Rare Dialogue Acts Common in Oncology Consultations
Mary McGee Wood, Richard Craggs
Department of Computer Science
University of Manchester
Manchester M13 9PL U.K.
mary, richard.craggs@cs.man.ac.uk
Ian Fletcher, Peter Maguire
Psychological Medicine Group
University of Manchester
Stanley House
Manchester M20 4BX U.K.
ian.fletcher, peter.maguire@cs.man.ac.uk
Abstract
Dialogue Acts (DAs) which explicitly en-
sure mutual understanding are frequent
in dialogues between cancer patients and
health professionals. We present exam-
ples, and argue that this arises from the
health- critical nature of these dialogues.
1 Background
We have described elsewhere (Wood, 2001; Wood
and Craggs, 2002) the use of dialogue analysis in
communication skills training for health profession-
als working with cancer patients. Our initial cor-
pus arises from a study of Macmillan Cancer Care
nurses undertaken by the Psychological Medicine
Group, University of Manchester, funded by the
Cancer Research Campaign. It consists of 37 dia-
logues between nurses and patients, each comprising
200-1200 utterances (mostly 300-600). The nurses’
goal is to learn as much as possible about the pa-
tients’ condition, both physical and mental, and to
inform the patients about their condition and treat-
ment. The dialogues are thus genuine, naturally oc-
curring conversations, but occurring in an unusual,
highly significant and emotionally charged situation.
We have not yet fully annotated a statistically
significant sample, but it is clear even from read-
ing through the corpus that a group of themati-
cally related DAs occur frequently which are rare
in previously studied corpora such as Switchboard.
These are DAs which explicitly establish or con-
firm accurate mutual understanding, either factual
or emotional, between the participants (collaborative
completions, summaries), or which build rapport
through courtesy and appreciation (thanks, apolo-
gies). Protracted closing sequences are charac-
teristic, and tend to have elements of both. We
interpret these patterns as direct responses to the
goal-directed and potentially health-critical nature
of these dialogues.
2 Rare dialogue acts
We take as our point of comparison the corpus of
some 200,000+ utterances from the Switchboard
corpus tagged with the SWBD-DAMSL tagset (Ju-
rafsky et al, 1998). Of these, 36% are “Statements”,
19% “Continuers”, and 13% “Opinions”, giving a
total of 68% of all utterances in the three most com-
mon categories. At the other end of the scale, an
original tagset of 220 was reduced to 42 because the
rarity of many made statistical analysis impossible.
Even of these 42, 32 occur with less than 1% fre-
quency, 25 less than 0.5%. Four of the five DAs we
will discuss here are among these last 25.
2.1 Mutual understanding
The first goal of the nurse in these dialogues is to as-
certain the subjectively perceived physical condition
of the patient, which reflects the success of previ-
ous treatment, and suggests directions for the future.
Factual accuracy is clearly essential if future treat-
ment is to be appropriate. It is also seen as important
that the patient have accurate knowledge of his / her
condition and treatment.
Secondly, the nurse is trying to elicit the mental
/ emotional state of the patient, and any particular
concerns or worries he or she has. Although this is
a somewhat different type of mutual understanding,
the same DAs - collaborative completions and
     Philadelphia, July 2002, pp. 196-200.  Association for Computational Linguistics.
                  Proceedings of the Third SIGdial Workshop on Discourse and Dialogue,
summaries - can effect both.
Collaborative completions
Collaborative completions are rare in most
dialogue types, where they would probably be seen
as pre-emptive interruptions. (The SWBD/ DAMSL
corpus includes 699/ 205,000, about 0.3%.) Here,
however, they are supportive, and relatively com-
mon. We have identified recurring patterns of both
factual and emotional use.
a. Factual completions: commonly, the patient is
not sure of the name of (e.g.) a drug or procedure,
and hesitates, whereon the nurse provides it:
P20 This is the, this is the ones I’m taking there.
N21 Right.
P21 Deta,
N22 Dexamethasone. (017-166)
P5 MST and,
N6 Antibiotics.
P6 Antibiotics and then by Sunday morning I had
come round a bit because all I did was sleep.
(038-395)
Sometimes the need for information is made ex-
plicit (here - as in the rest of the article - names have
been changed to maintain confidentiality):
P49 I did ask the Registrar, I forget his name Mr,
Mr Ferguson’s Registrar, I don’t, I forget his name.
N50 Birch, Mr Birch?
P50 That’s right, that’s the one. (047-291)
The same effect is also often achieved through an
overt question, more often on the part of the nurse:
N133 And the tablets that you take, the capsules
that you’ve just taken now?
P134 They’re Tylex, they’re a painkiller but I don’t,
I mean I don’t take them all the time they’re just
purely as a little top up. (035-215)
b. Emotional completions: the patient may be
struggling to find the right words, or reluctant to talk
about something. The nurse is showing understand-
ing and empathy with the patient, and encouraging
him/her to continue the conversation:
P80 : But that’s the sort of feeling I get a,
N81 : A tiredness. (042-277)
P328b But it’s not until it happens to you that,
N329 That sudden impact of gosh.
