distribution of tau distances assessing episodic and semantic contributions in serial recall...

1
Distribution of Tau Distances 0 4 8 12 16 20 24 28 32 0 0.2 0.4 0.6 bus 0 4 8 12 16 20 24 28 32 0 0.2 0.4 0.6 morning 0 4 8 12 16 20 24 28 32 0 0.1 0.2 0.3 0.4 w edding 0 4 8 12 16 20 24 28 32 0 0.1 0.2 yogurt 0 4 8 12 16 20 24 28 32 0 0.05 0.1 0.15 0.2 pizza 0 4 8 12 16 20 24 28 32 0 0.1 0.2 0.3 clay P rior E pisodic M em ory C hance Assessing Episodic and Semantic Contributions in Serial Recall Pernille Hemmer, Brent Miller & Mark Steyvers University of California, Irvine A Wisdom of Crowds Analysis Can we reconstruct the original order on the basis of the recalled orders of a group of individuals? How can memory models help us in this reconstruction process? The goal here is to show that the reconstructed order from a group of individuals is better than the recalled order from any particular individual – a wisdom of crowds effect. We developed a variant of the perturbation model (Estes et al., 1972) that can explain a number of memory errors in serial recall. In this model, items are originally encoded in the correct temporal sequence. Errors occur by locally perturbing the temporal encoding. In our variant of the model, we build in individual differences and item differences – assuming that some people have better memories (sigma parameter), and that some items are easier to rank (e.g., because they stand out temporally) Prior knowledge and expectations about events are known to influence episodic memory. Researchers often assume that prior knowledge leads to errors in recall. We investigate recall for the order of sequences of events, focusing on the potential benefit of prior knowledge. First, we assess the prior knowledge that people bring to the memory task. Second, we utilize memory tasks that are ecologically valid. The stimulus materials to be remembered have the same statistical regularities as can be found in the natural environment. The memory task involves a rank-ordering task in which people have to remember the study order of pictures of stereotyped and random event sequences. We expect to show that combining our knowledge of the regularities of the environment with these noisy memory representations improves the overall accuracy in episodic memory Introduction The Tasks Ten images were drawn from 6 different video clips. Three videos depicted stereotyped events sequences (getting up in the morning, a wedding, getting on the school bus) for which people have strong prior expectations. Three videos depicted more random event sequences (yogurt, pizza, claymation) for which people have weak prior expectations. Prior knowledge condition: Participants would order the randomized sequence of images based on their prior expectation and with out having previously viewed the images Memory condition: Participants would first study the correct sequence of images. They would then order the sequence from memory. probability for individual j that the item at the source position i gets perturbed to the destination position k. Original order Recalled order A B C D E F G H D Note: the mean tau expected for random sequences is 22.5 (=10*9/4) Median Tau to Truth Performance was measures using Kendall’s Tau: The number of adjacent pair-wise swaps between recalled and true order. = 1 = 1+1 Ordering by Individual A B E C D True Order A B C D E C D E A B A B C D E = 2 Measuring Performance B us Morning Wedding Pizza Y ogurt Clay 0 5 10 15 20 25 30 P rior N o P rior M edian Tau Study N o study C hance Data for Morning Sequence (see top banner) 1. event1 (1) 2. event2 (2) 3. event3 (3) 4. event4 (4) 5. event5 (5) 6. event6 (6) 7. event7 (7) 8. event8 (8) 9. event9 (9) 10.event10 (10) 0 2 4 6 8 10 12 R =0.982 1 2 3 4 5 6 7 8 9 10 1. event1 (1) 2. event2 (2) 3. event3 (3) 4. event4 (4) 5. event5 (5) 6. event6 (6) 7. event7 (7) 8. event8 (8) 9. event9 (9) 10.event10 (10) O utput P osition 1 2 3 4 5 6 7 8 9 10 O utput P osition W edding (Recall) 0 2 4 6 0 2 4 0 2 4 6 8 10 12 R =0.980 1. event1 (1) 2. event2 (2) 3. event3 (4) 4. event4 (5) 5. event5 (3) 6. event6 (6) 7. event7 (7) 8. event8 (8) 9. event9 (10) 10.event10 (9) Pizza (Recall) 0 5 10 15 20 25 30 R =0.920 1 2 3 4 5 6 7 8 9 10 1. event1 (1) 2. event2 (2) 3. event3 (3) 4. event4 (5) 5. event5 (6) 6. event6 (4) 7. event7 (7) 8. event8 (8) 9. event9 (9) 10.event10 (10) O utput P osition 1 2 3 4 5 6 7 8 9 10 O utput Position C lay (Recall) 0 0.5 1 1.5 2 2.5 0 2 4 6 0 5 10 15 20 25 R =0.924 Inferred Model Parameters Example Random Sequences Example Stereotyped Event Sequence Best individual Average Individual Inferred Item Accuracy Calibration of Individuals Participant number (first row) Tau score for each participant (second row) Tau scores indicate high accuracy from guessing with prior knowledge for stereotyped events For random events prior knowledge performance is similar to chance Taus are significantly lower for stereotyped (prior) than for random (no prior) events Taus in the memory task (study) are significantly lower than in the prior knowledge task (no study) Even when participants did not previously study the events, they performed better for all three stereotyped events Semantic and Episodic Contributions prior serial recall A B C D E F G H I J A B C D E F G H I J Data for Pizza Sequence (see bottom banner) prior serial recall Morning Sequence Wedding Sequence Pizza Sequence Clay Sequence Mean 1 10 20 30 0 5 10 15 Individuals Thurstonian M odel P erturbation M odel B orda count Individuals Mean Performance of Individuals and Model(s Note: the Thurstonian and Borda count model are two alternative models that were applied to this data

