conversation Analysis: the Study of Talk-in-Interaction',
George Psathas, Sage Publishers, intended as a quick introduction to
the field.
The motivation for looking at CA is that unlike standard linguistics,
CA focusses on the interactional aspects of language use, where the
activities of control-systems are most likely to manifest themselves
(as I've mentioned before, I consider grammar to be primarily a
modelling facility; when it works right, you don't see much evidence
for control-systems).
The idea of CA seems to be to understand interaction through a very
close examination of naturally occurring examples, beginnings of
telephone conversations being an especially favored topic of
investigation. It is a branch of `ethnomethodoloy', based on work
by Harvey Sacks, Erving Goffman and Harold Garfinkel, and is in part
a reaction against some approaches to interaction that arose in
sociology after WW2, which were based on the application of statistics
to the results of coded descriptions of interactions in terms of various
category systems. Features of CA include (a) no use of mass statistics
(b) allegedly no use of preconceived category systems (I'm not sure if
they really manage to do this) (c) no use of experiments, only
observation of natural, spontaneously occurring events (d) public
availability of the primary data. (a) and (d) would presumbly make
a good impression on PCT-ers, (c) is discrepant with PCT methodology,
so I'll say something in defense of it.
If you want to know whether something, such as, say control, is
important in everyday activities, you have to actually look at these
actvities, rather than something else, such as say lab experiments.
We know that basic natural processes don't care about social situations,
but this is definite false for people, so it's quite possible that
you could build an elaborate theory explaining lab observations that
had nothing at all to do with what goes on outside of the lab. So
if you you want to know about is ordinary conversation, that is what
you have to look at. And of course the possibilities for serious
experimentation in ordinary situations is extremely limited, since
these interactions always have some purpose for the participants, which
experiments (at least the ones that are easy to think of, such as not
responding to greetings, or repeating them when they've already been
responded to) would tend to screw up, with real damage to the lives
of the participants.
Of course, contemporary ethics requires that the participants in the
interactions know that they are being monitored; however the researches
don't think this knowledge has much effect on what's going on (I guess
an unscrupulous doubting Thomas could make recordings surreptitiously,
to really be sure ...)
There is quite a bit more than this to the ideology and methodology of
CA, I suspect that it's overdone in various ways (similarly to
generative grammar), but I don't want to linger on philosophical issues.
CA claims to uncover quite a variety of structures in ordinary
conversation, one of the most important of which is the `adjacency
pair'. Here are some samples, of various recognized types:
Telephone Summons-Answer:
(ring)
Greeting-Return Greeting:
B: Hi
Compaint-Apology/Justification:
B: Oh, sorry B: The traffic held me up
Psathas delineates the major features of adjacency pairs as follows
(unbracketed stuff are his exact words, there seems to be a lot
of redundancy):
1. They are (at least) two turns in length
[actually, I can't figure out what a >2 length adj. pair would be]
2. They have (at least) two parts
3. The first pair part is produced by one speaker
4. The second pair part is produced by another speaker
5. The sequences are immediate next turns
[even slight delays can be indicative of trouble]
6. the two parts are *relatively ordered* in that the first belongs
to the class of first parts, the second to the class of second
parts. [For the kind of pair in question, I think, different kinds
of pairs have different possibilities for first and second parts.]
7. The two are *discriminatively related* in that the pair type of
which the first is a member is relevant to the selection among the
second pair parts.
8. The two parts are in a relation of *conditional relevance* such that
the first sets up what may occur as a second, and the second
depends on what has occurred as a first.
I propose that adjacency pairs are the activities that result when the
initiator does something to attempt to correct an error-signal regarding
some perception of the mental/emotional state of the responder. If the
responder's response causes the initiator to perceive the responder to
be in the desired state, the pair concludes successfully, otherwise
there's `trouble' (seems to be a technical term in CA, I'm not sure I
know what they really mean by it).
