Journal of Experimental Psychology: Learning ...

Running head: NATURE OF SEMANTIC INTERFERENCE 1

Long-term Interference at the Semantic Level:

2

Evidence from Blocked-Cyclic Picture Matching

3 4 Tao Wei and Tatiana T. Schnur*

5 6 7 8 9 10 11 12

: y g o l o h c . sy n P o i l t a i t n n g e o m eri , and C p x E f ry o o l m a e n r ss M u e , r o g J P in n I n r a Le Rice University

13 14

Please address correspondence to:

15

Tatiana T. Schnur, Ph.D.

16

Rice University

17

Department of Psychology – MS 25

18

P.O. Box 1892

19

Houston, Texas 77251-1892

20

E-mail: [email protected]

21

Phone: +1 713 348 5054

22

Fax: +1 713 348 5221

23

2 NATURE OF SEMANTIC INTERFERENCE 1 2

Abstract Processing semantically related stimuli creates interference across various domains of

3

cognition, including language and memory. In this study, we identify the locus and mechanism of

4

interference when retrieving meanings associated with words and pictures. Subjects matched a

5

probe stimulus (e.g., cat) to its associated target picture (e.g., yarn) from an array of unrelated

6

pictures. Across trials, probes were either semantically related or unrelated. To test the locus of

7

interference, we presented probes as either words or pictures. If semantic interference occurs at

8 9 10 11 12

: y g o l interference should o h the stage common to both tasks, i.e., access to semantic representations, then c . sy n P o i l t a i t n interference effects n occur in both probe presentation modalities. Results showed clearo semantic g e m eri , and C p x independent of presentation modality frequency, confirming a semantic locus of E and lexical y f r o o nal To, testMtheem r ssinterference, we repeated trials across u e interference ino comprehension. mechanism of r g J P in n I n r a e and manipulated the number of unrelated intervening trials (zero vs. four presentationL cycles

13

two). We found that semantic interference was additive across cycles and survived two

14

intervening trials, demonstrating interference to be long-lasting as opposed to short-lived.

15

However, interference was smaller with zero vs. two intervening trials, which we interpret to

16

suggest that short-lived facilitation counteracted the long-lived interference. We propose that

17

retrieving meanings associated with words/pictures from the same semantic category yields both

18

interference due to long-lasting changes in connection strength between semantic representations

19

(i.e., incremental learning) and facilitation caused by short-lived residual activation.

20 21

Keywords: comprehension, interference, facilitation, semantics, incremental learning

3 NATURE OF SEMANTIC INTERFERENCE 1

How quickly we understand, name, or remember a word or picture depends on its

2

relationship with items recently or concurrently processed. For example, episodic memory

3

retrieval is impaired by the previous retrieval of a semantically related memory (i.e., retrieval-

4

induced forgetting; e.g., Anderson, Bjork, & Bjork, 1994). Similarly in the language domain,

5

naming a picture (e.g., CAT) is slowed down when simultaneously reading or hearing a

6

semantically related vs. unrelated word (e.g., dog vs. ball; Lupker, 1979), and both production

7

and comprehension of words is slowed after previously naming or understanding semantically

8 9 10 11 12

: y g o l & Cole-Virtue, o h related words (e.g., Campanella & Shallice, 2011; Howard, Nickels, Coltheart, c . sy n P o i l t a i t subserving n n 2006). In this study, we identify the locus and mechanism interference when g e o m C i r d e n pprocessing) a x , retrieving the meanings (i.e., semantic associated with a series of semantically E y f r o o emprocessing nalBecause, semantic r sissshared by various cognitive processes M u e related picturesoor words. r g J P in n I n r a (e.g., memory retrieval, Le language production, and object recognition), studying how interference

13

arises during semantic processing provides insight into the constraints which may govern

14

semantic interference effects across different cognitive domains.

15

People are generally slower and/or more error prone when matching a probe word

16

(spoken or written, e.g., cat) to its corresponding target picture (e.g., CAT) in the presence of

17

semantically related (e.g., MONKEY, RABBIT, DOG) vs. unrelated distractor pictures (e.g.,

18

TABLE, BOAT, APPLE; Biegler, Crowther, & Martin, 2008; Campanella & Shallice, 2011;

19

Jefferies, Baker, Doran, & Lambon Ralph, 2007; Warrington & Cipolotti, 1996). This direct

20

word-picture matching task was originally designed to test the nature of semantic deficits as a

21

result of brain damage (e.g., Warrington & McCarthy, 1983, 1987). Individuals with aphasia

22

commit more errors and show longer response times (RTs) in the semantically related vs.

23

unrelated condition in this word-picture matching task (e.g., Campanella, Crescentini, Mussoni,


& Skrap, 2013; Forde & Humphreys, 1995, 1997, 2007; Gardner, Lambon Ralph, Dodds, Jones,

2

Ehsan, & Jefferies, 2012; Jefferies et al., 2007; Warrington & Cipolotti, 1996; Warrington &

3

McCarthy, 1983, 1987). It has been generally assumed that semantic interference in direct word-

4

picture matching occurs within the semantic system for a couple reasons. First, semantic

5

interference effects occurred in patients with severely impaired output (i.e., patients unable to

6

produce any speech; Forde & Humphreys, 1995, 1997; Warrington & Cipolotti, 1996). The

7

assumption here has been that patients who were unable to produce speech no longer had access

8 9 10 11 12

: y g o l a deficit in lexical o h to the lexical and/or phonological representations for speech production (i.e., c . sy n P o i l t a i tat this stage oofgprocessing. n n processing). Thus, interference was not likely arising Second, these e m C i r d n pe (e.g., a x , patients exhibited intact presemantic processing intact visual and auditory input processing E y f r o o m al1997; Warrington e n s , and thus exaggerated interference r s1996) M u e cf. Forde & Humphreys, & Cipolotti, , r g Jo P in n I n r a e due to deficits at this level of processing. Together, this pattern of results effects were not L likely

13

suggested that the interference effect arises within the semantic system instead of outside of it.

14

Campanella and Shallice (2011) replicated the semantic interference effects in healthy subjects.

15

They proposed that in healthy subjects too, semantic interference occurred within the semantic

16

system because “no activation of postsemantic lexical representations is needed in order to

17

perform the word to picture matching tasks” (p. 427) and “Since the error pattern that we found

18

was analogous to that occurring in patients…it is proposed that the conditions necessary and

19

sufficient for producing this type of behaviour in normal subjects also, relate to processes within

20

the semantic system itself” (p. 427). Thus, for both aphasic and healthy subjects, semantic

21

interference in the direct word-picture matching task has been assumed to reflect interference

22

within the semantic system (e.g., Campanella & Shallice, 2011; Jefferies et al., 2007; cf. Biegler

23

et al., 2008).


