-0.005

-0.003

-0.001

0.001

0.003

0.005

0.00 0.25 0.50 0.75 1.00

sign

al c

hang

e

Percentage of time object color listed first -0.002

-0.001

0

0.001

0.002

0.003

COLOR + SHAPE SHAPE

r = 0.50, p < 0.01

Color+Shape Shape

R

sign

al c

hang

e

-1.5

-1

-0.5

0

0.5

1

1.5

1 2 3 4 5 6 7 8 9

Neu

ral S

imila

rity

(Pea

rson

r, F

ishe

r tra

nsfo

rmed

)

Color similarity rating

Color + Shape Shape

0

0.002

0.004

0.006

Left Fusiform Left Lingual

Color+Shape Shape

!"

Sig

nal c

hang

e

1

2

3

4

5

6

7

8

9

Different color Same color

Sim

ila

rity

ra

tin

g

Object pair

A central principle in feature-based theories of semantic memory is the differential weighting of some features over others [1-5]. Some of these features are diagnostic – they serve to distinguish or otherwise conspicuously differentiate one item from others [6,7]. In determining feature diagnosticity, we argue for a distinction between when a feature is available and needed, and when it is actually used.

Subjects learned one of two novel object sets over the course of four sessions:

1. Allport D (1985) Distributed memory, modular subsystems and dysphasia. In Current Perspectives in Dysphasia, Edinburgh: Churchill Livingstone. 2. Warrington EK, McCarthy RA (1987) Brain: A Journal of Neurology 110: 1273-1296. 3. Tyler LK, Moss HE (2001) TICS 5: 244-252. 4. Patterson K, Nester PJ, Rogers TT (2007) Nat Rev Neuro 8: 976-987. 5. Thompson-Schill SL (2003) Neuropsychologia 41: 280-292. 6. Gonnerman LM, Andersen ES, Devlin JT, Kempler D, Seidenberg MS (1997) Brain Lang 57: 254-279. 7. Devlin JT, Gonnerman LM, Andersen ES, Seidenberg MS (1998) JOCN 10: 77-94. 8. Connolly AC, Gleitman LR, Thompson-Schill SL (2007) PNAS 104: 8241-8246. Shape retrieval task

Thirty-two of these subjects (n = 16 for each group) performed a shape retrieval task while undergoing fMRI, answering yes/no shape questions about the objects. This task was followed by a functional color localizer.

Feature diagnosticity affects semantic representations of novel and common object categories

Nina S. Hsu, Margaret L. Schlichting, and Sharon L. Thompson-Schill Center for Cognitive Neuroscience, University of Pennsylvania

INTRODUCTION

+

+

RESULTS: BEHAVIORAL

REFERENCES

METHODS

CONTACT Nina Hsu: ninash@mail.med.upenn.edu

http://www.psych.upenn.edu/stslab/

D74

DISCUSSION

This work was funded by the NIH (R01 MH070850 awarded to S. L. T.-S.) and by a Ruth L. Kirchstein NRSA Predoctoral Fellowship awarded to N. S. H.

•  Features can vary both in how well we know and use them, and this distinction taps into semantic representations.

•  These results parallel previous work demonstrating differences in

conceptual knowledge for blind versus sighted subjects [8]. •  The neural instantiation of diagnostic features may vary along a

posterior-anterior gradient in ventral temporal cortex. •  Preliminary results suggest a similar neural basis for the interaction

of feature diagnosticity with representations of both novel and common object categories, supporting generalization of our findings.

klarve

1. purple

2. blobby

3. round

4.

thull

whemp

QuickTime™ and a decompressor

are needed to see this picture.

thull

whemp

1 2 3 4 5 6 7 8 9Very dissimilar Very similar

klarve

chulge

Using color as the diagnostic feature, we used a training paradigm to investigate how diagnostic features interact with semantic representations.

Following training, subjects performed a number of behavioral tasks, including adjective generation and pairwise general similarity ratings:

Color is necessary, shape is not sufficient: P (object | shape) = 0.33 P (object | color) = 0.50 P (object | shape AND color) = 1.00

Color is available, shape is sufficient: P (object | shape) = 1.00 P (object | color) = 0.50 P (object | shape AND color) = 1.00

Color perception task

When listing object adjectives, “color+shape” subjects (n = 29) listed color first 88% of the time, whereas “shape” subjects (n = 34) listed color first only 45% of the time (p < 0.001). Notably, the groups demonstrated comparable explicit object color knowledge.

“Color+shape” subjects assigned higher general similarity ratings to same-colored object pairs than did “shape” subjects (p < 0.03). We replicated this result when comparing stimuli shared across both groups.

fulch fulch

whemp whemp voothe voothe yerts yerts dorth dorth

klarve klarve chulge chulge screll screll

thull thull hinch hinch nidge nidge jarmed jarmed

Color is available but not needed in order to distinguish stop signs and yield signs.

Color is necessary in order to distinguish lemons and limes.

COLOR + SHAPE: SHAPE:

If you flipped a KLARVE over, would it stand up straight?

0.00%

20.00%

40.00%

60.00%

80.00%

100.00%

COLOR + SHAPE SHAPE

Co

lor

nam

ing

accu

racy

Condition

0.00%

20.00%

40.00%

60.00%

80.00%

100.00%

First Second

Nam

ing

%

Color adjective order

Color+Shape

Shape

*

!"#"$

Left fusiform gyrus (involved in color perception): greater activation during a shape retrieval task for “color+shape” subjects. Independent color similarity ratings approached significance in predicting neural similarity for “color + shape” subjects (rs = 0.22, p = 0.08), which differed significantly from “shape” subjects (Z = 1.98, p < 0.05).

RESULTS: FMRI R

Left inferior temporal gyrus (identified through exploratory analyses): showed an interaction of training condition and task. Prioritizing color positively correlated with brain activity.

ACKNOWLEDGEMENTS

RESULTS: GENERALIZATION

0.00%

20.00%

40.00%

60.00%

80.00%

100.00%

First Second

Na

min

g %

"Color" adjective order

!"

FV

HHI

We examined the extent to which these results generalized to common object categories via fruits and vegetables (FV, color is diagnostic) and household items (HHI, color is not diagnostic) through parallel behavioral and fMRI tasks.

If you flipped a CARROT over, would it stand up straight?

*

n.s.

Like “color+shape” subjects, those subjects describing FV objects were more likely to list color first. When combining novel and common object categories, we found that prioritizing color positively correlated with left fusiform activation.

Nam

ing

perc

enta

ge

Sim

ilarit

y ra

ting

Nam

ing

perc

enta

ge

For more on the color perception task, please see Poster G101 (Persichetti).

* n.s.

r = 0.28, p < 0.03