This claim works out quite nicely for the `Summons-Answer' type, the
kind of pair most thoroughly described by Psathas. In a summons-answer
pair, the initiator seems to be trying to attain a perception that
the addressee is (a) able to hear the initiator (b) willing to listen
(c) in an appropriate frame of mind for the intended sequel (you wouldn't tap
somebody on the back in order to get their attention to tell them that
their mother had just died). Psathas enumerates the following
characteristics of Summons-Answer pairs:
1. They are two turns in length
2. First speaker produces the first pair part.
3. Second speaker produces the second pair part. [these guys certainly
do get pedestrian ...]
4. They occur at the beginning of interaction sequences.
5. They are nonterminal in that they do not end encounters.
6. The speaker who speaks the first part, the summons, is obliged to
speak (though deferral may be requested, as in "wait a minute", or
a change of mind indicated, as in "oh, never mind")
7. The answerer of the summons is obliged to listen, thoough
availability to listen may be deferree (as in "just a moment").
8. They are nonrepeatable once an answer occurs.
9. They [actually, their first pairs, I think] may be repeated if not
answered
10. They need not be verbal but could, for example, involve taps on the
shoulder, hand waving nods, head shakes and so on.
8-9 are particularly indicative of control-system activity, as is 10,
since, fundamentally, any action to direct an intended audience's
attention to the intending speaker can serve the purpose of getting
the reference level for `the intended listener is listening' to be
satisfied (although there is a lot of ritualization about what kinds of
attention-getting activities are appropriate to different kinds of
social situations).
I think that the other features of Summon-Answer pairs can be derived
with the aid of an AI-inspired import into PCT, the idea of a perceptual
precondition for output. Suppose you want to taste your beer. If
the bottle is within reach, you stick out your hand towards it, but if
it's on the other side of the room, you don't do this until you are
within reach (or will be by the time the hand is reasonably extended).
The general point is that when the attainment of a goal involves a
number of activities, each activity typically doesn't commence until
certain preconditions for its potential success are satisfied
(under extreme situations, this can break down, people frantically
sticking out their arms through the bars towards the totally out-
of-reach keys to the cell, as the water level rises ...). There
has been some inconclusive discussions of what kinds of PCT
architectures can produce this sort of of effect, searching the
archives for `beer' & `grasp' would probably turn it up. One
possibility is to have a perception of the precondition `gate' the
reference for the activity, so that when the precondition (beer within
reach) is not satisfied, the perceptual reference (to be holding it)
does not make any contribution to the perceptual reference for
hand-position. Then, when the beer is perceived as within reach,
the gate opens, and the location of the beer becomes the reference
for position of the hand (there's a lot more to it than that of course,
with the grasping, etc.).
Regardless of these details, several of the additional properties of the
summons-answer pairs can be explained on the basis that `intended
addressee listening' is a precondition for some other activity that
is serving a deeper reference level (such as, for example, to perceive
someone has knowing that the garbage has to be taken out).
Here's a diagram, as close to PCT conventions as I can get it, although
the inherently discrete nature of this kind of system raises
difficulties, along with doubt about how to deal with perceptions &
references involving combinations of categories & relationships.
> ref = 'yes'
v
A. knows that gargage ----
must go out? |
>
^ --------------------------------------
> > >
> v neg gate |
> A. listening? ----------------------------->|
> ^ | |
> > (if A says `yes', | v
> > this perc. is set | perceive self having
> > to 'yes') | said something that
> > imples that the garbage
If A. says `ok' to v must go out |
statement, perc. is circuits for attention- ^ |
set to `yes'. getting activity (e.g. | v
^ 'honey?' (means I want | talk
> you to do something) |
> > linguistic
> proprioception
>
v
Note that in this case, the precondition is something that the
system can itself try to bring about, so that an error for the
higher-order perception sets a reference for both the precondition
and the perception for the activity that is supposed to satisfy the
higher-order perception; there could also be situations where the
system simply had to wait for the precondition to be satisfied (you
want to announce something to your wife, she's out shopping, so you
just have to wait until she gets back).