Some evidence suggests that interference in direct word-picture matching arises due to

2

brief changes in the residual activation of representations in the semantic system (e.g.,

3

Campanella & Shallice, 2011; Jefferies et al., 2007). According to the residual activation account,

4

if the current trial proceeds before the activation of semantic representations in prior trials decays

5

to baseline, the residual activation of prior trials will increase the competition among

6

representations associated with stimuli in the current trial, particularly when probes of prior trials

7

are semantically related with the current trial (e.g., Campanella & Shallice, 2011; Warrington &

8 9 10 11 12

: y g o lbetween semantic o h Cipolotti, 1996). As a result, it is more difficult to resolve the competition c . sy n P o i l t a i t the probe inogthenrelated vs. unrelated n representations to retrieve the meaning associatedewith m eri , and C p x condition. Assuming residual activation (~2 sec, e.g., Dell, 1988), then semantic E is short-lived y f r o o m al by increasing e n s r sresponse-stimulus-interval M u e interference should be reduced either r the (RSI) or , g Jo P n i n I n(lags). In fact, Campanella r a e intervening unrelated trials and Shallice (2011) found that healthy L

13

subjects made more errors matching probe words to pictures across semantically related trials

14

when the RSI was zero vs. one second, suggesting that the interference effect is short-lived. In

15

addition, subjects were prone to falsely select the picture corresponding to the probe word from

16

the preceding trial (i.e., perseverative errors), which suggests that the short-lived interference

17

effect is due to residual activation of representations from the preceding trial. Thus, Campanella

18

and Shallice concluded that semantic interference in direct word-picture matching is the result of

19

competition between related semantic representations due to residual activation of the preceding

20

trial.

21

However, several pieces of evidence temper the conclusion that in healthy subjects,

22

interference occurs at the level of semantic processing as the result of residual activation. First,

23

semantic interference in the word-picture matching task may occur because picture arrays are


named rather than as a result of processing the meanings associated with the probe words

2

(Biegler et al., 2008). Biegler et al. (2008) found a significant correlation between direct word-

3

picture matching RTs and the number of alternative names for the target picture in the picture

4

array. Because the number of alternative names is an index of lexical processing, the correlation

5

suggests that direct word-picture matching involves naming of the target pictures. This

6

speculation is consistent with similar semantic interference effects in naming. Subjects take

7

longer to name pictures when grouped in semantically related vs. unrelated condition (e.g.,

8 9 10 11 12 13

: y g o olsuggests that this h Damian, Vigliocco, & Levelt, 2001; Kroll & Stewart, 1994) andy evidence c . s n P o i l t a i t Belke, 2013; n et al., 2001). n interference occurs at a lexical, not semantic levele(e.g., Damian g o m eri , and C p x Together these results raise thef possibility interference in direct word-picture E thatosemantic y r o em arising nal ,byM r ssat a postsemantic lexical not semantic u e matching is at o least in part caused interference r g J P in n I n r a level of processing. Le Second, the residual activation account is challenged by the findings that semantic

14

interference in comprehension is not always reduced by manipulations of time intervals but

15

instead accumulates over long time intervals. If interference is the result of residual changes in

16

activation, this makes the prediction that it should decay with intervening time and should not

17

survive beyond several trials. Consistent with this prediction, Campanella and Shallice (2011,

18

Experiment 2) demonstrated that when RSI was short (0 ms) vs. long (1 sec) subjects made

19

quantitatively more errors directly matching a probe word to a picture. Biegler and colleagues

20

(2008, Experiment 2B) also found that when the same set of materials were repeatedly matched

21

interference did not accumulate across repetitions. Both these results suggest that interference

22

did not accumulate over long time intervals. However, in contrast to Campanella and Shallice,

23

Biegler et al. found that the manipulation of RSI (1 vs. 4 sec.) did not change the magnitude of


semantic interference. Similarly, in contrast to Biegler et al., Campanella and Shallice found that

2

interference increased across repetitions, suggesting that semantic interference lasts over several

3

trials. It is possible that the interference effect in comprehension is due to multiple mechanisms,

4

but it is unclear why the same kind of measurement of interference effects showed different

5

patterns across different studies. One possibility which may have contributed to this discrepancy

6

is that in the related vs. unrelated condition, not only are probe words from successive trials

7

semantically related, but also the pictures in the array on a single trial are semantically related to

8 9 10 11 12

: y g o lresult, it is unknown o h each other (e.g., Biegler et al. 2008; Campanella & Shallice, 2011). As a c . sy n P o i l t a i t nfrom a prior trial (when n whether the interference which temporally differed in these studies was g e o m eri , and C p x probe words across trials are semantically to each other, as when naming semantically E related y f r o o emet al., 2006; nal e.g.,, Howard s Schwartz, Brecher, & Hodgson, r sSchnur, M u e related picturesoone-at-a-time, r g J P in n I n r a e trial. This makes it difficult to draw conclusions concerning the source 2006) or from the Lcurrent

13

of interference. In sum, the experimental design of previous word-picture matching tasks does

14

not allow us to draw conclusions about the locus (i.e., at the semantic vs. lexical level) and

15

mechanism (i.e., temporal residual activation vs. long-lasting changes) of semantic interference

16

in comprehension, the focus of the current study.

17

To address the locus of semantic interference, we tested subjects on an associate

18

matching paradigm where we presented probes as either words or pictures. We developed the

19

associate matching paradigm based on Biegler et al.’s (2008) Exp. 3, where subjects saw a probe

20

word and selected the most associated target picture from an array of distractor pictures (e.g.,

21

probe: cat, array: YARN, HONEY, BAMBOO and BANANA; a contrast to the direct word-

22

picture matching task used in previous studies, e.g., Campanella & Shallice, 2011; Warrington &

23

Shallice, 1983, 1985). Critically, by using an associate match, correct responses cannot be


generated by directly comparing the probe name with the picture names (e.g., cat to YARN), as

2

can be done in direct word-picture matching (e.g., cat to CAT; cf. Biegler et al., 2008). Instead,

3

to select the correct associated target, subjects retrieve the semantic representations associated

4

with the probe, target and distractor pictures (Biegler et al., 2008; Forde & Humphreys, 1997).

