Ensemble Interactions - Queen Mary University of London

Ensemble Interactions
G. Mills (University of Edinburgh)
P. G. T. Healey (Queen Mary University of London)
Ensemble Interactions
G. Mills (University of Edinburgh)
P. G. T. Healey (Queen Mary University of London)
Interaction in Dialogue
But, dialogue typically involves more than 2 people.
The F-Formation
(Kendon, 1992)
Levels of participation in an interaction
Main
participants
Side -Participant
Overhearer
Eavesdropper
(Goffman, 1976, Clark 1996)
Levels of participation in an interaction
Ensemble
Primary participant
Primary participant
Peripheral
Establishing referring conventions within an ensemble
(0 mins)
(5 mins)
(10 mins)
(15 mins)
(20 mins)
(25 mins)
The man touching his toes, leaning forward, has a funny back
The man leaning forward to touch his toes with funny back
The man leaning forward with the sticking out back
Sticking out back guy with the square head
The hunchback with the square head
The hunchback
(Krauss and Weinheimer 1966, Clark and Wilkes-Gibbs 1986)
Production of referring expressions
• Primary participants use longer and different
descriptions with peripheral participants (Clark and
Wilkes-Gibbs 1982; Brennan and Clark, 1986)
• Primary speakers create different novel
descriptions than with peripheral participants
(Gann & Barr 2013, Oberlander 2011)
• Co-ordination develops at a different rate within
ensemble vs. with peripheral participants (Healey
and Mills, 2006)
Comprehension
• Peripheral participants are worse at identifying
referents (Schober and Clark, 1992)
• If a primary participant uses a different referring
expression to refer to the same object, this
interferes with comprehension.
If a peripheral speaker uses a different description
no such effect (Metzing and Brennan 2003)
• Interlocutors take into account referential intentions
of primary participants from earliest moment of
linguistic processing (Brown–Schmidt et al, 2007)
Ensembles
Conventionalization and partner-specificity are driven
by interaction in an ensemble.
Two questions:
1. Are referring conventions the only kind of
convention that develop in ensembles?
2. How dynamic are ensembles - can ensembles
(re)form during the interaction
Conventionalization in an ensemble
How do two or more people come to use the same
representations, ideally on subsequent turns, to refer to
the same objects, leading to the same behaviour or same
representations, or same beliefs about the same context?
Some initial concerns:
• Is emphasis on representational parity warranted?
• Dialogue is underpinned by normative sequential
constraints consisting of complementary contributions
(cf. Bekkering et al 2007)
Adjacency / projective pairs
First pair part
Second pair part
Greeting
Question
Request
Praise
Greeting
Answer
Compliance
Self-deprecation
Approach ticket window
"Where would you like to go?"
"Timber"
Place object on counter
Jump out of way
"That'll be X Pounds"
(Sacks, Schegloff, Jefferson, 1977; Clark, 2002)
Adjacency / projective pairs
First pair part
Second pair part
Greeting
Question
Request
Praise
Greeting
Answer
Compliance
Self-deprecation
Approach ticket window
"Where would you like to go?"
"Timber"
Place object on counter
Jump out of way
"That'll be X Pounds"
(Sacks, Schegloff, Jefferson, 1977; Clark, 2002)
Some more concerns…
• Adjacency pairs underpin procedural co-ordination
in dialogue (sequentiality, temporality)
• Cognitive and Conversation analytic approaches
presuppose they are already shared
(and known to be shared) by interlocutors.
• How are they established and sustained?
• How partner-specific are they?
Some more concerns…
• Adjacency pairs underpin procedural co-ordination
in dialogue (sequentiality, temporality)
• Cognitive and Conversation analytic approaches
presuppose they are already shared
(and known to be shared) by interlocutors.
• How are they established and sustained?
• How partner-specific are they?
Method
Question:
How does procedural co-ordination develop in dialogue?
Task:
Make reference as easy as possible, but present
participants with recurrent procedural co-ordination problem.
Experiment 1:
Test for development of procedural co-ordination
Experiment 2:
Test for partner-specific effects
The alphabetical task‫‏‬
Participant 1
Your words are:
Banana, peach, tomato
Participant 2
Your words are:
Apple, pear
Alphabetical task
• Participants must combine their lists into a single
alphabetically sorted list.
