predictive or reactive Laura Lindsay Chiara Gambi Martin Pickering amp Hugh Rabagliati Department of Psychology University of E dinburgh The turntaking puzzle During conversation we take turns between speaking and listening ID: 618894
Download Presentation The PPT/PDF document "Turn-taking in children and adults:" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Turn-taking in children and adults: predictive or reactive?
Laura Lindsay, Chiara Gambi, Martin Pickering, & Hugh RabagliatiDepartment of Psychology, University of EdinburghSlide2
The turn-taking puzzleDuring conversation, we take turns between speaking and listening:
Short gaps/overlaps: 200 ms (Stivers et al., 2009)Production processes take time
– at least 600ms for a single contentword (Indefrey &
Levelt, 2004)
Data from
Stivers
et al., 2009,
PNAS
GAPS
OVERLAPSSlide3
Three questions about turn-
takingPREDICTIONEARLY PREPARATION
DEVELOPMENT
Garrod and Pickering, 2015; Levinson 2016; see also De Ruiter et al., 2006; Magyari
et al., 2014.Slide4
1. Prediction
“cake” (CONTENT)
How can
you
not
like
birthdays
!? Are you not excited
at the idea of eating
the
…
From
2;0
, children
observing
a conversation look at the next speaker before she begins
speaking
(Casillas
& Frank,
2012, Lammertink et al, 2015; see also: Keitel et al., 2013; Keitel & Daumm, 2015)
From
2;0, children make semantic predictions (Mani & Huettig, 2012)
Do children predict timing?
??
400-600ms (
TIMING
)Slide5
2. Early preparation
How can
you
not like birthdays !? Are you not
excited at the idea of
eating
the
…
NO!
Bögels
et al., 2015
Might
be
possible
when
:
Speaker’s
turn
is
predictableListener’s turn is short and simple
However:
children might struggle with early preparation
because their planning abilities are slow to develop(Casillas
and
colleagues
)Slide6
3. The development of turn taking
0;3
0;9
1;0
1;6
2;0
3;0
700
ms
1100
ms
Hilbrink
et al., 2015
Corpus
studies
: 3
months
to 18
months
years
;
monthsSlide7
3. The development of turn taking
0;3
0;9
1;0
1;6
2;0
3;0
700
ms
1100
ms
Hilbrink
et al., 2015
Casillas
et al., 2016
900
ms
5
00
ms
years
;
months
Corpus
studies
: 2 to 3 and a
half
yearsSlide8
3. The development of turn taking
0;3
0;9
1;0
1;6
2;0
3;0
700
ms
1100
ms
Hilbrink
et al., 2015
1200
ms
8
00
ms
(
around
5;0)
Garvey
and
Berniger
, 1981
Mother-child
C
hild-
child
Casillas
et al., 2016
900
ms
5
00
ms
years
;
monthsSlide9
3. The development of turn-taking
Why does it take so long?Children’s production
abilities take long to develop early
preparation is limited
children predict
semantics
(CONTENT) and can
prepare
early, but they
cannot not yet
predict
when
the
speaker’s
turn
is
going
to end (TIMING)
Do children (and adults) predict
when the speaker will stop talking
by predicting what she
is about to say?Slide10
This study
Interactive iPad-based maze game (4 mazes) Slide11
This study
Example trial24 adults30 5-year-olds (one participant excluded, N=29)47 3-year-olds (13 participants excluded, N=34)
Participants
were reminded of the characters’
names before each
mazeSlide12
DesignFully crossed, within-subjects:
Answer Type Is the content of the question (and its answer) predictable (YES answer) or not (NO answer)?Participants expect Peter Pan to ask about the correct direction (he is always right on filler items)Scene Type
The game has 4 mazes, each with 36 trials (24 target trials and 12 filler trials) Slide13
Predictable
Predictable
Unpredictable
UnpredictableSlide14
Predictable
Predictable
Unpredictable
UnpredictableSlide15
DesignFully crossed, within-subjects:
Answer Type Is the content of the question predictable (YES answer) or not (NO answer)?Participants expect Peter Pan to ask about the correct direction (he is always right on filler items)
Scene TypeAre the names of the two characters the same length?MATCH: both short / longMISMATCH: one short / one long (mean difference: 430 ms
)Slide16
Match
Mismatch
Mismatch
MatchSlide17
Match
Mismatch
Mismatch
MatchSlide18
Match,
Predictable
Mismatch
,
Predictable
Mismatch
,
Unpredictable
Match
,
Unpredictable
Match
Mismatch
Mismatch
MatchSlide19
Hypotheses: gaps
ANSWER TYPE: longer for unpredictable (NO answer) than predictable questionsANSWER TYPE * SCENE TYPE: if participants predict when the question will end by predicting its content;
interaction:
When the characters’ names mismatch in length, we should find a larger difference between predictable and unpredictable questions (compared to when the names match in length) Slide20
Hypotheses: gaps
ANSWER TYPE: longer for unpredictable (NO answer) than predictable questionsANSWER TYPE * SCENE TYPE: if participants predict when the question will end by predicting its content;
interaction: Slide21
Hypotheses: gaps
ANSWER TYPE: longer for unpredictable (NO answer) than predictable questionsANSWER TYPE * SCENE TYPE: if participants predict when the question will end by predicting its content;
interaction: Slide22
Hypotheses: gaps
ANSWER TYPE: longer for unpredictable (NO answer) than predictable questions
ANSWER TYPE * SCENE TYPE: if participants predict when the question will end by predicting its content;
interaction:
Decrease with age
Adults < 5;0 < 3;0Slide23
Results
* - Answer TypeAll age groups take longer to respond to unpredictable (NO answer) than predictable questions (YES answer).