5

Unlike direct word-picture matching, associate word-picture matching RTs do not correlate with

6

lexical variables of target pictures (e.g., lexical frequency or name agreement) suggesting that

7 8 9 10 11 12

: y g o las words (associate o h we manipulated the modality of the probe presentation, displaying probes c . sy n P o i l t a i t matching). nWe hypothesized that if n word-picture matching) or pictures (associate picture-picture g e o m eri , and C p x semantic interference occurs in both occurs during a process shared by both tasks E tasks, it likely y f r o o m al e n r ss rather than from non-shared aspects M u e (access to semantic representations from words or pictures) , r g Jo P in n I n r a of these tasks (e.g., Leword vs. picture recognition input processing; for a similar rationale to

this task reduces the likelihood of a naming strategy by subjects (Biegler et al., 2008). In addition,

13

identify language deficits at a semantic vs. presemantic locus, see Forde & Humphreys, 1995,

14

1997; Gardner et al., 2012). If semantic interference only appears when the probes are words

15

(associate word-picture matching) vs. pictures (associate picture-picture matching), this would

16

suggest that semantic interference occurs during access to the lexical representations or the links

17

between lexical and semantic representations 1. By contrast, if semantic interference only appears

18

in associate picture-picture matching, this would suggest that interference occurs somewhere

19

between access to the structural descriptions associated with pictures (e.g., object recognition

1

In language production, it has been proposed that semantic interference occurs during access to lexical representations or the links between semantic and lexical representations (cf. Damian et al., 2001; Damian & Als, 2005; Humphreys, Lloyd-Jones, & Fias, 1995; Oppenheim et al. 2010; Vitkovitch & Humphreys, 1991). If we assume that language production and comprehension share semantic and lexical representations as well as their links (e.g., Levelt, Meyer, & Roelofs, 1999; Rogers et al., 2004), then it is possible that semantic interference occurs in comprehension at the same locus. Thus, this generates the prediction that semantic interference will occur during access to semantic representations from lexical representations (i.e., during associate word-picture matching) but not during access to semantic representations from structural descriptions (e.g., object recognition input system, cf. Forde & Humphreys, 1997) associated with pictures (i.e., during associate picture-picture matching).


input system, cf. Forde & Humphreys, 1997) and access to the pictures’ associated meanings.

2

Lastly, although null results could occur for a number of reasons, if we find no semantic

3

interference in either associate word-picture or picture-picture matching, one possible

4

explanation would be that semantic interference only occurs during access to lexical

5

representations from semantics (i.e., during naming; Belke, 2013; Damian et al. 2001).

6 7 8 9 10 11 12

To address the mechanism of semantic interference, we modified Biegler et al.’s (2008)

: y g o l semantically related owere h trials but not within trials. In the related condition, probes acrossytrials c . s n P o i l t a i twere unrelated neach other (e.g., trial 1, n to each other but within a trial the pictures in the e array to g o m eri , and C p x probe: CAT, array: YARN, BAMBOO, and BANANA; trial 2, probe: PANDA, array: E HONEY, y f r o o m al and BANANA). e n r ss we rule out the possibility that M u e YARN, BAMBOO, HONEY, In this way, , r g Jo P in n I n r a semantic interference Le is caused by processing the picture array within a trial because the array

associate matching task in the following ways. First, we manipulated semantic relatedness across

13

pictures were unrelated to one another in both the related and unrelated conditions. Second, it is

14

unlikely that semantic interference will arise from a difficulty in selecting the associated target

15

picture because we presented the same probe-target pairs in both the related and unrelated

16

conditions. Moreover, we also controlled for the association strength between the probe and

17

distractor pictures in the related and unrelated condition (see Materials for details) and thus it is

18

unlikely that semantic interference will arise from a difficulty in rejecting the distractors in the

19

related vs. unrelated condition. With this design we can establish whether retrieving the

20

meanings associated with probes across trials generates semantic interference (cf. Biegler et al.,

21

2008; Campanella & Shallice, 2011).

22 23

Critically, manipulating semantic relatedness across (but not within) trials allows us to test whether residual activation of probe representations from previous trials affects access to the


meanings associated with the probe on the current trial. Specifically, to test the residual

2

activation account we examined the longevity of semantic interference by manipulating the

3

number of semantically unrelated intervening trials (0 vs. 2 lags). If semantic interference is due

4

to temporary residual activation (Campanella & Shallice, 2011), then increasing unrelated

5

intervening trials should reduce semantic interference. Alternatively, if semantic interference is

6

not caused by temporary residual activation but instead some long-lasting mechanism, unrelated

7

intervening trials should not reduce semantic interference. Third, we presented the same set of

8 9 10 11 12 13

: y g o l should not extend o h materials four times (i.e., cycles). If semantic interference is short-lived, it c . sy n P o i l t a i t from earlier ntrials in one cycle should n beyond the duration of a cycle (~6s) because interference g e o m eri , and C p x not affect trials in a subsequent cycle. the prediction that interference should E Thus, thisorgenerates y f o m al are presented e n r ssdifferent cycles. By contrast, if M u e not change when materials repeatedly across , r g Jo P in n I n r a semantic interference Le is long-lasting, semantic interference should accumulate across cycles (cf. Oppenheim, Dell, & Schwartz, 2010; Schnur et al., 2006).

14 15 16

Method Subjects We recruited 48 native English speakers from Rice University, where 24 subjects

17

participated in associate word-picture and picture-picture matching tasks respectively. All

18

subjects participated for course credit and gave informed consent in accordance with the

19

protocol approved by the Institutional Review Board of Rice University.

20

Materials

21

We selected eight probes from each of eight categories (i.e., four-legged animals, tools,

22

appliances, furniture, people, birds, vehicles and body parts), for a total of 64 probes. In order for

23

subjects to not see the same probe-targets repeated across lag conditions (see Design), we


organized the probes in two groups (Group1 and Group2), so that each group consisted of eight

2

categories of four exemplars each (see Appendix). Filler probes used as intervening lag items

3

were an additional 68 objects not belonging to the eight experimental probe categories, where

4

four items were practice materials. We selected pictures from the Bank of Standardized Stimuli

5

(Brodeur, Dionne-Dostie, Montreuil, & Lepage, 2010) and various internet sources. Based on the

6

Nelson, McEvoy and Schreiber (2004) association norm database and informal ratings, we paired

7 8 9 10 11 12

: y g o lthe same group o h associated target), which did not share association with other probes from c . sy n P o i l t a i t 15 subjects nrated the degree of n (semantically related or unrelated). Moreover, ane additional g o m eri , and C p x association between the experimental associated targets and un-paired distractor E probes and y f r o o l m e nascale s r M u e associates on ao5-point (1= not associated at all, 5s = very highly associated). The probes , r g J P in n I n r a and associated targets Le were rated as strongly associated (mean: 4.60, range: 3.67-5.00) where the each of the132 probes (experimental and filler) with a categorically unrelated associate (i.e.,

13

association ratings for the probe-target pairs were significantly higher than for the probe-

14

distractors in both the semantically related (mean: 1.18; t1(14) = 46.83, p < .001; t2(63) = 65.43,

15

p < .001) and unrelated groups (mean: 1.14; t1(14) = 56.04, p < .001; t2(63) = 72.20, p < .001).

