discomfort

[Kerstin Fischer]

Dear Professor Samuels,

I have recently started thinking about the questions you have posted
myself, and I found an answer for my problem, which probably won't be
useful for your field methods class, but here we go:

You have asked:

> Anyway.  She's faced with two broad problems.  First, the
> ethnomethodological one, of how she can reasonably interpret certain cues
> as indexing discomfort rather than something else (the blink/wink thing);
> second, you know, ethnographically, it's hard for her to go up to someone
> and say "gee, you look really uncomfortable, can I interview you about it?"
> That would seem to exacerbate the problem.

The concept I am dealing with is emotionality in human-computer
interaction, and the problems I have are similar to those of your student:
How do I know that speakers are really angry and that the behaviour they
display isn't rather caused by anything else? What I have done to
overcome this problem is to design a Wizard-of-Oz sceanrio, that is, a
scenario in which speakers believe to be talking to an automatic speech
processing system but in fact the system output is simulated by a human
"wizard". In this sceanrio I employ a fixed dialogue structure with a
number of phases which consist of fixed sequences of utterances. For
instance, such a phase may consist of a request to propose a date (the
project is in the appointment scheduling domain, the Verbmobil
speech-to-speech translation project), followed by a statement about the
holidays, that new year is a holiday and other irrelevant utterances, a
rejection of a proposal and a claim not to have understood. The system
"utters" its prefabricated units irrespective of what the user is saying
in a fixed order. Each phase, i.e. sequence of utterances by the system,
are repeated several times throughout the dialogue. Sometimes proposed
dates are also accepted. The speaker believes to be talking to a computer
which sometimes understands and sometimes misinterprets her utterances.

The methodological point about this setting is now that since each
sequence occurs several times in the dialogue, I can compare how the
speaker reacts to the same (simulated) system output when she is
confronted with it for the first time, the second, and so on. Observable
changes can then be attributed only to changing speaker attitude towards
the system. Thus I can study how irrelevant utterances vs. claims not to
have understood, etc. influence the speaker's linguistic behaviour, and
how this behaviour may change through time.
(Results are, for instance, that initially speakers reformulate their
utterances in order to facilitate understanding, they begin to
hyper-articulate, to lengthen their syllables, etc. (some of these
strategies being contra-productive since hyper-articulated speech is even
more difficult to understand for an automatic speech processing system),
yet after some time they kind of give up and keep repeating their
utterances only. Sometimes they sigh, laugh, or use swear-words.)

Self-evidently I ask speakers after the recording whether they were
emotionally involved; so far they all said yes, though not all were angry,
a few were also just amused.

However,

> If anyone's got any ideas about how these questions can be addressed in the
> context of a fieldwork class, I'd be really appreciative.

I have no idea how to extend this method to natural settings. Regarding
the methodology of analysing emotionality in conversation I find the book
by Reinhard Fiehler (1990): Kommunikation und Emotion. Berlin, New
York: Mouton de Gruyter. very helpful. He takes those situations as a
starting point in which emotional experience (I've had an awful day
today) or emotional interpretation (You look so angry, what's the matter?)
is made a topic of the interaction. I don't know, however, whether
speakers talk about feeling uncomfortable, that is, probably you'll be
mostly interested in situations in which people feel too uncomfortable to
say that they are uncomfortable. However, what you could do it take those
situations about which you know for sure that speakers are uncomfortable
(either because they say so themselves, or their partners brought it up,
or because you made them undergo a situation similar to mine) as a
starting point and look for the same indicators in other situations which
you suspect to be uncomfortable. May be tricky since most signs are
multi-functional, but I would guess it's the only thing you could do. I'm
curious about whether someone else on the list has an idea.

Good luck, yours sincerely,

                    Kerstin Fischer.