Sie sind auf Seite 1von 12

Journal of Experimental Psychology: Learning, Memory, and Cognition 1990, Vol. 16, No.

5, 760-771

Copyright 1990 by the American Psychological Association, Inc. O278-7393/9O/SO0.75

Controlling Stroop Interference: Evidence From a Bilingual Task


Joseph Tzelgov, Avishai Henik, and David Leiser
Department of Behavioral Sciences Ben Gurion University, Beer Sheva, Israel

Skilled performance is often associated with automaticity. Automatic processes are generally thought of as uncontrollable so that automaticity implies the lack of control. The Stroop colornaming task is one of the most cited examples of automaticity and uncontrollability of word reading. This task is also employed extensively to investigate the structure of the bilingual lexicon. The present work employed a Hebrew-Arabic bilingual Stroop task in two separate experiments. To induce controlled processing, we varied the subjects' expectations regarding the written (irrelevant) color words. The Stroop interference effect was always present but subjects were able to control (reduce) it in their native language but not in their second language. In addition, the presumed structure of the bilingual lexicon seems to change in accordance with proficiency in the second language. It is suggested that automaticity and control are both characteristics of skilled performance with each of them reflecting a different aspect and each subserving a different function of our cognitive system.

The present study addressed two related issues: a) automaticity and control, and b) the structure of the bilingual knowledge representation. To this end we employed a Stroop task in a bilingual situation. Automaticity and Control Most theorists argue that automatic processes are effortless, unconscious, and involuntary (e.g., Hasher & Zacks, 1979; Posner, 1978; Shiffrin & Schneider, 1977). The involuntary nature of automaticity is reflected in several properties: automatic processes occur without intention; when activated they run to completion; their operation cannot be fully suppressed, and their products are hard to ignore (Hasher & Zacks, 1979; Posner & Snyder, 1975; Shiffrin & Schneider, 1977; Zbrodoff & Logan, 1986). This implies that automatic processes are uncontrollable. Indeed, automaticity and control are suggested as two opposing concepts (Schneider, Dumais & Shiffrin, 1984). The characteristics of automaticity have been reserved for highly skilled performance, the origin of which is usually related to heredity or practice (Hasher & Zacks, 1979; Shiffrin & Schneider, 1977). According to Hasher and Zacks (1979), large amounts of practice result in the development of automatic processes, which in turn function to prevent subcomponents of complex skills from overWe are grateful to Osnat Milman and Limor Kaplan for coordinating the project and to Jacqueline Berger for her comments. We also thank Gordon Logan, Colin MacLeod and three anonymous reviewers for their valuable comments and suggestions. They are not to be blamed for any shortcomings of this article. Earlier versions of this article were presented at the 21st Scientific Meeting of the Israeli Psychological Association in March 1987 and the 6th Australian Language and Speech Conference in August 1988. Correspondence concerning this article should be addressed to Joseph Tzelgov, Department of Behavioral Sciences, Ben-Gurion University of the Negev, Beer-Sheva, Israel 84105.
760

loading our limited-capacity system. The subcomponents of skilled performance, which become automatic with practice, serve the skilled performer in being executed effortlessly and without the need to be monitored closely as performance proceeds. The equivalence between proficiency and automaticity is also implied in studies of the bilinguals' lexicon. For example, Magiste (1984) suggested that the amount of interference from a language is directly related to experience in that language. In contrast, it has often been suggested that skill involves control (see Logan, 1985, for a discussion of this point). We usually think about skill or mastery as manifested in ease of punto manipulation, or in the ability to accommodate to situational para colocar en demands and constraints. According to this point of view, skill would also be evidenced in an ability to withhold certain H1 processes. This has been demonstrated with skilled typists (Logan, 1982) and adult speakers (Levelt, 1983). Considering these conclusions together reveals that current theorizing on skill and automaticity is inconsistent. The development of a skill involves both automaticity and control. As skill develops, the performer learns to control the components of the skill. However, at the same time various processes become automatized and less controlled to the point of being executed involuntarily or autonomously (Zbrodoff & Logan, 1986). Logan (1985) has recently discussed this issue and suggested that the general belief that automaticity and control are opposites may be overstated. Posner (1978) suggested one solution when discussing the reflexive nature of eye movements: he suggested that the eye movement system might be viewed as one that requires minimal attention but which can be brought under conscious control if so desired. However, this solution cannot apply to processes that are carried out automatically and at the same time show a certain level of control. One of the most cited examples for automaticity and uncontrollability is the Stroop effect (Stroop, 1935). In this situation, subjects are asked to name the ink color of a word

BILINGUAL STROOP CONTROL

761

and refrain from reading the word itself (e.g., when presented with the word GREEN written in red they are supposed to answer "red"). Ink color naming is faster when the two dimensions of the stimulus are congruent (e.g., GREEN^ written in green) than when they are incongruent (e.g., GREEN written in red). That is, the irrelevant, to-be-ignored dimension interferes with processing the relevant dimension. Word reading is initiated without intention and in spite of the subjects' attempts to suppress it. This pattern of results is suggested as an example of the automaticity of reading. Accordingly, the Stroop effect should increase as reading experience increases. In the Stroop paradigm, control means the ability to inhibit the irrelevant reading process. Developmental studies show that the effect increases and then decreases as age and reading experience increase (Schiller, 1966; Schadler & Thissen, 1981). The decrease of the effect is not predicted by the notion of automaticity. However, it does fit rather nicely with the suggestion that improvement in skill implies control. Thus, it may be that the Stroop effect evidenced in the performance of children shows automaticity and control simultaneously: automaticity is shown by the color name interference, whereas control is suggested by the reduction of this effect with the increase in reading experience. There are some hints in the literature that subjects can modulate the activation of concepts (e.g., Neely, 1977). Some experiments investigating control in Stroop situations have been reported (Logan, 1980; Logan & Zbrodoff, 1979). However, the evidence is equivocal. While Stroop interference seems to be sensitive to expectations when two colors are employed, it is not affected by expectations when the number of colors increases (Logan, Zbrodoff & Williamson, 1984). Our study attempts to unravel effects of controlled processing in a bilingual Stroop situation. The Bilingual Lexicon Collins and Loftus (1975) suggested a two-level model of knowledge representation. They distinguished between semantic and lexical levels of representation. Concepts are represented at the semantic level, usually referred to as the "semantic network," whereas their names are represented at the lexical level. We will refer to this level of representation as the internal "dictionary".1 The Stroop effect implies automatic activation of the color name in the semantic network. The bilingual Stroop task has been employed extensively in studies aimed at investigating the structure of the bilingual's knowledge representation. These research efforts have documented the existence of a between-language Stroop interference (e.g., Dalrymple-Alford, 1968; Dyer, 1971; Fang, Tzeng, & Alva, 1981; Kiyak, 1982; Preston & Lambert, 1969). Most investigators report that the within-language interference is larger than the between-language interference (Dyer, 1971; Kiyak, 1982; Preston & Lambert, 1969, but see Ehri & Ryan, 1980). Recent studies have indicated that the between- as well as the withinlanguage interference is determined by language proficiency (Magiste, 1984, 1986). The difference between the within-language and the between-language effect suggests that the internal lexicon is