P329 That’s it. (024-113)
P162 I could feel the panic coming in me and before
the operation definitely but I’m,
N163 You’re okay. (057-356)
Summaries
The SWBD/ DAMSL “summarize/reformulate”
DA would seem laboured and unnatural in most
types of conversation, and indeed their corpus
includes only 919/ 205,000, about 0.5%. In our
domain, however, it is an entirely natural and useful
way of checking correct mutual understanding,
factual or emotional.
a. Repetition: simple repetition signals to the first
speaker that the second has heard and understood:
N297 I would have no problem in recommending
that you go from the 50 to the 75.
P298 Yes to the 75. (035-215)
The first speaker, on the other hand, may repeat in-
formation if the second appears not to have under-
stood (note the use of a collaborative completion in
this example):
P56 : 0 to 10, it went down then to about three.
...
N59 : So it went down to,
P59 : It went down to three. (042-277)
This sequence shows both patterns:
R128 Well she’s on 50mg of,
P129 Durogesic.
N129 Fifty?
R129 Durogesic, fifty yes.
N130 Fifty right. (035-215)
Where there is no apparent problem over informa-
tion, repetition suggests encouragement to continue
the conversation on that topic:
157 And you get very dizzy don’t you.
N158 That’s been an additional problem hasn’t it
that dizziness.
R158 The dizziness. (016-128)
b. Summary / paraphrase: a simple summary asks
for confirmation that the speaker has understood the
preceding dialogue correctly. Repetition may be
used to express the confirmation:
N25 : Mmmm. So yesterday you were sick twice.
P25 : Twice. For the first and only time. (042-277)
Summaries can also be used to bring the conversa-
tion back on track after a digression. The nurse may
initially wish to pursue a digression, in case it leads
to the revealing of a concern, but also needs to keep
the conversation focussed and ensure its goals are
met within an acceptable length of time:
P132 : I mean, and the parent, their separated
parents, I think it’s dreadful.
N133 : But I can also hear that, that from what
you’re saying you’re cross with, with being ne-
glected maybe for five hours or not having the
back-up ... (042-277)
N140 : Right. But coming back to what you were
saying earlier about, I’ve lost my frame of thought,
you mentioned earlier about wanting to make
sure that you get the right information, that it is
consistent. (042-277)
All these examples have been factual, but summaries
are also used to show empathy and understanding of
mental or emotional states:
P425 There’s nothing really honestly, if I took to my
bed or depended on someone.
N426 Right.
P426 That would be the end, oh it would.
N427 Your control, your independence.
P427 Yes, that’s more important to me at the minute.
N428 You want to stay really in the security of your
own home.
P428 Yes. (024-113)
2.2 “Social glue”
The dialogues in our corpus are not only more
important than most, they also occur in a complex
wider context. The nurse is part of an organisation
which is trying to save the patient’s life, using treat-
ments with painful, embarrassing, and depressing
side-effects. Everyone involved has more than usual
to be thankful or sorry for. Emotions run high. At
the same time, the whole enterprise depends on trust
and cooperation. Explicit courtesy, consideration,
and appreciation are essential “social glue”. No
wonder that thanks and apologies, both barely
represented in SWBD/DAMSL, are common here.
Thanks
Thanks occur only 67 times in SWBD/DAMSL,
probably not enough to be statistically recognised as
a separate tag, but clearly signalled lexically:
N115 Alright then.
P115 Right. Thank you very much. (027-334)
In our corpus, the nurse thanks the patient with
surprising frequency (even ignoring the artificial
cases of thanking the patient for allowing the con-
versation to be recorded). These seem to be part of
a general pattern of positive attitude and encourage-
ment:
N6 So thanks for doing this Karen, I just wanted
to come down and see you this morning because I
know we changed your medication a couple of days
ago. (027-334)
P80 So I’ve been doing, I’ve done everything you’ve
said and it’s working so far.
N81 Oh you’re wonderful, you’re wonderful, thank
you very much that’s really kind. (024-113)
Sometimes it is hard to draw a clear line between
thanks and appreciation:
N199 Is there anything you’d like to ask me?
P199 Mmmm no I think you’ve been very kind and
helpful. (030-318)
P204 But apart from that I’ve had some excellent
help and advice this year and from this week and
from the nurses, all nursing staff.
N205 Right.
P205 They’ve been excellent.
N206 Good, good.
P206 And it’s very helpful to your recovery.
N207 That’s good...
P207 Very helpful. (042-277)
Apologies
Apologies are similar to thanks (76 in
SWBD/DAMSL). Typically the nurse is reas-
suring the patient of her interest and attention,
perhaps after an interruption or some seeming
oversight:
N1 So go on I’m sorry to have interrupted you but
it was just that she had to get out. So it’s numb and
you can’t, (031- 198)
N105 Barbara I’m sorry I didn’t check out how
you are in here, how are you in this room, is that
alright? (057- 356)
N208 Bearing in mind there are a lot of questions I
get asked that I can’t give the answers to.