Post on 19-Dec-2015

218 views

Category:

Documents


3 download

TRANSCRIPT

Page 1: Distribution of Tau Distances Assessing Episodic and Semantic Contributions in Serial Recall Pernille Hemmer, Brent Miller & Mark Steyvers University of

Distribution of Tau Distances

0 4 8 12 16 20 24 28 320

0.2

0.4

0.6bus

0 4 8 12 16 20 24 28 320

0.2

0.4

0.6 morning

0 4 8 12 16 20 24 28 320

0.1

0.2

0.3

0.4 wedding

0 4 8 12 16 20 24 28 320

0.1

0.2

yogurt

0 4 8 12 16 20 24 28 320

0.05

0.1

0.15

0.2 pizza

0 4 8 12 16 20 24 28 320

0.1

0.2

0.3clay

PriorEpisodic MemoryChance

Assessing Episodic and Semantic Contributions in Serial RecallPernille Hemmer, Brent Miller & Mark Steyvers

University of California, Irvine

A Wisdom of Crowds AnalysisCan we reconstruct the original order on the basis of the recalled orders of a group of individuals? How can memory models help us in this reconstruction process? The goal here is to show that the reconstructed order from a group of individuals is better than the recalled order from any particular individual – a wisdom of crowds effect.

We developed a variant of the perturbation model (Estes et al., 1972) that can explain a number of memory errors in serial recall. In this model, items are originally encoded in the correct temporal sequence. Errors occur by locally perturbing the temporal encoding.

In our variant of the model, we build in individual differences and item differences – assuming that some people have better memories (sigma parameter), and that some items are easier to rank (e.g., because they stand out temporally)

Prior knowledge and expectations about events are known to influence episodic memory. Researchers often assume that prior knowledge leads to errors in recall. We investigate recall for the order of sequences of events, focusing on the potential benefit of prior knowledge.

First, we assess the prior knowledge that people bring to the memory task. Second, we utilize memory tasks that are ecologically valid. The stimulus materials to be remembered have the same statistical regularities as can be found in the natural environment. The memory task involves a rank-ordering task in which people have to remember the study order of pictures of stereotyped and random event sequences.

We expect to show that combining our knowledge of the regularities of the environment with these noisy memory representations improves the overall accuracy in episodic memory

Introduction

The TasksTen images were drawn from 6 different video clips.

Three videos depicted stereotyped events sequences (getting up in the morning, a wedding, getting on the school bus) for which people have strong prior expectations.