organized in terms of language-specific units. Several authors have discussed such a structure of the lexicon (Kirsner, Smith, Lockhart, King, & Jain, 1984; Potter, So, Von Eckardt, & Feldman, 1984). According to this approach, words from the same language are closer to each other in the internal lexicon than words from different languages. Bilingual versions of the basic structure suggested by Collins and Loftus (1975) that assume language-specific "dictionaries" and a common semantic representation have recently been suggested by several investigators (Chen & Leung, 1989; Gerard & Scarborough, 1989; Groot & Nas, 1989; Grainger & Beauvillain, 1989; Tzelgov & Henik, 1989). Potter et al. (1984) discussed two hypotheses about the relations between the language-specific lexicons: 1. Concept mediation hypothesis: There are no direct connections between the parallel words in the different languages. Parallel words (from the different languages) are connected via an amodal conceptual system (i.e., the semantic level of representation). 2. Word association hypothesis: Here, comprehension of a word is always mediated by access to units of the more proficient language. That is, access to and from a word in the less proficient language is always mediated by the activation of the parallel word in the more proficient language. Studies that show a larger within- than between-language interference effects are consistent with the concept mediation hypothesis. When the task involves two languages (i.e., between-language condition), activation has to spread from one area of the lexicon to another, via the amodal system, in order to create an interference effect. In contrast, when the task involves only one language (i.e., within-language condition), there is no need for the activation to spread via the amodal system to create an interference. Studies that present the influence of proficiency on the size of the interference effects support the word association hypothesis. For example, Magiste (1984) reported that the different language interference effect increases with proficiency in a given language. Chen and Ho (1986) investigated the development of Stroop interference in Chinese-English bilinguals. They found that at the first stages of learning the second language (English), the interference created by the first language (Chinese) is larger than the interference created by the second language. This is true for both the different language and same language conditions. However, as experience with the second language increases, the same language interference effect becomes greater than the different language interference effect in that language. Chen and Ho proposed a mixed model of language representation: They suggested that Chinese-English bilinguals use lexical associations to start with, but these associations are gradually replaced by concept mediation links with the increase of the proficiency in the second language. Chen
' Collins and Loftus (1975) refer to the representation of concepts' names as the "lexical network" (see p. 413). In contrast, in the bilingual literature the term "lexicon" is used to describe the internal representation of knowledge in general. We use the term lexicon in a similar way. We use the term "dictionary" (which is also used by Collins & Loftus, 1975) to denote the representation of concept names.

762

J. TZELGOV, A. HENIK, AND D. LEISTER

and Ho's (1986) suggestions imply an interdependence between proficiency and the structure of the lexicon. This structure would change as subjects become more skilled with a new language. We have claimed that the ability to control language processing is reflected in a decline in Stroop interference in developmental studies. It seems that the first stages of language acquisition reflect the development of the automatic component by an increased Stroop interference effect. In the more advanced stages, control is reflected by the decline of Stroop interference. It is possible that these processes, automatization on the one hand and increased control on the other, play a major role in changing the pattern of the between-languages and the within-language Stroop interference. For example, improved ability to control (i.e., inhibit) reading of the first language contributes to a reduction in the between-language effect of the first language. At the same time, the increased automatization in reading the second language contributes to an increase in the interference of the second language. Thus, in an advanced stage of reading proficiency the amount of the interference reflects both automatization and control processes. Most students of the field relate a major role to automatizaion. The present study attempts to unravel the role played by the control processes in bilinguals' reading.

control. When the subjects expect Arabic stimuli (i.e., when 80% of the stimuli are Arabic words) they may be able to inhibit the irrelevant reading process. This would reduce Stroop interference relative to the unexpected condition for Arabic written stimuli. Would this reduction be larger in the between-languages (i.e., response language Hebrew) than in the within-language (i.e., response language Arabic) condition? In the between-languages condition one has to inhibit or attenuate one whole system (i.e., Arabic), whereas in the within-language condition one has to inhibit processes within a system currently employed (i.e., inhibit reading Arabic while responding in Arabic). This pattern is not expected for the second language that the subjects cannot control. Such a pattern of results should be reflected in an interaction involving all four variables. In contrast, a lack of a fourth-order interaction would be consistent with MacNamara and Kushnir's (1971) assertion about independent activation mechanisms for input (stimulus) and for output (response) language processing.