P209 Yes.
N209 I’m afraid, but I’ll come and see you next
week. (063-489)
Apologies can also be indirect or implicit:
(after talking about problems)
P314 ...poor girl you’ve got to listen to all that ...
(06-017)
2.3 Closures
“Conventional-closing” is the tenth most frequent
tag in SWBD/DAMSL (although still only 1%
(2486/ 205,000)). Our corpus is distinguished not
so much by the frequency as the nature of closures.
The evidence is incomplete, as in many dialogues
the tape runs out or is switched off before the end,
but typically the closure of a dialogue is long, and
explicitly negotiated. The nurse makes it clear to
the patient that the conversation is ending, in a way
which does not leave the patient feeling cut off or
abandoned (the imminent arrival of lunch is often
given as a reason):
N131 Alright then. I’ll cut it short so you can have
your lunch. (018-168)
Expressions of thanks and/or appreciation are com-
mon:
N114 Right, I wouldn’t envy you that job, you
have loads of problems with Councils. Shall we let
Charlotte back in?
P114 Yes, yes, yes.
N115 And then we’ll be able to have, we’ll see what
she has to do, you can fill the forms in because it
will be getting towards lunch time anyway, what
time is your lunch here?
P115 Oh usually about now.
N116 Right I’ll swap over and let Charlotte in ...
thanks very much that was brilliant. (025-153)
We often find a series of summarising statements
within a closing sequence:
N160 No. Right well I’ll see if there’s anything I
can do about that then and I’ll pop back next week
but give me a ring in the meantime if there’s any
problems. I’ll speak to Dr Clarke and see what he
thinks about this pain in your back.
P161 Yes.
N161 And see if there’s anything else that we can,
bearing in mind you’ve only been on the MST
properly for a short time.
P162 Mmmm.
N162 It might, we’ll just have to see how that works.
P163 Its not too bad today is my back.
N163 Right. Anyway I’ll pop in next week.
P164 Yes.
N164 And, you know, just monitor how things are
going and how you’re managing. Alright?
P165 Yes. (020-139)
N107 Have a great weekend.
P107 Yes.
N108 And we’ll catch up with you.
P108 Yes.
N109 Either in person or on the phone.
P109 Right.
N110 Next week when we know what else is hap-
pening.
P110 Okay.
N111 Other than that we’ll actually now step back.
P111 Right, yes, yes.
N112 Unless you need us for anything specific.
P112 Okay well I know where you are.
N113 Because you’re back in the hands of,
P113 Just the medical team.
N114 That’s right yes for your treatment.
P114 Yes, yes.
N115 Alright then.
P115 Right. Thank you very much.
N116 Okay see you soon.
P116 Yes, yes.
N117 Shall I walk you back down?
P117 Right thank you, I’m not quite straight yet but
I’m getting there. (027-334)
Both participants seem to be taking their last oppor-
tunity to check that mutual understanding is com-
plete. Sometimes contact details are given or new
topics arise at this point:
N558 Oh, I’ll finish off now anyway. (023-111)
The tape runs out at P604, after some repetition of
earlier topics and some new ones.
3 Analysis
Cancer care dialogues are health-critical. Misunder-
standings in casual conversation are unlikely to have
dire consequences: here, they easily could. Both
participants need to be unusually clear, and to ensure
that the clarity is mutual. This results in an unusual
predominance of DAs which establish and monitor
mutual understanding, both factual and emotional.
These dialogues are also emotionally charged, and
are eased by explicit appreciation and courtesy.
(Formal design of patterns to ensure clarity in dia-
logues can be found in safety-critical situations such
as military command and aviation. The use of repe-
tition to check and confirm understanding is charac-
teristic of Air Traffic Control:
Tower: BA117 descend 3,000 feet QNH 1017.
Pilot: Descend 3,000 feet QNH 1017, BA117.
Our dialogues are at the other end of the scale for
openness and unpredictability: it is interesting to see
similar surface devices used for the same purpose in
such different environments.)
These findings are somewhat impressionistic, and
taken from a relatively small corpus. As soon as
we have analysed a larger sample in more detail,
it will be possible to verify and quantify these pat-
terns, and to analyse the linguistic characteristics of
DAs which have previously eluded us. Also, further
comparisons can then be made with other corpora
and previous work on dialogue analyses.
Acknowledgements
Appreciation for their support of this research goes
to Manchester University Computer Science Depart-
ment for funding Richard Craggs and Cancer Re-
search UK for funding Ian Fletcher.
References

Daniel Jurafsky, Elizabeth Shriberg, Barbara Fox
and Traci Curl. 1998. Lexical, Prosodic, and
Syntactic Cues for Dialog Acts. ACL/COLING.
Mary McGee Wood 2001. Dialogue Tagsets in
Oncology. Proceedings of SIGdial2.
Mary McGee Wood and Richard Craggs. 2002.
Language Engineering meets Oncology Training.
Paper submitted to the Edilog workshop on Seman-
tics and Pragmatics of Dialogue, Sept 2002.