16

Further, the association ratings for probe-distractors did not significantly differ between the

17

related and unrelated groups (t1(14) = 1.71, p = .11; t2(63) = 1.26, p = .21).

18

Design

19

The design was the same for both associate matching tasks, except that probes were

20

presented as words in associate word-picture matching and as pictures in associate picture-

21

picture matching. In each trial, subjects saw a probe (e.g., cat) and chose from a 4-picture array

22

the correct associated target (e.g., YARN). The 4-picture array on each trial was comprised of the

23

correct associated target to the probe and three distractors not associated with the probe but the


correct associates for other probes within the same block (See Figure 1). Probes across trials

2

were either from the same category (semantically related block, e.g., PANDA, CAT, BEAR, and

3

MONKEY) or different categories (unrelated block, e.g., NOSE, CAT, BLENDER, and

4

MOTORCYCLE) in the same group. Subjects saw the same probe paired with the same

5

associated target in related and unrelated blocks. Each block consisted of four different probes

6

repeating four times (cycles) with the same 4-picture array in different arrangements. In the lag2

7

condition, experimental probes were interleaved with two unrelated trials, whereas in the lag0

8 9 10 11 12

: y g o ol h condition experimental probes immediately followed each other.yHalf of the subjects c . saw probes s n P o i l itcondition, and the ta n n from Group1 presented in lag0 condition and probes from Group2oinglag2 e C im r d e n p a x , subject completed 33 blocks (one practice, other half of subjects saw the reverse. E Together,oreach y f o m etotal nal blocks), for r ss 512 experimental and 512 M e 16 related and o 16uunrelated a of 16 rpractice, g J P in n I n r a intervening unrelated Le trials. The order of lag condition was counterbalanced across subjects and

13

the order of blocks within each lag condition remained the same. Thus, semantic relatedness

14

(related vs. unrelated), cycle (1 – 4) and lag (0 vs. 2 unrelated intervening trials) were within-

15

subject and within-item factors. Task (probes presented as words or pictures) was a between-

16

subject and within-item factor.

17

Apparatus and procedure

18

We used the DMDX program (Forster & Forster, 2003) to present stimuli and collect RTs

19

and accuracy. Using the standard protocol in blocked-cyclic paradigms to reduce trial error (e.g.,

20

Damian & Als, 2005; Schnur et al., 2006), before the experiment began subjects saw all 132

21

word-picture or picture-picture pairs (depending on the task), presented in random order.

22

Association pairs stayed on the screen until the subjects recognized the association and pressed

23

the spacebar to initiate the next trial. Subjects then participated in a practice session with 16 trials.


Trials were presented in the same way in both practice and experimental sessions. Each trial

2

began with a fixation point (+) in the center of the screen for 500 ms, followed by a probe and a

3

4-picture array. The probe appeared in the center and four pictures were 350 pixels (above, below,

4

left and right) from the center (See Figure 1). Subjects placed their right hand on the number pad

5

and used the arrow keys corresponding to picture location (top, bottom, left and right) to indicate

6

the correct response. After subjects made a response or a 1500 ms deadline was reached, a

7

fixation was presented for 500 ms, followed by another probe along with the same 4-picture

8 9 10 11 12

: y g o l probes were o h array in a different arrangement. This procedure was repeated until all four c . sy n P o i l t a i ton to the nextogblock. n All pictures were n presented four times in a block. Then subjects moved e m eri , and C p x scaled to fit a 250 × 250 pixelfsquare. word-picture matching, the probe words were E In associate y r o o m al New Roman e n r ss lasted ~45 minutes including M u e presented in 20-point Times font. The experiment , r g Jo P in n I n r a three breaks spaced Leat equal intervals.

13

Results

14

Following Campanella and Shallice (2011), we included in the error analyses and

15

excluded from RT analyses trials with incorrect associate target selection or no response within

16

1500 ms. As a result, we removed 10% of the data points in the RT analyses. We computed

17

ANOVAs on RTs and error rates with subjects (F1) and items (F2) as random variables. Fixed

18

within-subject and within-item variables included relatedness (related vs. unrelated), cycle (1-4)

19

and lag (0 vs. 2). We treated task (associate word-picture, picture-picture matching) as a fixed

20

between-subject and within-item variable.

21

See Table 1 for an overview of mean RTs and error rates per condition. When analyzing

22

RTs, as shown in Figure 2A, subjects were slower to make a response in the semantically related

23

than unrelated condition (F1(1, 46) = 274.44, MSE = 2211, p < .001; F2(1, 63) = 53. 90, MSE =


31386, p < .001), which was not modulated by task (F1(1, 46) = 1.17, MSE = 2211, p = .29; F2(1,

2

63) = 1.67, MSE = 4966, p = .20). This interference effect increased with two vs. zero unrelated

3

intervening trials (F1(1, 46) = 12.47, MSE = 2491, p < .01; F2(1, 63) = 14.77, MSE = 6464, p

4

< .001), independent of task (F’s < 1). The interference effect also increased across cycles (see

5

Figure 2B), as shown by a significant linear component of the relatedness by cycle interaction

6

(F1(1, 46) = 67.51, MSE = 1016, p < .001; F2 (1, 63) = 28.22, MSE = 5231, p < .001). This

7

increasing interference effect was larger in the associate picture-picture than word-picture

8 9 10 11 12

: y g o l = 3684, p < .05). oMSE h matching task (F (1, 46) = 8.29, MSE = 1016, p < .01; F (1, 63)y =c 5.92, . s n P o i l t a i t n increased across n Unrelated intervening trials did not affect the degree to which interference g e o m eri , and C p x cycles (F (1, 46) = 1.43, MSEf= E 1324, p = .24; Fry (1, 63) = 2.71, MSE = 5549, p = .11) and this o o m e nal different r s(1,s 46) = 1.14, MSE = 1324, p = .29; M u e pattern was notosignificantly between tasksr(F , g J P in n I n r a e = 3437, p = .24). MSE F (1, 63) = 1.40,L 1

1

2

2

1

2

13

Error analyses revealed that subjects made significantly more errors in the related (13%)

14

than unrelated (7 %) condition (F1(1, 46) = 96.46, MSE = .007, p < .001; F2(1,63) = 63.05, MSE

15

= .03, p < .001), which was larger in the Lag2 (∆ = 8%) than Lag0 condition (∆ = 4 %; F1(1, 46)

16

= 25.97, MSE = .003, p < .001; F2(1,63) = 27.18, MSE = .007, p < .001). No other effects related

17

to semantic interference were significant in the error analyses (p’s > .21).