Method
Subjects. Sixteen Arabic-Hebrew bilingual students of Arabic origin participated in the experiment. Arabic was the native language of all the subjects. All of them began to study Hebrew in elementary school and thus studied Hebrew for at least twelve years. They were quite proficient in Hebrew, but knew Arabic better. They spoke only Arabic at home and both Hebrew and Arabic with friends. Outside of the home, they used mainly Hebrew for everyday interaction. Participation in the experiment was in partial fulfillment of course requirements or for payment. Stimuli and apparatus. Five colors were employed in the experiment: red, green, blue, yellow, and black. The names of these colors consist of four-letter words, in both Hebrew and Arabic. There is no phonetic similarity betweeni Hebrew and Arabic for the color names employed. Hebrew and Arabic are both written from right to left, but use different alphabets. The color names were written on white cards. Each word encompassed a rectangle of 30 x 8 mm centered at the middle of the card. The stimuli were presented 80 cm away from the subjects' eyes. The apparatus consisted of a Gerbrand's Harvard 3-field tachistoscope, connected to a voice operated relay and a Monsanto HT 8510 countertimer. Design. There were four experimental sessions which were at least three days apart from each other. In two sessions subjects were asked to respond in Hebrew and in two in Arabic. In one of tht two sessions Hebrew was more frequent (i.e., 80% of the words were written in Hebrew and 20% in Arabic) and in the other Arabic was ; frequent (i.e., 80% of the words were written in Arabic and : in Hebrew). Color-word congruency was manipulated within session; Each session was composed of one block of 50 trials that was presented twice. Forty trials in the block were written in the more frequent language and 10 in the less frequent language. Half of the trials within each block were congruent and half were incongruent. This arrangement resulted in at least 10 trials in each experimental condition. We decided to limit each session to 100 trials to avoid boredom effects, especially in the last two sessions of the experiment. The 20 incongruent stimuli in the frequent language were created by printing each of the five color names in four different color inks. The 20 congruent stimuli were created by printing each of the five color names in its color ink four times. The ten stimuli of the less frequent language consisted of five congruent and five incongruent

Experiment 1
We exposed Arabic-Hebrew bilinguals to a bilingual version of the Stroop task and manipulated their expectations regarding the written language. If the subject has no control, or cannot suppress the reading of the irrelevant color word even when it is expected, Stroop interference would be insensitive to the proportion of words in a given language. If the subject can control his or her reading, Stroop interference in the expected language should diminish relative to the, interference in the same language when unexpected.! However,: it may be that the relations between expectations andStooop interference are more complicated than this. *As suggested above, the notion of skill implies a close relationship between proficiency and control. Such a relationship between skill and control implies that subjects should be better at controlling the better known language. This should contribute to a reduction in Stroop interference in that language when expected as compared to the language unexpected condition, whereas expectation may not affect Stroop interference in the less known language. Native speakers of Arabic who were quite proficient in Hebrew participated in the experiment. In the present context, control can be exerted by inhibition of the irrelevant reading processes. This should be evidenced in a reduction of Stroop interference. Because the subjects were more proficient in Arabic, one may expect such a reduction for Arabic. It is not clear whether the same applies also to Hebrew. In addition, to the extent that the subjects were proficient in Hebrew, we may expect that Stroop interference in the within-language conditions will be larger than in the between-language conditions. This would support the concept mediation hypothesis. Suppose that the present subjects could control only Arabic, whereas Hebrew was processed automatically and without

BILINGUAL STROOP CONTROL stimuli. Each group of five stimuli consisted of all five color names and color inks. For each session the five incongruent trials were selected out of the twenty possible stimuli. Each session had a different set of such five stimuli so that the twenty possible stimuli were presented in the four sessions. To recapitulate, response language and frequent language were manipulated between sessions, whereas color-word congruency and stimulus language were manipulated within session. All variables were manipulated within subjects. Hence, this is a four-way repeated measures design. The experiment is composed of 16 conditions: 2 response languages (Hebrew vs. Arabic) x 2 levels of language frequency (80% Hebrew vs. 80% Arabic) x 2 congruency levels (congruent vs. incongruent) x 2 stimulus languages (word written in Hebrew vs. word written in Arabic). Procedure. Each subject participated in four sessions. Each session was homogeneous with respect to the response language. There were four different sessions according to the response and frequent language conditions. Order of sessions was counterbalanced across subjects by applying a Latin square arrangement. Thus, there were four groups of subjects according to the four possible orders of sessions. The trials in the block were randomly ordered and presented to each subject either from the first to the 50th or vice versa. As mentioned before, each block was presented twice within each session. The task was explained to the subjects at the beginning of each session. They were asked to name the ink color and ignore the written word. They were asked to respond as fast as possible. Each session was preceded by an explanation of its language structure (i.e., the frequency relations between stimuli in the two languages). A session began with a practice block often trials, designed in the same way as the experimental block. Each trial started with a fixation point (a black cross on a white background) that was on for 300 ms and that was replaced by a 200ms blank (dark field) interval. The color word appeared at the end of the interval and stayed on until the subject's naming response. The subject's response stopped the timer and eliminated the stimulus.

763

There were significant congruency effects in the same language conditions also, F([, 15) = 127.04, p < .001, for Hebrew; F(l, 15) = 74.06, p< .001, for Arabic. The congruency effect was also dependent on the correspondence between the stimulus language and the frequent language. This was indicated by the significant triple interaction among these variables, F(l, 15) = 8.53, p < .01. The relevant means appear in Table 3. For the Hebrew written stimuli there was no significant congruency by frequent language interaction [F < 1]. The congruency effect, across the two frequent language conditions, was very clear, F(l, 15) = 240.49, p < .001. For the Arabic written stimuli, congruency interacted with frequent language, F(l, 15) = 6.66, p < .01. This interaction was reflected in a reduction of the interference effect when Arabic was frequent (76 ms) relative to the condition in which Arabic was infrequent (i.e., Hebrew was the frequent language, 121 ms). Note, however, that the congruency effect was significant both when Hebrew was frequent, F(l, 15) = 65.01, p < .001 and when it was infrequent, F(l, 15) = 38.07, p < .001. The interaction between frequent language and response language (across congruency) was significant F(l, 15) = 106.55, p < .001. When most of the stimuli were written in Arabic the response language did not make any difference, F < 1. When the majority of the stimuli were written in Hebrew, responses in Arabic were 88 ms faster than responses in Hebrew, F(l, 15) = 21.31, p< .001. Stimulus language also interacted with response language, F(l, 15) = 17.85, p< .001, and with frequent language, F(l, 15) = 45.57, p < .001. These effects, however, appeared also as part of higher order interactions and therefore were not analyzed any further.