• Participants can only select each other's words
• Participants see effects of each other's actions
• Participants select words by typing the word
preceded with backslash: "/"
• Participants can restart at any time by typing
/restart
Procedural co-ordination problem
Reference (EASY):
Refer to list of mutuallyknown words
Sequencing (HARD)
Signal initiation and
completion of each
selection.
The alphabetical task‫‏‬
Participant 1
Your words are:
Banana, peach, tomato
Participant 2
Your words are:
Apple, pear
Alphabetical task
Participant 1
/apple
Participant 2
/banana
/peach
/pear
/tomato
Conventionalized adjacency pairs
apple
/apple
/apple
“One of the words you will need to select is apple, tell me
my list in alphabetical order, and then, if you can start with
the first word then please do and then tell me when.”
“I've just selected the sub-list of words you told me to. You
need to select the list that starts with apple, and then tell
me when you’re done by telling me the next word in the list
that I need to select.”
I’ve just selected the sub-list of words that ends with apple,
you need to select the next word in your list, when you’re
done tell me what the last word is that you selected.
Procedural conventionalization in dialogue
Question 1
How partner-specific are these conventions?
Question 2
How dynamic are ensembles - can ensembles
(re)form during the interaction
3 participant version of alphabetical task
Participants alternate between:
• An ensemble of 3
• An ensemble of 2 and 1 peripheral participant
Interaction is constrained so that
• Some conventions can only be established within
ensemble of 3 vs. ensemble of 2.
Testing for partner-specific effects in both ensembles
• Insert artificial clarification requests into dialogue
that probe for partner-specific effects.
3 participant version of alphabetical task
Participants alternate between:
• An ensemble of 3
• An ensemble of 2 and 1 peripheral participant
Interaction is constrained so that
• Some conventions can only be established within
ensemble of 3 vs. ensemble of 2.
Testing for partner-specific effects in both ensembles
• Insert artificial clarification requests into dialogue
that probe for partner-specific effects.
3 participant version of alphabetical task
Peripheral
Primary
The words that I
have are peach and
banana for Main
Ensemble of 3
All 3
participants
must
combine
lists
Primary
mine are banana,
apple and peach
Ok I have peach
for Main2
Is that all?
Yeah,
I’ll start with apple
Ensemble of 2
Only primary
participants
need to make
selections.
Peripheral does
not need to
make selections
OK
/APPLE
/BANANA
Right I did it
So you did peach?
/PEACH
Now I have,
OK great
3 participant version of alphabetical task
Peripheral
Primary
The words that I
have are peach and
banana for Main
Ensemble of 3
All 3
participants
must
combine
lists
Primary
mine are banana,
apple and peach
Ok I have peach
for Main2
Is that all?
Yeah,
I’ll start with apple
Ensemble of 2
Only primary
participants
need to make
selections.
Peripheral does
not need to
make selections
OK
/APPLE
/BANANA
Right I did it
So you did peach?
/PEACH
Now I have,
OK great
Low co-ordination (Initially)
Peripheral
Primary
Primary
Ensemble of 3
Ensemble of 2
Low co-ordination (Initially)
Peripheral
Primary
Primary
Ensemble of 3
Medium
co-ordination
(25 min)
Periph
Primary
Primary
Ensemble of 3
Ensemble of 2
Ensemble of 2
Low co-ordination (Initially)
Peripheral
Primary
Primary
Ensemble of 3
Medium
co-ordination
(25 min)
Periph
Primary
Primary
Ensemble of 3
High
co-ordination
(50 min)
Periph
Primary
Primary
Ensemble of 3
Ensemble of 2
Ensemble of 2
Ensemble of 2
Testing for partner-specificity
Method:
Use the chat server to insert “spoof” clarification requests
into the dialogue that probe the procedural function of the turn.
Spoof
clarification
generated
by the server
Primary
: It’s apple now
Peripheral: apple?
Primary
: yeah select it in 5 sec
Compare responses to clarification from Primary vs. Peripheral
Artificial “spoofed” clarification requests
A: Next is tomato
A: Next is tomato
A: Next is tomato
Server relays message to other participants
Artificial “spoofed” clarification requests
A: Next is tomato
A: Next is tomato
A: Next is tomato
Server detects target word tomato
Artificial “spoofed” clarification requests
A: Next is tomato
A: Next is tomato
A: Next is tomato
Network error: Please wait
Network error: Please wait
SERVER instructs other chat clients to block text-entry
Artificial “spoofed” clarification requests
A: Next is tomato
B: tomato?