18

General Discussion

19

In this study we designed a new paradigm to test the locus and mechanism of semantic

20

interference in comprehension. Subjects selected from an array of unrelated pictures, a picture

21

semantically associated with a probe word or picture. Probes were either semantically related or

22

unrelated across trials. To address the locus of interference (semantic vs. lexical), we tested

23

whether semantic interference occurred independently of probe presentation modality (word vs.


picture) and minimized naming by using an associate matching (e.g. YARN to cat) task. To

2

address the mechanism of semantic interference in comprehension (short-lived residual

3

activation vs. other long-lasting mechanisms), we manipulated semantic relatedness across trials

4

(as opposed to within trial). We tested whether interference was reduced by unrelated intervening

5

trials, and whether interference accumulated across cycles. We found that whether probes were

6

words or pictures, subjects took longer to match the probe with the correct associated picture

7

when probes of successive trials were semantically related vs. unrelated. Interference increased

8 9 10 11 12

: y g o l vs. lag0). Below, o h across cycles and was greater with increased unrelated intervening trials (lag2 c . sy n P o i l t a i t n of interference in n we discuss how these results provide insight into e the locus and mechanism g o m eri , and C p x comprehension and the implications models involving semantic processing. E for computational y f r o o m al e n r ss M u e A semantic locus of interference in comprehension , r g Jo P in n I n r a We designed Leour associate matching task to test the hypothesis that semantic interference

13

in comprehension occurs during access to the semantic system. Associate matching involves

14

several cognitive steps, including presemantic input processing of the probe and picture array

15

(e.g., visual processing of pictures and recognition of words), activation of associated semantic

16

information, rejection of the distractors and selection of the correct associate to the probe. It is

17

unlikely that the longer RTs in the related vs. unrelated condition were due to a difference in

18

difficulty between selecting the correct associated targets in the related vs. unrelated condition as

19

we used the same probe-target pairs in the related and unrelated conditions. Second, interference

20

is unlikely to be due to a difficulty in rejecting the distractors in the related vs. unrelated

21

condition because there was no difference in association strength between the probe and

22

distractors between conditions. Third, semantic interference was the same magnitude whether the

23

probe was a word or picture, which suggests that semantic interference occurred during a stage


shared by word/picture comprehension, i.e., the retrieval of the semantic representations

2

associated with probes, not the stages which differ between word and picture comprehension

3

(e.g., input processing). Fourth, even though naming was not required in the associate matching

4

task (cf. Biegler et al. 2008), we can rule out the possibility that we observed semantic

5

interference because subjects named the probes while performing the matching tasks. We

6

correlated the 64 experimental probes’ log-transformed lexical frequencies (CELEX, Baayen,

7

Piepenbrock, & Gulikers, 1998) with their corresponding RTs, collapsing across subjects, tasks,

8 9 10 11 12

: y g o ol h lags, relatedness, and cycles. We found no relationship betweeny the log-transformed c . lexical s n P o i l it ta n n frequencies and mean probe RTs (r = .16, p = .21). Thus, these results localize semantic g e o C im r d e n p a x , within the semantic system consistent interference in associate word/picture-picture matching E y f r o o m alaccounts (e.g., e n s r M u e with other semantic locus Campanella &sShallice, 2011; Jefferies et al., 2007). , r g Jo P in n I n r a However,L ae semantic locus account of interference is inconsistent with studies suggesting

13

that semantic interference only arises at a postsemantic level, during the process of naming (e.g.,

14

Belke, 2013; Damian et al., 2001). Belke (2013) and Damian and colleagues (2001) among

15

others (e.g., Howard et al., 2006; Schnur et al., 2006) assume that semantic interference requires

16

lexical processing because semantic interference is consistently observed in naming but not in

17

judgment tasks designed to tap semantic processing. We propose that we found interference in

18

our semantic tasks whereas Damian et al. (2001) and Belke (2013) did not because the semantic

19

judgment tasks they used tap a different kind of semantic processing compared to that in

20

associate matching. In judgment tasks such as judging the direction in which an object is faced

21

(Damian et al., 2001), or whether an object is natural vs. man-made (Belke, 2013), subjects do

22

not need to identify specific semantic information to make a correct response. For example, if the

23

probe is a “cat”, subjects can make a correct decision concerning whether the object is natural vs.


man-made merely based on retrieving the semantic information shared by animals, but subjects

2

do not need to specify the difference between a cat and other animals. Thus, repeatedly retrieving

3

the same set of semantic representations across semantically related vs. unrelated trials in this

4

task should facilitate semantic judgments, exactly the effects observed by Belke (2013) and

5

Damian et al. (2001) 2. In contrast, in associate matching more detailed semantic information

6

especially that which distinguishes different concepts is required for the retrieval of a correct

7 8 9 10 11 12

: y g o lThus, we argue that the o h subjects’ subsequent comprehension of semantically related probe items. c . sy n P o i l t a i treflects interference n during semantic n semantic interference in associate matching seen e here g o m eri , and C p x processing, rather than duringfother associated with the task. E potential processes y r o o m al in comprehension: e n r ss interference and short-lived M u e Temporally distinct effects Long-lasting , r g Jo P in n I n r a facilitation Le response (e.g., the probe is a cat rather than a dog), which as discussed below, slows down

13

We observed two temporally distinct effects during the retrieval of semantic

14

representations associated with words and pictures. First, semantic interference was not

15

eliminated but increased with unrelated intervening trials. That interference survived across

16

unrelated intervening trials suggests that interference is long-lasting. Second, interference was

17

reduced with no intervening trials. We interpret the reduction of interference at the lag0

18

condition as caused by short-lived facilitation based on a large body of evidence from priming

19

and picture naming paradigms. In word priming paradigms, semantic facilitation effects are

20

reduced with one unrelated intervening trial between semantically related prime and target pairs

21

(e.g., Joordens & Besner, 1992). In naming, semantic interference is more evident when

22

semantically related pictures are separated by two vs. zero unrelated trials (e.g., Vitkovitch,

23

Rutter, & Read, 2001; Wheeldon & Monsell, 1994). The short-lived semantic facilitation effect 2

Damian et al. (2001, Exp. 1) found an 8 ms facilitatory, but non-significant semantic judgment effect.