Results
For each subject in each condition the median reaction time (RT) of the correct responses was calculated.2 These medians were subjected to a four-way repeated measures analysis of variance. Means and standard deviations of the median reaction times in the various conditions of Experiment 1 are presented in Table 1. Congruent ink colors were named faster than incongruent stimuli, F(\, 15) = 176.35, p < .001. This interference effect depended on the different language and same language conditions, resulting in a significant triple interaction among stimulus language, response language, and congruency, F(l, 15) = 37.26, p < .005. This interaction appears in Table 2. As indicated in Table 2, Stroop interference was much larger when the stimulus and the response languages matched than when they mismatched. The two simple interaction effects, between congruency and stimulus language, were significant, both when the subjects responded in Hebrew, F(l, 15) = 59.55, p < .001, and when they responded in Arabic, F(l, 15) = 11.86, p < .01. Further analyses of simple main effects indicated that there were significant congruency effects in the different language conditions, F(l, 15) = 54.46, p < .001, wherr. subjects responded in Hebrew and the words were written in Arabic; F(l, 15) = 27.69, p < .001, when subjects responded in Arabic and the words were written in Hebrew.

Discussion
As expected, we found a within-language and a betweenlanguages Stroop interference effect. Furthermore, as in most similar experiments (e.g., Fang, Tzeng, & Alva, 1981), the between-languages effects were smaller. The two within-language Stroop effects were about the same size, and the same findings held for the between-languages conditions. Arabic was our subjects' first language, and they reported higher proficiency in Arabic than in Hebrew. ,Yet, it seems that the between- and the within-languages Stroop interference did not reveal Arabic dominance. This is consistent with the subjects' high proficiency in Hebrew. This pattern of results supports the concept mediation hypothesis (Chen & Ho, 1986; Chen & Leung, 1989; Potter et al., 1984).
2

Subjects made about 2% errors. Almost all errors were colornaming errors. In a few cases, subjects made response language errors (i.e;, responded in a wrong language). It should also be noted that in a "non-frequent language" condition, an error rate of 1 % was equivalent to one trial. The same rate of errors in a "frequent-language condition" was equivalent to four trials. It follows that the standard errors are different in the frequent and the infrequent conditions. Therefore, after verifying that there was no indication for a speedaccuracy tradeoff in our data, we focused on the reaction times.

764

J. TZELGOV, A. HENIK, AND D. LEISTER

Table 1 Means (in Milliseconds) and Standard Deviations in the Experimental Conditions of Experiment 1
Frequent language: Hebrew Stimulus language: Hebrew Statistic C I Stimulus language: Arabic C I Frequent language: Arabic Stimulus language: Hebrew C I Stimulus language: Arabic C I

M
SD

724 84

877 94

824 93

Response language: Hebrew 897 718 882 94 117 110 Response language: Arabic 829 805 885 144 120 113

759

85
718 90

808 91

M SD

706 84

776 92

662 95

821 132

Note. C and I represent congruent and incongruent conditions, respectively.

Is there any evidence for controlled processing? In the introduction, we suggested that control may be induced by expectations (i.e., the frequent language variable). We found a similar Stroop interference for stimuli written in Hebrew regardless of their expectancy level. That is, even though on certain sessions the subjects knew that the next stimulus would probably be written in Hebrew, they could not suppress reading the irrelevant color name. In contrast, expectation influenced the size of the interference effect of color names written in Arabic. The knowledge that, on certain sessions, the next stimulus would probably be written in Arabic did enable the subjects to reduce the amount of interference. In the case of Hebrew written stimuli the Stroop effect was insensitive to expectations. However, expectations did affect the processing of Hebrew written stimuli; when such stimuli were expected they were processed faster than when unexpected. This effect was not dependent on congruency relations. We would like to claim that the reduction in the Stroop effect found in the experiment reflects control processes. Therefore, it is important to analyze how the decrease in the interference effect is accomplished. The basic two conditions used in the original Stroop task (Stroop, 1935) were the incongruent condition (e.g., BLUE written in red) and a

Table 2 Mean Reaction Times in Experiment 1 (in Milliseconds) as a Function of Stimulus Language, Response Language, and Color Name Congruency
Stimulus language Response language Hebrew Arabic Hebrew Cong. 721 791 Incong. 880 852 I.E. 159 61 Cong. 757 689 Arabic Incong. 830 825 I.E. 73 136

neutral condition (e.g., XXXX written in red). Another experimental condition frequently used in Stroop tasks is the congruent condition (e.g., RED written in red). The difference between the incongruent and the neutral trials is usually referred to as "inhibition," whereas the difference between the neutral and the congruent trials is referred to as "facilitation." In the reported experiment we used an incongruent and a congruent condition.3 The Stroop effect, or "interference effect," we are discussing is therefore the sum of the components of facilitation and inhibition. A reduction in the interference effect reflects either a reduction in the incongruent RTs or an elevation of the congruent RTs or both. Assuming that the "neutral" RT is more or less constant, a reduction in the incongruent RTs implies weakening of the inhibition component. In parallel, elevation of the congruent RTs implies reduction in the facilitation component. We believe that control reflects adaptive mental operations that evolved in order to deal with specific situational demands (see also Rozin, 1976). It follows that only the weakening of the inhibition (or the strengthening of the facilitation component) are consistent with the idea of control. Keeping this analysis in mind, we again examined the data. As can be seen in Table 3, when subjects whose first language was Arabic were expecting an Arabic written word and such a stimulus appeared, it reduced the RTs in the incongruent condition, F\l, 15) = 10.66, p < .01, without affecting the congruent trials (F < 1). In contrast, expectation for Hebrew words affected congruent and incongruent Hebrew written stimuli to an equal amount, (F < 1, for the relevant interaction). Thus, it seems fair to conclude that expectations affect mainly the inhibition com-

Note. I.E. is the interference effect defined as the difference between reaction times in the incongruent and congruent conditions; Cong. = Congruent; Incong. = Incongruent.

