BALDWIN DARE Infant Looking Time Events 2010
<p class="MsoNormal" >DARE BALDWIN
Dec 3 2010
Dwelling on ACTION Univ of Oregon</p>
<p>
psych prof at Univ of Oregon</p>
<p>
<b>Bridgette Martin Hard</b> is now back at stfd, a former student</p>
<p>
a FriSem..</p>
<p>
Meredith Meyer is also here today (Univ of Michigan)</p>
<p>
<b>What info do infants track as events unfold</b></p>
<p>
<b>how do infants draw causal inferences</b></p>
<p>
how do causal inferences alter infants unfolding processing</p>
<p>
existing looking-time methodologies are only suited for qualitative info</p>
<p>
<b>eye-tracking wud seem valuable ... but hard to do for rapidly unfolding events</b>.</p>
<p>
dwell-time paradigm seems a potential candidate</p>
<p>
<b>dwell time is what they use.</b></p>
<p>
*
Bridgette now speaks (Dare gave the intro)</p>
<p>
her advisor was Barbara Tversky</p>
<p>
measure moment by moment changes in attention across an intential action sequence
monitor how attn changes with event structure</p>
<p>
eg organizing dorm room (making bed, folding laudry, straigtening up desk
or putting on make-up or
assembling somethng from Ikea.</p>
<p>
extract still frames every 1 second... self-paced viewing measure <b>looking time LT</b></p>
<p>
how to segment the event... what is the level of granularity.</p>
<p>
ask each participant individually to hit space bar when action begins or ends (break points)</p>
<p>
Hard, Recchia, and Tversky under review</p>
<p>
looking times LTs fall off within units (follows a power function decay: detrend the data)</p>
<p>
what is the hierarchical structure of the entire event:
reliably dwelling on breakpoints;</p>
<p>
fine brkpoints have small LT; med have med LT and coarse grained brk points have long LT.</p>
<p>
eating brkfst... randomly organized;</p>
<p>
cleaning has v clear hierarch order;</p>
<p>
how many actions could they report for each event;</p>
<p>
how does attn go up as people transition from goal or intention to another goal or intention
(indicates a change in what they're trying to accomplish)</p>
<p>
are position changes characteristic of breakpoints... big changes in position</p>
<p>
pass image thru convolution filter
and then do position change claculation</p>
<p>
dwell-time effect... predictable incr in attn at brkpoints</p>
<p>
human speechome project (deb roy) ceiling mount video image</p>
<p>
magnitude of a position change predicts both brkpts and dwelltime increases.</p>
<p>
Qs: 1) is position change solely responible for incr dwell-time.
2) can dwell-time effect be observed for segmental structure that is not grounded in position changeds.
3) can dwell-time effects INDEX newly acquired event segments.</p>
<p>
grasp knife to chop food then grasp fridge then open fridge</p>
<p>
makes an analogy with speech acts and grouping sentence or word fragments.</p>
<p>
statistical structure of event sequence.</p>
<p>
Baldwin, Andersson, Saffron, and Meyer 2007
exposure corpus: blow-inser-drink</p>
<p>
participants first view video of exposure corpus</p>
<p>
they then extract still frames extracted from exposure corpus at 3 per sec.</p>
<p>
then S watches the stills self-paced.</p>
<p>
Bridg expects that they will get incr in dwelltime whnever a new event seq begins.</p>
<p>
blow-insert-drink vs scrub-inspect-wipe etc.</p>
<p>
result: reliable dwell-time incr at beginning of each segment.</p>
<p>
people who were exposed to the corpus video first had obvious effect:
first motion effect had bigger than second had bigger than third
(these were detrended by removing fatigue/ boredom effect)</p>
<p>
it is correlated with high recognition accuracy:
ie for those Ss who recognized the events, it was really true
ïe first motion event had highest mean looking time score and thereafter, a monotonic decrease
(even after detrending it by normalizing with decay function)</p>
<p>
heightened attn does help people encode more;</p>
<p>
Dare's comment: system is designed for efficient collection of info;
so u deploy it most when looking is most important (at beginnigns of segments)</p>
<p>
end of Bridge's segment
*
now Dare begins again talking abt kids</p>
<p>
Bridge worked with Dare as apost-doc in Oregon.</p>
<p>
tried it with 3 and 4 year olds
hoped that kids wud be able to do it like kids do it;
use kid-freindly video
with two hierarchical levels of action rep (fine and coarse)
then sldies extracted at one-second increments 88 sldies total
kids advance at their own pace with touch of a mouse.
dwell times (looking times were records (not just their touch of the mouse)
cuz kids get distracted and look away alot.</p>
<p>
doing tower of hanoi</p>
<p>
coarse-grain finsihed stacking a partic pole.</p>
<p>
she shows a video of child doing task: rapt attn:</p>
<p>
saw same effect as adults</p>
<p>
small dwelltime within units med time for fine bndryies; and long time for major (coarse) bndries.</p>
<p>
kids with hihgh memory scores did show strong linear trend (incr dwell time with incr coarseness of boundaries.</p>
<p>
can we do dwell time in infants</p>
<p>
did it with 6 mont old
dwell time in infants sitting in moms lap</p>
<p>
need to measure their looking time; infants look all over the place:</p>
<p>
19 slides total; strict alternation of break pts and within.</p>
<p>
meredith is the actress; she rech sor a cup and then stacks it;</p>
<p>
10 month old doing task</p>
<p>
lose abt half of data to inattentiveness
but do show effect (ie more dwell time at breakpoints vs within-segment)
it actually was significant.</p>
<p>
perhaps dwell is also sensitive to other aspectsof infants event processing</p>
<p>
how abt social cues or pedagogy
or changes due to incr knowledge or familiarity or
changes due to apprehension of causal structure.</p>
<p>
she has shown that if kids see cane-pulling working, then they
are better at understanding it.</p>
<p>
task: have 10 to 13 month old infants attemt cane pulling in a basseline action task</p>
<p>
next an experiemtner domonstrrates can-pulling
a) with accompanying social/pedagogical cues
b) ina a non-social manner (hands only; no vocalizations)</p>
<p>
next infants page thru slide-show versions of the possible and impossibel events</p>
<p>
(an impossible event wud be cane not in hand and cane flipped around so it cannot grasp the object.</p>
<p>
next infants page thhru sldie shows of both possibloe and impossible events in alternation
next infants attemtp cane-pulling themselves again in a second action task.</p>
<p>
Q1: infants dwells do incr at breakpoints: grasping the cane, pulling, retrieving the toy.</p>
<p>
Q2 des dwell time correlate with infants subsequent performance.</p>
<p>
in pedagogical case (with prior teaching) kids really show peaks at dwell times;
but in contrast, during the nonsocial condition, only dwell when the cane is being pulled
(ie they're only responsive to the motion of pulling (shallow processing rather than being sensitive to
the actual task segments ie they didn't "get it")</p>
<p>
dwell times seem to become more systematic with repetiion (ie they LEARN btwn run-thru 1 and 2)
ie they can deploy their attn on ru-thru 2.</p>
<p>
she shows the impossible case clip: pulling of toy occurs despite the fact that the cane is not around it...
it just pulls in by magic.
the baby really attend (big prolonged dweltime (ie they startle when impossibel pulling of the toy occurs)</p>
<p>
infants dwell times DO seem to reflect sensitivity to causally relevant aspects of the evnt
(ie cuz they know that pulling in the toy requries the cane, they are shocked when it moves in on its own)</p>
<p>
how do adults deploy their attention? hard to eval</p>
<p>
conclusions: dwell times are useful for capturing reorg of event processing as observers accrue knowledge of the events
they observe across time.
</p>