is commonly attributed to spreading activation in the semantic system (e.g., Damian & Als,

2

2005). That long-lasting interference and short-lived facilitation both occur during

3

comprehension is further suggested by the finding of semantic interference accumulating across

4

cycles. If interference is long-lasting, semantic interference should accumulate over several trials

5

and cycles. By contrast, because the facilitation effect is due to temporary residual activation, it

6

should not accumulate across trials and cycles. That is, the effect of semantic relatedness across

7

cycles reflects the build-up of long-lasting interference only, without the contamination of the

8 9 10 11 12 13

: y g o oislpredicted, and unlike h facilitation effect. Thus, an increasing interference effect across y cycles c . s n P o i l t a i tinterferenceoeffect n will not be affected by n the main effect of semantic relatedness, the increasing g e m eri , and C p x unrelated intervening trials. Our results demonstrated both these patterns. Together, these results E y f r o o m al e n r ss M u e suggest both semantic interference and facilitation in comprehension, where facilitation but not , r g Jo P n i n ntemporary residualI activation. r a e interference is the result of L However, that semantic interference was not reduced but increased by unrelated

14

intervening trials is inconsistent with Campanella and Shallice’s (2011) and Biegler et al.’s (2008)

15

results concerning manipulations of time in direct word-picture matching. Because Campanella

16

and Shallice only assessed the effect of RSI on matching performance in the related condition

17

(short vs. long RSI created lower accuracy), it is unknown whether RSI also affects matching

18

performance in the unrelated condition, which, if that were the case, would have yielded no

19

effect of RSI on the semantic interference effect. As such, a comparison between results is

20

uninterpretable. That Biegler et al. saw no change in semantic interference with different RSIs

21

may be explained by the time intervals manipulated. Biegler et al. manipulated RSI as 1000 vs.

22

4000 ms, while here we manipulated zero vs. two intervening unrelated trials, corresponding to

23

500 vs. ~ 2800 ms RSI. It is possible that semantic facilitation was largely attenuated by the


1000 ms RSI and thus no semantic facilitation was observed in Biegler et al.’s study. In sum, our

2

study suggests two temporally distinct effects occurring at the level of semantic processing

3

during comprehension – long-lasting interference and short-lived facilitation.

4

Implications for computational models of semantic processing

5

These results have implications for models of cognitive processes which involve

6

semantics. First, that semantic interference occurs at the level of semantic processing in

7

comprehension challenges the single postsemantic lexical locus of semantic interference in

8 9 10 11 12

: y g o l oOppenheim h computational models of speech production (e.g., Howard et al.,y 2006; c . et al., 2010). s n P o i l itin semantic blockedta task asoisgused n n We used the same experimental design in a comprehension e C im r d e n p a x , but instead there were no postsemantic cyclic naming, engaging similar semantic processing E y f r o o m alidentify specific e n r ss to make a correct response in M u e lexical requirements (i.e., semanticrinformation , g Jo P in n I n r a the absence of naming). Le Even without postsemantic lexical processing, we found similar long-

13

lasting semantic interference in terms of the main effect of semantic relatedness, increasing

14

interference across cycles, and increased interference at lag2 vs. lag0 as seen in naming (e.g.,

15

Belke, 2013; Damian et al., 2001; Damian & Als, 2005; Navarrete, Del Prato, & Mahon, 2012;

16

Schnur et al., 2006). Assuming that comprehension and naming share the same semantic system

17

(e.g., Levelt, Meyer, & Roelofs, 1999; Rogers et al., 2004), the similar results in semantic

18

blocked-cyclic comprehension and naming suggest that interference does not occur exclusively

19

during lexical access in naming, but also during the retrieval of semantic representations. Thus,

20

computational models should take these findings into account when simulating semantic

21

interference during speech production.

22 23

Second, because semantic interference showed long-lasting effects, this suggests that semantic memory is dynamic, continually changing (i.e., learning) from relevant but not


irrelevant experience. Thus, a learning mechanism which adjusts connection weights between

2

representations based on experience should be applied to semantic processing as has been done

3

for lexical processing (Howard et al., 2006; Oppenheim et al., 2010). For example, within the

4

framework of the computational model of semantic memory proposed by Rogers and colleagues

5

(2004), the meaning of a stimulus is represented in a network, where semantic features across

6

different sensory/motor modalities are bound by amodal semantic hubs. An incremental learning

7

mechanism could account for long-lasting semantic interference in the following ways. After

8 9 10 11 12

: y g o ol between its h activating the meaning associated with a probe (e.g., DOG), the y connections c . s n P o i l t a i tstrengthened.oBecause n stimuli from the n semantic features and an amodal semantic hub(s)eare g m eri , and C p x same semantic category share fsome the probe from the same category (e.g., E semantic features, y r o o m atriall will activate e n r sshub while also erroneously priming the M u e CAT) on a subsequent its own semantic , r g Jo P in n I n r a previously activated Lesemantic hub (i.e., “DOG”) via the strengthened connection. At this point, if

13

these semantically related representations compete with each other for retrieval, then it will take

14

longer to resolve the competition (see Howard et al., 2006 for a similar discussion in speech

15

production). Another way to account for long-lasting semantic interference with incremental

16

learning is to assume that after activating the meaning of a probe (e.g., DOG), the connections

17

between its semantic features and an amodal semantic hub are strengthened, while connections

18

between semantic features of the probe and the amodal semantic hubs of other exemplars from

19

the same category (e.g., CAT) are weakened. As a result, if the probe in the next trial is a

20

semantically related stimulus, the weakened connections will result in reduced activation of its

21

amodal semantic representation and thus a slower response time (see Oppenheim et al., 2010 for

22

a similar discussion in speech production). Intervening unrelated trials have little effect on

23

semantic interference because connection weight change only occurs for representations which


share semantic features, and does not occur (or occurs to a lesser degree) for representations

2

which share few semantic features. Incremental learning has been assumed to reflect continual

3

adjustments of the cognitive system to its environment (e.g. Gupta & Cohen, 2002) and has been

4

widely used to explain long-lasting effects, such as repetition priming (e.g., Monsell, Matthews,

5

& Miller, 1992), syntactic priming (e.g., Bock & Griffin, 2000), and form-based priming (e.g.,

6

Bowers, Damian, & Havelka, 2002). Future research should examine how different

7 8 9 10 11 12

: y g o l o h semantic processing. c . sy n P o i l t a i t n n g Conclusion e o m eri , and C p x We addressed the locusf and E mechanismoofryinterference during comprehension by o nal , Mem r ss matching tasks. The results u e implementing o new associate word-picture and picture-picture r g J P in n I n r a provide strong evidence Le that semantic interference in comprehension occurs during access to implementations of incremental learning can best account for interference at the level of

13

semantic representations rather than during lexical access. In addition, we found that interference

14

increased across presentation cycles and with unrelated intervening trials, indicating that two

15

different mechanisms occur at the level of semantic processing – long-lasting interference and

16

short-lived facilitation. Furthermore, that interference occurs at the level of semantic processing

17

challenges the single lexical locus of semantic interference in computational models of speech

18

production. We see the next important steps in this line of inquiry are to explore the degree to

19

which semantic interference in naming arises at a lexical vs. semantic level of the production

20

system, and to understand how cognitive models can be implemented to account for the

21

relatively temporary facilitation and persistent interference effects at the level of semantic

22

processing.