* Gaps < 2 sec; Linear mixed-effects
models with maximal random structure; |t| > 2 means
p<.05
56
ms
, t = 3.25
103
ms
, t = 3.37
119
ms
, t = 4.29Slide24
Results
* - Answer Type: Scene TypeNO interaction between Answer Type and Scene Type in any age group.
t = 0.58
* Gaps < 2 sec ; Linear mixed-effects models
with maximal random structure; |t| > 2
means
p
<.05
t = 0.46
t =0.58Slide25
Results*
– Early preparation (?)Exploratory analysis: is there evidence for early preparation?Character names varied in length
If answer preparation starts before question end:
Longer
character
names
Shorter
gaps
More
preparation
timeSlide26
Results
* – Early preparation (?)The longer
the character name, the faster the
participant’s response
r(45) = -.66
r
(45)=-0.70
r
(45)=-.50
*
Gaps < 2
sec; by-item
correlationsSlide27
Results –
Distributional AnalysisDo children just get faster?
Ex-gaussian
distributionThree parameters:Mu (mean
)Sigma (standard deviation)
Tau (
thickness
of the
tails
)
Tau = 100
Tau = 200
See
Ratcliff
, 1979Slide28
Results*
* Parameters jointly
estimated with Bayesian linear regression
Children’s slow responding
is driven by differences in the right
tail
of the
distribution
Effect
of
Answer T
ype and lack of interaction
Answer
Type
* Scene
Type
replicate on muSlide29
Predicting Timing?
Neither children nor adults timed their answers to questions by predicting the length of the final word
Predictions:
“Fireman Sam” (CONTENT)
600-800ms (TIMING)
Should
we
go
past
…
Fireman Sam?
Do
listeners
predict
question
length
/
structure
?Slide30
Early Response Preparation
Instead, they rapidly prepared their answer as soon as possible, and responded reactively
Predictions:
“Po” (CONTENT)
Should
we
go
past
…
[p]?
Po is a short word, so there is less time to prepare a response
NO!
LONG RESPONSE TIMESlide31
Early Response Preparation
Instead, they rapidly prepared their answer as soon as possible, and responded reactively
Predictions
:“Fireman Sam” (CONTENT)
Should
we
go
past
…
[f]?
Fireman Sam is a long word, so there is more time to prepare a response
YES!
SHORT RESPONSE TIMESlide32
How does the system develop?
3 yo and 5 yo leave longer gaps than adultsHowever:Children are often as fast as adultsChildren are no more variable than adultsInstead, children experience occasional “breakdowns”, leading to very long gaps.Slide33
Conclusion
Children and adults were able to take turns quite rapidly without fine-grained timing predictionEarly preparation of simple responses can afford reactive turn-taking strategiesThe building blocks of the turn-taking system are in place by age 3Children leave long gaps, not because they are slow at producing answers, but because their turn-taking system is less stableSlide34
References
Casillas, M., & Frank, M. C. (2013). The development of predictive processes in children’s discourse understanding. In CogSci 2013: The 35th annual meeting of the Cognitive Science Society (pp. 299-304). Cognitive Society.Casillas, M., & Frank, M. C. (2012). Cues to turn boundary prediction in adults and preschoolers. In
SemDial 2012 (SeineDial
) (pp. 61-69). Université Paris-Diderot.De
Ruitter, J. P., Mitterer, H., Enfield, N. J., (2006). Projecting the end of a speaker’s turn: a cognitive cornerstone of conversation,
Language
, 82(3), 515-535
Garrod
, S., & Pickering, M. J. (2015). The use of content and timing to predict turn transitions.
Frontiers in psychology
, 6.
Lammertink, I., Casillas
, M., Benders, T., Post, B., &
Fikkert
, P. (2015). Dutch and English toddlers' use of linguistic cues in predicting upcoming turn transitions.
Frontiers in psychology
,
6
.
Levinson, S. C., &
Torreira
, F. (2015). Timing in turn-taking and its implications for processing models of language.
Frontiers in psychology, 6.Slide35
References
Maygari, L. & De Ruitter, J. P., (2012). Prediction of turn-ends based on anticipation of upcoming words, Frontiers in Psychology, 3(376), 1-9Magyari, L.,
Bastiaansen, M. C., de Ruiter, J. P., & Levinson, S. C. (2014). Early anticipation lies behind the speed of response in conversation.
Journal of cognitive neuroscience, 26(11), 2530-2539.
Riest, C., Jorschick, A. B., & De Ruiter, J. P. (2015). Anticipation in turn-taking: mechanisms and information sources.
Frontiers in psychology
,
6
.
Stivers
, T., Enfield, N. J., Brown, P., Englert, C., Hayashi, M., Heinemann, T., ... & Levinson, S. C. (2009). Universals and cultural variation in turn-taking in conversation. Proceedings of the National Academy of Sciences,
106(26), 10587-10592.Slide36
The development of turn taking
0;3
0;9
1;0
1;6
2;0
3;0
sensitivity to
contingent
vs.
random
exchanges
(Bloom
,
et al.
1987)
decrease in overlaps:
%40
%20
(
Hilbrink
et al., 2015)
5% overlaps
(Garvey &
Berninger
, 1981)
OVERLAPSSlide37
The development of turn taking
0;3
0;9
1;0
1;6
2;0
3;0
s
ee
also
Garvey
&
Berniger
, 1981
Casillas
et al., 2016
900
ms
5
00
ms
The
duration
of gaps
varies
hugely
with the
c
omplexity
of the
response
and
its
predictability
yes/no:
400-500
ms