Properties 4-6, I claim, follow from the idea that the Summons-Answer
sequence is there to satisfy a precondition for the satisfaction of the actual
perceptual reference that precipitates the communication. It's got to
be at the beginning, since it's a precondition for anything else,
its `pre-terminal' because it gets the addresee in to a state where
they're expecting to hear something, and for the same reason it's the
initiator who has to say that something.
Still unexplained is property 7: why is the answerer obliged to listen,
while another issue is whether and when the summonee is obliged to
answer. I think that this has to do with the references for various
perceptions involving social relationships, but I don't want to try to
work out the details here. In general, I suspect that working out what
listeners references are might be pretty challenging, and an area
where PCT might engender some real understanding.
The more general point is that the logic of control-systems and preconditions
produces temporal scheduling of a kind that is somewhat similar to
what might be gotten out of a `grammar of discourse', but in a much more
flexible manner. Such a grammar might have rules:
Conversation -> Summons-Answer Body-of-Conversation
Summons-Answer -> A:Summons B:Answer
Summons -> "Hello", "Hi", "Honey", "Yo <Name>", ...
Answer -> "yes", "yeah", ....
But it is then a complete surprise that if the Summons goes unanswered,
it may be repeated (this requires an elaboration of the rule-system),
while this follows from the logic of preconditions and control-systems.
Likewise there is no account of the wide variability of fillers for the
Summons position, while the usability of various means to attain a given
reference is an essential feature of control-system architecture (the
choice will be governed by the combination of the attention-getting
reference with other references for the social situation, such as level
of seriousness).
And I believe that the adjecency-pair structuring itself follows from
certain features of these control systems, such as the use of an
auditory channel that can accomodate only one sender at a time, but
that remains to be worked out in detail.
The utility of precondition logic shows up in another popular area of
study, participatant identification in phone calls. For many kinds of
phone-calls, the first order of business is for the caller and answerer
to identify themselves to each other. (Schegloff 1979 `Identification
and Recognition in Telephone Conversation Openings', in Psathas (ed)
_Everyday Language: Studies in Ethnomethodology_, Irvington: New York).
This is, I suggest, largely because establishing these identities is
usually a precondition for attaining whatever perceptual reference the
call is in aid of. Schegloff shows that recognition can be intertwined
with various other activities, that if mis-recognition is detected,
identity-establishing activities will recommence, etc., just as one
would expect from a preconditions-based account.
And of course it also follows that these recognition steps may well be
omitted in transactions where personal identities aren't relevant,
such as calls to emergency services, or `getting-the-right-person'
transactions:
1. (ring)
2. A; Hello
3. B: Is Owen there?
4. A: I'll see
Here B has a perceptual reference for talking to Owen, since that's a
precondition for whatever B really wants to do; B detects from A's voice
quality that B is not talking to Owen, so in 3. says something that
would tend to get this error corrected, and in 4. A says something that
will cause B to think that progress is being made towards the
satisfaction of his reference (why does A want B to perceive this,
something to do with maintaining social relationships, I conjecture).
In this case, the identities of A and B aren't crucial, so might not be
asked. Interestingly, in our culture at least it seems to be polite for
A to want to know who B is (step 4. could be `who's calling please'),
while acceptable inquiry of B about A is a lot more limited, I think
to asking whether A is some specific person who B knows and would want
to greet on any casual encounter. So there's a lot to pick apart here,
and it remains to be seen whether PCT can be applied in detail in a
useful way (as opposed to being just another system of preconceived
categories being foisted upon the data). But, in general, preconditions
logic seems able to generate the basic temporal phenomena in an
insightful way, which for example `discourse grammars' on the model
of sentence-level syntax could not, since they can't *explain* the
early ordering of identificational activites, their interleaving with
others, their resumptions in case of error, or their omissability in
certain circumstances.
So I'd suggest that CA is worth further looking into, for people
interested in language and PCT.
···
A: hello
A: Hi
A: You're late for class A: You're late for class