23


Acknowledgments

2

We presented these results at the 55th Annual Meeting of the Psychonomic Society in

3

Long Beach (2014). We thank Ja Young Choi for helping collect data, and Randi Martin and

4

three anonymous reviewers for their helpful comments.

: y g o l o h c . sy n P o i l t a i t n n g e o m eri , and C p x E f ry o o l m a e n r ss M u e , r o g J P in n I n r a Le

23 NATURE OF SEMANTIC INTERFERENCE References Anderson, M. C., Bjork, R. A., & Bjork, E. L. (1994). Remembering can cause forgetting: retrieval dynamics in long-term memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20(5), 1063-1087. Baayen, R. H., Piepenbrock, R., & Gulikers, L. (1998). The CELEX Lexical Database. Philadelphia, PA: Linguistic Data Consortium, University of Pennsylvania.

: y g o l o h necessarily conceptually mediated: Implications for models of lexical-semantic c y . encoding. s n P o i l it ta n n Journal of Memory and Language, 69(3),e 228-256. g o C im r d e n p a x , Consequences of an inhibition deficit for Biegler, K. A., Crowther, J. E.,f &E Martin, R. C. (2008). y r o o l m aand e n r ss the semantic blocking paradigm. M u e word production comprehension: evidence from , r g Jo P in n I n r a e CognitiveL Neuropsychology, 25(4), 493-527. Belke, E. (2013). Long-lasting inhibitory semantic context effects on object naming are

Bock, K., & Griffin, Z. M. (2000). The persistence of structural priming: Transient activation or implicit learning?. Journal of Experimental Psychology: General, 129(2), 177-192. Bowers, J. S., Damian, M. F., & Havelka, J. (2002). Can distributed orthographic knowledge support word-specific long-term priming? Apparently so. Journal of Memory and Language, 46(1), 24-38. Brodeur, M. B., Dionne-Dostie, E., Montreuil, T., & Lepage, M. (2010). The bank of standardized stimuli (BOSS), a new set of 480 normative photos of objects to be used as visual stimuli in cognitive research. PloS ONE, 5(5), e10773. Campanella, F., & Shallice, T. (2011). Refractoriness and the healthy brain: A behavioural study on semantic access. Cognition, 118(3), 417-431.

24 NATURE OF SEMANTIC INTERFERENCE Damian, M. F., & Als, L. C. (2005). Long-lasting semantic context effects in the spoken production of object names. Journal of Experimental Psychology: Learning, Memory, and Cognition, 31(6), 1372-1384. Damian, M. F., Vigliocco, V., & Levelt, W. J. M. (2001). Effects of semantic context in the naming of pictures and words. Cognition, 81, B77-B86. Dell, G. S. (1988). The retrieval of phonological forms in production: Tests of predictions from a

: y g o l on semantic oaphasia: h Forde, E., & Humphreys, G. W. (1995). Refractory semantics inyglobal c . s n P o i l t a i t in neuropsychology. n Memory, 3, 265-307. n organisation and the access–storage distinction g e o m eri , and C p x Forde, E. & Humphreys, G. W. (1997). locus for refractory behaviour: Implications E A semantic y f r o o aldistinctionsMandem n s memory. Cognitive r ssemantic u e for access storage the naturerof , g Jo P in n I n r a Neuropsychology, Le 14, 367-402. connectionist model. Journal of Memory and Language, 27, 124-142.

Forster, K. I., & Forster, J. C. (2003). DMDX: A Windows display program with millisecond accuracy. Behavior Research Methods, Instruments, and Computers, 35(1), 116-124. Gardner, H. E., Lambon Ralph, M. A., Dodds, N., Jones, T., Ehsan, S., & Jefferies, E. (2012). The differential contributions of pFC and temporo-parietal cortex to multimodal semantic control: exploring refractory effects in semantic aphasia. Journal of cognitive neuroscience, 24, 778-793. Gupta, P., & Cohen, N. J. (2002). Theoretical and computational analysis of skill learning, repetition priming, and procedural memory. Psychological Review, 109(2), 401-448. Howard, D., Nickels, L., Coltheart, M., & Cole-Virtue, J. (2006). Cumulative semantic inhibition in picture naming: Experimental and computational studies. Cognition, 100, 464-482.

25 NATURE OF SEMANTIC INTERFERENCE Jefferies, E., Baker, S. S., Doran, M., & Lambon Ralph, M. A. (2007). Refractory effects in stroke aphasia: A consequence of poor semantic control. Neuropsychologia, 45(5), 10651079. Joordens, S., & Besner, D. (1992). Priming effects that span an intervening unrelated word: implications for models of memory representation and retrieval. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18(3), 483-491.

: y g o olrepresentations. Journal h Evidence for asymmetric connections between bilingual y memory c . s n P o i l t a i t n of Memory and Language, 33, 149-174. en g o m eri , and C p x Levelt, W. J. M., Roelofs, A., f& E Meyer, A. S. (1999). A theory of lexical access in speech y r o o m al and Brain e n s r s1-38. M u e production. Behavioral Sciences, 22, , r g Jo P in n I n r a Lupker, S. J. (1979). LeThe semantic nature of response competition in the picture-word Kroll, J. F., & Stewart, E. (1994). Category interference in translation and picture naming:

interference task. Memory & Cognition, 7(6), 485-495. Monsell, S., Matthews, G. H., & Miller, D. C. (1992). Repetition of lexicalization across languages: A further test of the locus of priming. The Quarterly Journal of Experimental Psychology, 44(4), 763-783. Navarrete, E., Del Prato, P., & Mahon, B. Z. (2012). Factors determining semantic facilitation and interference in the cyclic naming paradigm. Frontiers in psychology, 38, 1-15. Nelson, D. L., McEvoy, C. L., & Schreiber, T. A. (2004). The University of South Florida free association, rhyme, and word fragment norms. Behavior Research Methods, Instruments, and Computers, 36, 402-407.