A: Next is tomato
A: Next is tomato
Network error: Please wait
Network error: Please wait
SERVER sends fake clarification request to participant A
Artificial “spoofed” clarification requests
A: Next is tomato
B: tomato?
A: Next is tomato
A: Next is tomato
Network error: Please wait
Network error: Please wait
SERVER waits for response from participant A
Artificial “spoofed” clarification requests
A: Next is tomato
B: tomato?
A: Next is tomato
A: Next is tomato
Network error: Please wait
Network error: Please wait
do it now
SERVER waits for response from participant A
Artificial “spoofed” clarification requests
A: Next is tomato
B: tomato?
A: do it now
A: Next is tomato
A: Next is tomato
Network error: Please wait
Network error: Please wait
SERVER captures response
Artificial “spoofed” clarification requests
A: Next is tomato
B: tomato?
A: do it now
A: Next is tomato
A: do it now
A: Next is tomato
A: do it now
SERVER unblocks screens and relays message to others
Artificial clarification requests
• Inserted after 45 mins (to allow co-ordination to develop)
• Inserted randomly (every ~ 40 turns)
• Sent only to Primary participants
Appear to originate either from:
– Primary participant
– Peripheral participant
Inserted either in :
– 3 participant Ensemble
– 2 participant Ensemble
=> 2 x 2 Factorial design (Apparent Origin / Ensemble)
High co-ordination
(50 min)
Peripheral Primary
Primary
Peach
Peach?
It’s yours
Artificial CR
Response
tomato
Tomato?
Yeah
/PEACH
peach
Peach?
Done
/TOMATO
tomato
Tomato?
What?
Artificial CR
Response
Compare
responses to
clarification
Compare
partner-specific
effects in both
Ensembles
Artificial CR
Response
Artificial CR
Response
Compare
responses to
clarification
Hypotheses
(H1) “Coarse” partner-specificity
• Clarification requests from side-participant will cause more
overall disruption than clarification from main participant.
(H2) Specificity of ensemble
• In 3-participant ensemble: No difference
• In 2-participant ensemble: Clarification from the
peripheral participant will cause more disruption.
Typing time (msecs) of response
Primary
Peripheral
Typing time
of response
(msecs)
3 Ensemble
2 Ensemble
Typing time (msecs) of response
Primary
Peripheral
Typing time
of response
(msecs)
3-Ensemble
2-Ensemble
Length (chars) of responses
Primary
Peripheral
Chars
3-Ensemble
2-Ensemble
Self-corrections
Primary
Self-corrections
Non-corrected
*
Peripheral
3-Ensemble
2-Ensemble
Task performance (% correct)
Correct
Error
Primary
Peripheral
*
3-Ensemble
2-Ensemble
Partner-specific effects
In the 3 participant ensemble, clarification from Peripheral participants
• Cause more disruption to participants’ turns
• Cause less disruption to task performance
Why?
• In the 2-participant ensemble Primary participants know they can
safely ignore Peripheral participants
• In the 3-participant ensemble, clarification from the Primary
participants are interpreted as instructions, resulting in wrong
selections
Partner-specific effects
In the 3 participant ensemble, clarification from Peripheral participants
• Cause more disruption to participants’ turns
• Cause less disruption to task performance
Why?
• In the 2-participant ensemble Primary participants know they can
safely ignore Peripheral participants
• In the 3-participant ensemble, clarification from the Primary
participants are interpreted as instructions, resulting in wrong
selections
Clarification request or instruction
Peripheral
Primary
Primary
apple and
banana
Peach tomato
Blackberry, tomato
blackberry?
/blackberry (ERROR)
Selected it
“Can you select
blackberry now?”
Conclusions
• Interlocutors rapidly develop highly elliptical
language and routines for managing procedural
co-ordination.
• Interlocutors rapidly establish Ensembles
– Differentiation of “procedural” contexts for
primary vs. peripheral participants
– Participants alternate between different participatory
structures during single interaction.
Conclusions
• Interlocutors rapidly develop highly elliptical
language and routines for managing procedural
co-ordination
• Interlocutors rapidly establish Ensembles
– Differentiation of “procedural” contexts for
primary vs. peripheral participants
– Participants alternate between different participatory
structures during single interaction.
Thanks
Herb Clark, Eleni Gregoromichelaki, Kerstin Fischer.
Marie Curie PIOF-GA-2009-236632-ERIS