Three videos depicted more random event sequences (yogurt, pizza, claymation) for which people have weak prior expectations.

Prior knowledge condition: Participants would order the randomized sequence of images based on their prior expectation and with out having previously viewed the images Memory condition: Participants would first study the correct sequence of images. They would then order the sequence from memory.

probability for individual j that the item at the source position i gets perturbed to the destination position k.

Original order

Recalled order

A B C D E F G H

D

Note: the mean tau expected for random sequences is 22.5 (=10*9/4)

Median Tau to Truth

Performance was measures using Kendall’s Tau: The number of adjacent pair-wise swaps between recalled and true order.

= 1

= 1+1Ordering by IndividualA B E C D

True OrderA B C D E

C DEA B

A B C D E

= 2

Measuring Performance

Bus Morning Wedding Pizza Yogurt Clay0

5

10

15

20

25

30

Prior No Prior

Med

ian

Tau

Study

No studyChance

Data for Morning Sequence (see top banner)

1. event1 (1)2. event2 (2)3. event3 (3)4. event4 (4)5. event5 (5)6. event6 (7)7. event7 (6)8. event8 (8)9. event9 (9)

10. event10 (10)

Bus (Recall)

0

2

4

6

8

10

12

R=0.721

1. event1 (1)2. event2 (2)3. event3 (3)4. event4 (4)5. event5 (5)6. event6 (6)7. event7 (7)8. event8 (8)9. event9 (9)

10. event10 (10)

Morning (Recall)

0

2

4

6

8

10

12

R=0.982

1 2 3 4 5 6 7 8 9 10

1. event1 (1)2. event2 (2)3. event3 (3)4. event4 (4)5. event5 (5)6. event6 (6)7. event7 (7)8. event8 (8)9. event9 (9)

10. event10 (10)

Output Position1 2 3 4 5 6 7 8 9 10

Output Position

Wedding (Recall)

0 2 4 6

0 2 40

2

4

6

8

10

12

R=0.980

1. event1 (1)2. event2 (2)3. event3 (3)4. event4 (5)5. event5 (7)6. event6 (4)7. event7 (6)8. event8 (8)9. event9 (9)

10. event10 (10)

Yogurt (Recall)

0

5

10

15

20

25

R=0.885

1. event1 (1)2. event2 (2)3. event3 (4)4. event4 (5)5. event5 (3)6. event6 (6)7. event7 (7)8. event8 (8)

9. event9 (10)10. event10 (9)

Pizza (Recall)

0

5

10

15

20

25

30

R=0.920

1 2 3 4 5 6 7 8 9 10

1. event1 (1)2. event2 (2)3. event3 (3)4. event4 (5)5. event5 (6)6. event6 (4)7. event7 (7)8. event8 (8)9. event9 (9)

10. event10 (10)

Output Position1 2 3 4 5 6 7 8 9 10

Output Position

Clay (Recall)

0 0.5 1 1.5 2 2.5

0 2 4 60

5

10

15

20

25

R=0.924

Inferred Model Parameters

Example Random Sequences

Example Stereotyped Event Sequence

Best individual Average Individual Inferred Item Accuracy Calibration of Individuals

Participant number (first row)

Tau score for each participant (second row)

Tau scores indicate high accuracy from guessing with prior knowledge for stereotyped events

For random events prior knowledge performance is similar to chance

Taus are significantly lower for stereotyped (prior) than for random (no prior) events

Taus in the memory task (study) are significantly lower than in the prior knowledge task (no study)

Even when participants did not previously study the events, they performed better for all three stereotyped events

Semantic and Episodic Contributions

prior serial recall

A B C D E F G H I J

A B C D E F G H I J

Data for Pizza Sequence (see bottom banner)

prior serial recall

Morning Sequence

WeddingSequence

Pizza Sequence

ClaySequence

Me

an

t

1 10 20 300

5

10

15

Individuals

Thurstonian ModelPerturbation ModelBorda countIndividuals

Mean Performance of Individuals and Model(s)

Note: the Thurstonian and Borda count model are two alternative models that were applied to this data