During the last year we ran several (monolingual) Stroop experiments in which we manipulated the proportion of neutral trials. A consistent finding emerged from these experiments: the size of the Stroop effect (and, to be more accurate, the inhibition component) is positively correlated with the proportion of the neutrals. In other words, increasing the proportion of neutrals decreases control. Thus, according to our best knowledge the presence of neutral trials affects the very process of control: it decreases controllability.

BILINGUAL STROOP CONTROL

765

Table 3 Mean Reaction Times in Experiment 1 (in Milliseconds) as a Function of Stimulus Language, Language Frequency, and Color Name Congruency
Stimulus language Hebrew Arabic Frequent language Cong. Incong. I.E. Cong. Incong. I.E. Hebrew 717 826 109 742 863 121 Arabic 762 884 814 738 76 122 Note. I.E. is the interference effect defined as the difference between reaction times in the incongruent and congruent conditions; Cong. = Congruent; Incong. = Incongruent.

ponent of the Stroop effect, which is consistent with the idea of control. This finding is not necessarily consistent with Logan's (1980) model of Stroop control that implies symmetric effects of control on facilitation and inhibition. Thus it may be that several mechanisms of control exist. However, additional data are needed before reaching such a conclusion. We would like to briefly mention two points. One is related to the relationship between expectations and the betweenversus within-language pattern of interference. Expectations (i.e., the frequent language variable) did not influence the within-language versus the between-language pattern of Stroop interference. To the degree that the relations between a within-language and a between-language pattern of interference reflect the underlying structure of the bilingual lexicon, the present pattern of results suggests that the issue of control versus automaticity is independent from that of the structure of the lexicon. It also implies independence of the input and the output aspects of language processing (MacNamara & Kushnir, 1971). The second point is related to control and language proficiency. Our subjects reported better proficiency in Arabic than in Hebrew. They succeeded in reducing the interference effect for Arabic expected words but not for Hebrew expected words. This pattern of results supports the claim that one by-product of proficiency is control. In the Stroop paradigm, control allows the subject to reduce the interference effect. Accordingly, one may predict a reversed pattern of interference for subjects who are more proficient in Hebrew than in Arabic. That is, for such subjects one expects to find a reduced Stroop interference for Hebrew expected words and little or no effect of expectation on Arabic words. Our next experiment addressed this issue. In it, we used subjects who were more proficient in Hebrew than in Arabic.

of Israeli high school graduates study Arabic for about four years. Out of this population, we recruited a sample of students for the present experiment. These subjects were therefore less proficient in Arabic (i.e., their second language) than were Experiment 1 subjects in Hebrew (i.e., their second language). We assumed that the two samples were equally proficient in their respective first languages. With respect to automaticity and control, we expected to find evidence for control, which should be reflected in a reduction in the amount of Stroop interference in the first language. We found such a pattern in Arabic in Experiment 1 and expected to find a similar pattern for Hebrew in Experiment 2. With respect to the structure of the lexicon, reflected in the within-language and the between-languages pattern of interference, it was possible that Experiment 2 subjects would deviate from the pattern obtained in Experiment 1. Because Experiment 2 subjects were less proficient in their second language than Experiment 1 subjects were in theirs, the present within-language interference may not be larger than the between-languages interference. Instead, we predicted that Hebrew written words would cause a relatively large interference effect both when Hebrew was the response language (i.e., a within-language condition), and when Arabic was the response language (i.e., a between-languages condition).

Method
Subjects. Sixteen subjects participated in the experiment, either in partial fulfillment of course requirement or for payment. All subjects were of Jewish origin and studied Arabic as a second language for 4 years or more. All of them were using Arabic in the restricted context of work or academic studies and Hebrew in their daily interactions. Three of the subjects were high school teachers of Arabic, but even they declared that their knowledge of Hebrew was much better than their knowledge of Arabic. Procedure. All other details of the method were exactly the same as those of Experiment 1.

Results
For each subject in each condition, we calculated the median RT of the correct responses. These medians were subjected to a four-way repeated measures analysis of variance. Means and standard deviations of the median reaction times in the various conditions of Experiment 2 appear in Table 4. Congruent ink colors were named faster than incongruent stimuli, F(\, 15) = 79.56, p < .005. This interference effect was moderated by the same two triple interactions that were significant in Experiment 1. The interference effect depended on the correspondence between the stimulus and the response language, resulting in a significant triple interaction among stimulus language, response language, and congruency, F(\, 15) = 12.42, p < .01. This interaction appears in Table 5. It seems that there was no clear difference between the within- and the between-language conditions. The interference effects in three of the four combinations of stimuluslanguage and response-language were very similar. There was

Experiment 2
The purpose of Experiment 2 was to replicate the main findings of the previous experiment; namely, to show that Stroop interference can be diluted when subjects expect a stimulus in their first language. We used the same set of stimuli but with subjects whose first language was Hebrew and whose second language was Arabic. In Israel, Arabic is studied in high school as an elective, so a certain percentage

766

J. TZELGOV, A. HENIK, AND D. LEISTER

Table 4 Means (in Milliseconds) and Standard Deviations in the Experimental Conditions of Experiment 2
Frequent language: Hebrew Stimulus language: Hebrew Statistic M SD M SD C 662 97 I 774 94 Stimulus language: Arabic C I Frequent language: Arabic Stimulus language: Hebrew C I Stimulus language: Arabic C 693 74 691 95 I 713 81 781 138

Response language: Hebrew 731 746 674 807 92 102 109 125

Response language: Arabic 721 811 708 816 758 899 106 150 134 161 136 172 Note. C and I represent congruent and incongruent conditions, respectively.

one combination that was at odds with the other three. The effect shrank markedly when the words were written in Arabic and the subjects responded in Hebrew. The simple interaction between congruency and stimulus language was not significant when the subjects responded in Arabic, F( 1, 15) < 1, but it was significant when they responded in Hebrew, F(\, 15) = 33.58, p < .001. Under the latter conditions there was a clear 121-ms effect when the stimuli were written in Hebrew. In addition, there was a small and insignificant effect, in the same direction, when the stimuli were written in Arabic, F(l, 15) = 3.65,/;< .10. The congruency effect was also dependent on the correspondence between the stimulus language and the frequent language. This was indicated by the significant triple interaction among these variables, 7-1(1, 15) = 4.84, p < .05. The relevant means appear in Table 6. This interaction resembles the one reported in Experiment 1. For the Hebrew written stimuli there was a significant congruency by frequent language interaction, F(l, 15) = 4.60, p < .05. The congruency effect was reduced when Hebrew was the frequent language (100 ms) relative to conditions in which Hebrew was infrequent (136 ms). For the Arabic written stimuli there was no significant congruency by frequent language interaction, F < 1. The congruency effect, across the two frequent language

conditions, was clear, F(l, 15) = 39.19, p < .001. In contrast, the frequency effect for such stimuli did not reach significance, F(l, 15) = 3.28,p>.05. Given the distinction between the effects on facilitation and inhibition components of the Stroop effect, we examined another partition of this interaction. Expectations for Hebrew words resulted in significantly shorter RTs in the incongruent condition, F(l, 15) = 6.32, p < .05, but not in the congruent condition, F(l, 15) = 1.50, ns. Stimulus language interacted with response language, F\l, 15) = 6.30, p < .05, with frequent language, F( 1, 15) = 51.84, p < .001, and with congruency, F(l, 15) = 19.48, p < .001. In addition, response language interacted with congruency, F( 1,15) = 9.46, p < .01. These effects, however, were included in the two triple interactions discussed above and therefore were not analyzed any further. Two additional main effects were significant. Responses in Hebrew were faster than responses in Arabic, F\l, 15) = 6.27, p < .025, and Hebrew written stimuli resulted in longer RTs than Arabic written stimuli, F(\, 15) = 13.64, p < .01.

Discussion
The subjects of the present experiment were less proficient in Arabic than in Hebrew. In general, their responses in Hebrew were faster than their responses in Arabic. Furthermore, when they responded in Hebrew there was only a very small interference by Arabic written stimuli. The within-language interference effect was not necessarily stronger than the between-languages interference effect. The within-language interference effect was 121 ms for Hebrew and 99 ms for Arabic stimuli. The between-language interference effect of Hebrew written stimuli was 115 ms, whereas it was 17 ms for Arabic written stimuli. Hence the withinlanguage and the between-languages Stroop effects revealed Hebrew dominance (which was in agreement with the faster responses in Hebrew). This pattern of results does not fit the concept mediation hypothesis, which was in agreement with Experiment 1 results. In fact, at least the relative sizes of these

Table 5 Mean Reaction Times in Experiment 2 (in Milliseconds) as a Function of Stimulus Language, Response Language, and Color Name Congruency
Stimulus language Hebrew Arabic Response language Cong. Incong. I.E. Cong. Incong. I.E. Hebrew 17 669 790 121 713 730 Arabic 740 855 115 700 799 99 Note. I.E. is the interference effect defined as the difference between reaction times in the incongruent and congruent conditions; Cong. = Congruent; Incong. = Incongruent.

BILINGUAL STROOP CONTROL

767

Table 6 Mean Reaction Times in Experiment 2 (in Milliseconds) as a Function of Stimulus Language, Language Frequency, and Color Name Congruency
Stimulus language Frequent language Hebrew Arabic Hebrew Cong. 692 717 Incong. 792 853 I.E. 100 136 Cong. 720 692 Arabic Incong. 781 748 I.E. 61 56

Note. I.E. is the interference effect defined as the difference between reaction times in the incongruent and congruent conditions; Cong. = Congruent; Incong. = Incongruent.

interference effects are in agreement with the word association hypothesis. A similar pattern of results was reported by Chen and Ho (1986) for Chinese speakers who were at the first stages of learning English. It also fits the proficiency hypothesis suggested by Magiste (1984). As in the previous experiment, we found evidence for control. The pattern here was similar to the results found in Experiment 1. In Experiment 2, as in the previous experiment, only the RTs in the incongruent conditions were affected by expectations for stimuli written in Hebrew. We found similar Stroop effects for stimuli written in Arabic regardless of their expectancy level. Even though on certain sessions the subjects knew that the next stimulus would probably be written in Arabic, they could not suppress reading the irrelevant color name. In contrast, expectation influenced the size of the interference of color names written in Hebrew. It should be noted that expectations did not influence the relations between the within-language and the between-language pattern of Stroop interference. The same was true in Experiment 1. General Discussion We used the same set of stimuli in two experiments with Arabic-Hebrew bilinguals. The subjects in the two experiments differed in what they defined as their first language and in their proficiency in their first and second languages. The subjects who took part in Experiment 1 were quite proficient in the two languages and their first language was Arabic. The first language of the subjects who took part in Experiment 2 was Hebrew; however, these subjects were much more proficient in their first language (Hebrew) than in their second language. Certain differences in the results of the two experiments stem from these differences. The results of the two experiments have significant implications for the two issues addressed earlier, namely, the structure of the bilingual lexicon and the issue of automatic versus controlled characteristics of skilled performance. In the first experiment the within-language Stroop effect was larger than the between-language effect. The pattern of the between- and the within-languages interference effects did not reveal dominance of either Arabic or Hebrew. In contrast, the interference created by Hebrew in the second experiment

was relatively large in both the within-language and the between-language's conditions, whereas the interference created by Arabic was relatively large only under the within-language condition. Thus, in the second experiment, Hebrew emerged as the dominant language. The pattern of results in the first experiment is in agreement with the concept mediation hypothesis, whereas that of the second experiment is in agreement with the word association hypothesis. Taken together, the results of the present experiments fit the mixed model of concept representation advocated by Chen and Ho (1986). They suggested that at the first stage of learning a second language, subjects access the new language via the representations of the first language (i.e., the word associations hypothesis). As proficiency in the second language increases, dependence on the first language decreases and subjects tend to rely on amodal concepts. Chen and Ho (1986) and Chen and Leung (1989) reported such a pattern of results with Chinese-English bilinguals. However, Potter et al. (1984) did not find indications for such a mixed model of concept representation. Hence, it is possible that such a model is limited to very concrete concepts such as colors. The fact that the pattern of results in Experiment 1 is consistent with the concept mediation hypothesis, whereas the results of Experiment 2 are consistent with the word association hypothesis, may also be interpreted in a different way.4 The subjects in Experiment 1 learned their second language in early childhood, whereas the subjects in Experiment 2 acquired their second language as adolescents. Recently, Johnson and Newport (1989) extended the concept of a critical period of language acquisition to second language acquisition. Thus the differences found between the groups may reflect the fact that the subjects in Experiment 1 acquired their second language during the critical period, whereas subjects of Experiment 1 acquired their second language after the end of the critical period. What about the proficiency hypothesis suggested by Magiste (1984)? According to this hypothesis, the degree of interference created by a language is directly related to the proficiency in that language. Note that this monotonic relationship between proficiency and the Stroop effect is not limited to learning a second language. It can be found when children practice their first language (e.g., Schiller, 1966). However, this pattern of results, which fits the word association hypothesis, is limited to the initial stages of learning a language (be it the first or the second language). At these initial stages of learning, interference reflects the progress towards automatization of the reading process. The proficiency hypothesis does not capture the phenomenon of control that apparently characterizes the more advanced stages of language learning. In other words, further practice does not necessarily increase interference; it may even reduce it. A pattern of results consistent with this interpretation was reported by Ehri and Ryan (1980). They investigated interference of words in a picture-naming task. They found that words written in the

4 This interpretation was implied in a comment by one of the anonymous reviewers of an earlier version of this article.

768

J. TZELGOV, A. HENIK, AND D. LEISTER

second language of their bilingual subjects produced more interference than words written in their first language. In addition, Fang et al. (1981) found larger Stroop interference in the second language of their Chinese-English and SpanishEnglish bilinguals. Fang et al. (1981) have shown that the magnitude of reduction from within- to between-language Stroop interference is a linearly decreasing function of orthographic similarity. Arabic and Hebrew differ in their orthography. Accordingly, taking Fang et al.'s (1981) analysis as a starting point, one should expect a reduction of at least 110 ms (see Fang et al., 1981, Table 3, p. 613) in the between-language's (ArabicHebrew or Hebrew-Arabic) in comparison to a within-language condition. Such a reduction would be comparable to the reduction in a switch from a within- to a betweenlanguages condition in Hungarian-English and Japanese (Kana)-English bilinguals. In contrast, the reduction of the interference effect in the between-languages condition of our Arabic-Hebrew bilinguals was considerably smaller (see Experiment 1, Table 2). In Experiment 2 the interference effect in the Hebrew-Arabic condition was larger than the interference in the Arabic-Arabic condition (see Table 4). We are not claiming that orthographic similarity is not important. It seems, however, that the relative language proficiency, as reflected in the mixed model presented earlier, is the dominant factor in determining the pattern of bilingual Stroop effect. Can the sensitivity of the interference to language frequency be explained as resulting from increased practice? Dyer (1971) compared the interference to color naming of (incongruent) color words written in seven different languages in a bilingual Stroop task. He found that the conditions that produced the greatest interference in color naming showed the greatest improvement over sessions. In parallel, it may be suggested that the smaller interference in the 80% condition reflects the fact that the subjects had more opportunity to practice with stimuli in a frequent language and that this effect of practice was stronger for the language that interferes more. To reject this possibility, we reanalyzed our data in search of possible practice effects. As described in the method section of Experiment 1, there were four groups of subjects that differed in the order of sessions. Because the sensitivity of the Stroop effect to frequency was limited to the first language of our subjects, we focus first on that language. The practice hypothesis predicts that sensitivity of the Stroop effect to language frequency should be greater in the groups in which the (first) language was frequent in the first two sessions than in the groups that were run in an opposite order of sessions. This prediction is based on the idea that the effect of practice is dominant mainly during the first two sessions. Furthermore, at the end of the fourth session the amount of practice was equal in all four groups. Therefore, the practice effect should act to amplify the influence of the frequency variable when the frequent conditions preceded the infrequent conditions. It should also act to reduce the influence of the frequency variable when the infrequent conditions came before the frequent conditions. This in turn implies an interaction between the "group" variable and the remaining variables (ex-

cluding response language) of the experiments. However, in both experiments, none of the interactions including simultaneously group and language frequency approach significance. Assuming a significance level of .05, the critical F ratio for each of these interactions is 3.49, whereas the highest F ratio obtained in the case of the above mentioned interactions was 2.33. Thus an "increased practice" explanation of our results can be rejected. It may be assumed that an increase in language proficiency, and in reading in particular, affects mainly reading time and not color-naming time. It usually takes less time to read a color word than to name its color (Dyer, 1971). Increased language proficiency should result in an easier discrimination of the arrival times of color names and word names. Thus, if the (irrelevant) color word consistently proceeds the (relevant) color name, these arrival time relations could be used as a cue for selecting the relevant (or suppressing the irrelevant) information. It would also leave more time for disregarding the irrelevant information. Such a model predicts a decrease in the interference with increased proficiency. However, it is hard to see how it could explain the sensitivity of the interference to frequency unless one claims that our frequency manipulation reflects additional practice that leads to additional proficiency. We have already shown that an "increased practice" interpretation of the frequency effect is unlikely. In addition, it seems unrealistic to assume that this relatively short "practice," in which reading is in fact undesirable, has a significant effect on reading proficiency. Expectations may affect processing time of stimuli in the second language as well. In Experiment 1 the processing of Hebrew written stimuli was accelerated by expectations that most stimuli in a block were going to be in that language. It should be stressed that in this case it is not the Stroop interference that is influenced. When Hebrew words are expected they are processed faster than when they are unexpected. Note that this happens even though the meaning of the word is irrelevant and that the effect is independent of congruence. The Stroop effect reflects differences in the (color) word meaning, and as such it reveals processes in the semantic network. In contrast, expectations for the less known language do not affect the semantic level of processing, but they still may enhance the general level of performance by speeding up the access to the lexicon of the expected (but to be ignored) language. We believe that this general effect on reaction times reflects increased automaticity rather than control. It seems that in the process of increasing language proficiency, humans acquire the ability to focus on that language. This selectivity of processing because of increased proficiency is similar to the idea of "tuning" as used by Anderson (1982). In Experiment 2, expectations for stimuli in the second language (Arabic in this case) did not reduce reaction times. This pattern of results is consistent with a mixed model of representation. The subjects in the second experiment were at the first stages of learning Arabic and, according to the mixed model, their access to the semantic level was always mediated by Hebrew. Therefore they did not benefit from expectations for Arabic; their access to the semantic network was through the Hebrew lexicon in any case.

BILINGUAL STROOP CONTROL

769

Logan (1985) pointed out that skilled performers were usually able to control their performance better than unskilled performers, in spite of the fact that skilled performance was characterized by automaticity. Thus, control increases rather than decreases with skill. Similarly, our results indicate that Stroop interference is controllable, and that proficiency is a precondition for such control. In both experiments we found that subjects were able to decrease the interference of the expected language. Hebrew-speaking Arabs, in Experiment 1, were able to do it only for the Arabic stimuli, whereas the Arabic-speaking Jews, in Experiment 2, were able to do it only for the Hebrew stimuli. Note that control was limited; the Stroop effect did not disappear, it shrank. This fact was quite prominent in Experiment 2; interference of Hebrew stimuli was smaller in the 80% Hebrew stimuli condition than in the 20% Hebrew stimuli condition. Nevertheless, the interference created by Hebrew words was still larger than the interference created by Arabic words. The results of the two experiments indicate that language proficiency was a precondition for control. This implies a curvilinear relationship between Stroop interference and reading proficiency in a given language. A minimal level of reading proficiency is required to obtain the interference at all. One has to know that the letters R-E-D mean "red" for their appearance to interfere with performance. Automatization of the reading tasks that follow practice speeds up the reading process, and as a result interference increases. In the advanced stages of acquiring the proficiency in reading, one learns to apply control mechanisms to the reading processes, and as a result interference diminishes. According to this analysis, one should expect interference to increase at the first stages of learning a language and to decrease later on. This pattern was indeed found by Schiller (1966). We would like to suggest that the present (and similar) evidence for controlled processing reflects the operation of consciousness. Several authors have emphasized the role of consciousness in the development of new behaviors (Carr, 1979; Rozin, 1976). This idea is in accord with conceptualizations of consciousness as a mechanism that controls and selects specific subroutines during performance of various tasks (Frith, 1981; Keele, 1973; Shallice, 1972). The present results do show that control, reflected in the ability to withhold certain automatic processes, may be evidenced in skilled readers. Furthermore, it seems that such control emerges only when the readers are really proficient in a given language. Note, however, that the present results evidence automaticity (i.e., Stroop interference) and control (i.e., a reduction of Stroop interference under certain conditions) simultaneously. We would like to agrue that automaticity and control are both characteristics of skilled performance, but each reflects a different aspect of the system, and each subserves, essentially, a different function. The present pattern of results supports the notion that automatic and controlled processing may exert their influences simultaneously. Our last comment is concerned with two related issues, the locus of the Stroop effect and the source of the control process. One group of models attributes the effect to the perceptual

encoding stage (e.g., Hock & Egeth, 1970). Such models assume that the interference effect discloses a breakdown of selective attention. The perceptual encoding models have been criticized because they cannot explain why congruent stimuli, which have the maximal semantic similarity between the word and its color, do not cause maximal interference (Seymour, 1977). Another group of models (e.g., Morton, 1969; Posner, 1978; Stroop, 1935) explains the interference as resulting from response competition between the process of color naming and the relatively faster process of word reading. Such "horse race" explanations of the Stroop effect that predict a monotonic relationship between proficiency and interference have been recently evaluated and criticized by Dunbar and MacLeod (1984). Our results, which point to the decline in interference in the better known language, constitute a further challenge to the simple horse race model interpretation of the Stroop phenomenon. In addition, the ability to control the interference was independent of the response language. In other words, the ability to reduce the interference was not affected by the proficiency of language of response. This is another indication against a response competition explanation of the Stroop effect. It seems that our results are best explained in terms of Seymour's (1977) "conceptual encoding" model. The model places the locus of interference at the semantic level and attributes it to the simultaneous activation of two nodes in the semantic network. The extra processing time reflects the processes involved in the suppression of the irrelevant activation. Such an interpretation is consistent with Neill's (1977; Neill & Westberry, 1987) and Tipper's (1985) proposal that selection of relevant information may proceed by inhibition of irrelevant information. Tipper and his coworkers (Tipper, 1985; Tipper & Cranston, 1985; Tipper & Driver, 1988; Driver & Tipper, 1989) have shown that unaffected stimuli have negative priming effects. This implies that the unattended primes were identified before being inhibited. It may be that a similar process is reflected in the Stroop effect. If that is the case, our results imply that language proficiency and expectations determine the efficiency of the suppression process. References
Anderson, J. R. (1982). Acquisition of cognitive skill. Psychological Review, 89, 369-406. Carr, T. H. (1979). Consciousness in models of human information processing: Primary memory, executive control and input regulation. In G. Underwood and R. Stevens (Eds.), Aspects ofConsciousness, Vol. 1. New York: Academic Press. Chen, H., & Ho, C. (1986). Development of Stroop interference in Chinese-English bilinguals. Journal of Experimental Psychology: Learning, Memory and Cognition, 12, 397-401. Chen, H. C, & Leung, Y. S. (1989). Patterns of lexical processing in non-native language. Journal of Experimental Psychology: Learning, Memory, and Cognition, 15, 316-325. Collins, A. M., & Loftus, E. (1975). A spreading activation theory of semantic processing. Psychological Review, 82, 407-428. Dalrymple-Alford, E. (1968). Interlingual interference in the Stroop task. Psychonomic Science, 10, 215-216. Driver, J., & Tipper, S. P. (1989). On the nonselectivity of selective

770

J. TZELGOV, A. HENIK, AND D. LEISTER

seeing: Contrast between interference and priming in selective the color-word Stroop task. Bulletin of the Psychonomic Society, attention. Journal of Experimental Psychology: Human Perception 22, 135-138. and Performance, 15, 304-314. McCormack, P. D. (1976). