on the nature of prototype effects in visual working memory ... - Brandeis

0 downloads 0 Views 881KB Size Report
their memory representa ons of s muli held in visual working memory. 2) Subjects showed larger prototype effects immediately following a shift when the.
R E I NN

Background  

•  SRmulus  strength  was  varied,  by  using  RDCs  whose  moRon  Coherence  was  low  (50%)   or  high  (90%).  

Analyses:     •  Errors  were  sign-­‐corrected  (“Prototype  Error”):  Errors  toward  the  mean  of  a   distribuRon  were  posiRve,  away  were  negaRve.     •  Analyzed  Prototype  Errors  made  to  RDCs  whose  direcRons  fell  between  the  means   of  the  Pre  and  Post  Shic  distribuRons.   •  We  defined  two  subclasses  of  Prototype  Errors:  Local  errors  were  signed  relaRve  to   the  current  distribuRon’s  mean  direcRon;  Previous  errors  were  signed  relaRve  to   the  previous  distribuRon’s  mean.  

5

6

7

8

9

10

11

12

13

14

20

Local Previous

0

Post Shift

Pre Shift

Quadrant A 1

2

3

4

5

6

Stimulus Set Presentation Block

Timepoint

Post Shift

Quadrant B 7

8

9

10

11

12

13

14

15

16

Timepoint

The  range  of  each  uniform  distribuRon  was  doubled,  to  60o  wide.  DistribuRons  were  shiced  by  a  larger  amount:  20°.    

•  •  This  produces  less  similar  direcRons  of  moRon,  both    within  and  between  distribuRons.   Post Shift

3)    Subjects’  reliance  on  the  prototypical  sRmulus  was  modulated  by  both  sRmulus   strength  (coherence)  and  similarity:  coherence  effects  were  larger  when  direcRons  in   Post  Shic  were  more  similar  to  the  established  prototype.    This  represents  a  omission   in  Wilken  and  Ma  (2004)’s  hypothesis  about  the  adapRve  nature  of  prototype  effects.   4)    These  results  are  not  easily  explained  by  models  assuming  equal  weight  is  given  to   all  prior  direcRons  in  compuRng  a  prototype  (Morgan,  Watamaniuk,  &  McKee,  2000).    

for ID= 26.00 120

Mean = - 3 . 2 9 Std. Dev. = 65.7 N = 319

•  Our  results  may  thus  be  consistent        with  recent  dual-­‐process  models  of        working  memory  that  include  a          uniform  distribuRon  of  guesses        (Zhang  &  Luck,  2009).  

80

60

40

Post Shift

Pre Shift

•  DistribuRons  of  errors  for  individual        subjects  were  posiRvely  kurtoRc,          approximaRng  Von  Mises  distribuRons.  

Histogram

100

Experiment  2:  Low  Similarity  

Pre Shift

2)    Subjects  showed  larger  prototype  effects  immediately  following  a  shic  when  the   new  direcRons  of  moRon  were  more  similar  to  the  prototype  established  before  the   shic.    This  suggests  inter-­‐item  similarity  is  an  important  variable  controlling  reliance  on   prototypes.  

•  Explore  our  results’  implicaRons  for  models  of  visual  working  memory   Pre Shift

16

1)    Subjects  relied  more  on  the  prototypical  direcRon  when  RDC  coherence  was  low.     This  supports  the  main  contenRons  of  Wilken  and  Ma  (2004),  and  Huang  and  Sekuler   (2010a,b):  the  prototype  effect  allows  subjects  to  compensate  for  imperfecRons  in   their  memory  representaRons  of  sRmuli  held  in  visual  working  memory.  

•  Examine  strength  and  similarity  effects  on  prototype  reliance  in  domains  other  than   direcRon  of  moRon.    Extend  a  computaRonal  model  (under  development)  to  these   domains.  

-40

15

   

Future  DirecRons  

-20

20 0 -20

4

 

•  However,  guesses  that  are  biased  by          the  protoype  may  produce  a          non-­‐uniform  distribuRon.    This  suggests        a  potenRal  modificaRon  to  the  model.    

20

0 -200.00

-100.00

.00

100.00

200.00

error

Time

                             

N = 27

Local Previous

Pre Shift

Post Shift

Pre Shift

Quadrant A 1

2

3

4

5

6

Post Shift

Quadrant B 7

8

9

10

Timepoint

11

12

13

14

40

•  Prototype  effect  established  in  the  Pre  Shic  condiRon  (p

Suggest Documents