26 NATURE OF SEMANTIC INTERFERENCE Oppenheim, G. M., Dell, G. S., & Schwartz, M. F. (2010). The dark side of incremental learning: A model of cumulative semantic interference during lexical access in speech production. Cognition, 114, 227-252. Rogers, T. T., Lambon Ralph, M. A., Garrard, P., Bozeat, S., McClelland, J. L., Hodges, J. R., & Patterson, K. (2004). Structure and deterioration of semantic memory: a neuropsychological and computational investigation. Psychological Review, 111(1), 205-

: y g o l interference during o h Schnur, T. T., Schwartz, M. F., Brecher, A., & Hodgson, C. (2006). Semantic c . sy n P o i l t a i t Journal ofoMemory n and Language, 54(2), n blocked-cyclic naming: Evidence from aphasia. g e m eri , and C p x 199-227. E f ry o o l m a e in serialrverbal nStudies of interference r ss reactions. Journal of Experimental M u e Stroop, J. R. (1935). , o g J P in n I n r a Psychology, Le18(6), 643-662. 235.

Vitkovitch, M., Rutter, C., & Read, A. (2001). Inhibitory effects during object name retrieval: The effect of interval between prime and target on picture naming responses. British Journal of Psychology, 92(3), 483-506. Warrington, E. K., & Cipolotti, L. (1996). Word comprehension The distinction between refractory and storage impairments. Brain, 119(2), 611-625. Warrington, E. K., & McCarthy, R. (1983). Category specific access dysphasia. Brain, 106(4), 859-878. Wheeldon, L. R., & Monsell, S. (1994). Inhibition of spoken word production by priming a semantic competitor. Journal of Memory and Language, 33(3), 332-356.

27 NATURE OF SEMANTIC INTERFERENCE Appendix

Probes

Group1

ANIMALS

VEHICLES

APPLIANCES

BODY

BIRDS

PEOPLE

TOOLS

FURNITURE

Rabbit

Car

Vacuum

Eye

Dove

Waiter

Corkscrew

Bed

Whisk

Cupboard

Scissors

Chair

Peeler

Dresser

Wrench

Crib

Saw

Hatrack

Shovel

Bench

Knife

Shelf

Cow Dog Squirrel Group2

Bear Cat Panda Monkey

Associated

Group1

Targets

Group2

gy: Airplane Iron Foot HummingbirdoloCowboy h c y . s n P o Bicycle Refrigerator Mouth Duck Photographer i l t i ta n n g e o C Boat Toaster erimNose Owl Teacher d n p a x , Emachine orLegy f Truck Washing Seagull Police o l m a e n s , M n Ear res Parrot our ingStove Referee JMotorcycle P I n r Blender Finger Turkey Chef LCabea Train

TV

Neck

Eagle

Doctor

Carrot

Gas station

Carpet

Glasses

Olive branch

Menu

Wine

Comforter

Milk

Railway

Antenna

Scarf

American flag

Stethoscope

Egg

Tableware

Bone

Sky

Crease

Shoes

Flower

Wagon

Paper

Cushion

Acorn

Bell

Ice

Lipstick

Pond

Film

Potato

Clothes

Honey

Anchor

Bread

Tissues

Night

Chalk

Screw nut

Baby

Yarn

Boxes

Laundry basket

Pants

Sea

Handcuffs

Wood

Coat

Bamboo

Helmet

Flame

Headphone

Cage

Basketball

Soil

Park

Banana

Hail

Fruits

Ring

Thanksgiving

Spatula

Chopping board

Books

28 NATURE OF SEMANTIC INTERFERENCE Figure Captions

Figure 1. Example of sequential trials in associate picture-picture matching. Presentation times (ms) on left. Probes are outlined in red, associated target and distractors are outlined in black (n.b., in associate word-picture matching, words replaced pictures as probes). In the related condition, probes within a block are from the same category (Panel A). In the unrelated condition,

: y g o l o h c . sy n P o i l t a i t n (A) and increasing n Figure 2. Semantic interference in terms of the main effect of relatedness g e o m eri , and C p x interference across cycles (B) fas E a function of relatedness and lag in associate word-picture and y r o o em within-subject nalError bars, M r ss 95% confidence intervals. u e picture-pictureo matching. represent r g J P in n I n r a Le probes are from different categories (Panel B).

29 NATURE OF SEMANTIC INTERFERENCE Figure 1. A.


30 NATURE OF SEMANTIC INTERFERENCE

B.


31 NATURE OF SEMANTIC INTERFERENCE Figure 2. A.

RT (ms)

1000

900 Related

: y g o l o h c . sy n P o i l t a i t 700 n n g e o m Lag0 Lag2 dC eriLag0 , anLag2 p x E Word-Picture Picture-Picture f ry o o l m a e n r ss M u e , r o g J P in n I n r a Le 800

B.

Unrelated

Difference (Related - Unrelated)

160

120 Cycle 1

80

Cycle 2 Cycle 3 40

Cycle 4

0

-40

Lag0

Lag2

Word-Picture

Word-Picture

Lag0

Lag2

Picture-Picture Picture-Picture

32 NATURE OF SEMANTIC INTERFERENCE Table 1. Mean response times (in milliseconds; calculated from participant means, with standard deviations in parentheses) and error rates (as percentage means) for different conditions (task, lag, relatedness, and cycle).

Task Associate word-picture matching

Associate picture-picture matching

Lag0 Lag2 Lag0 Lag2

y: g o l o4 h Cycle 1 Cycle Cycle 1 c y .10 s n P o i l Related 887 (65) 902 (70) t ta 849og(82)ni n e Unrelated 852 (62) 5 m C i r d e Related 980 (80) xp 15 an 979 (74) , E y f r o(68) emo 7 Unrelated 916 858 (68) l a n s s 919 (99) Related 16 , M904 (73)n P893 re(86) our 930in(75) g J Unrelated 9 earn932 (82) 856 (84)I 846 (81) 845 (86) L Related 971 (71) 911 (85) 913 (89) 937 (96) 19 Response time (SD) Cycle 2 Cycle 3 863 (68) 884 (65) 815 (61) 837 (73) 932 (88) 943 (74) 854 (75) 866 (65)

Unrelated

937 (83)

859 (93)

853 (96)

829 (90)

10

Error rate (%) Cycle 2 Cycle 3 Cycle 4 8 9 10 5 6 7 17 15 16 8 7 6 10 6

11 8

11 7

13 7

16 7

14 8

Journal of Experimental Psychology: Learning ...

Journal of Experimental Psychology: Learning ...

Suggest Documents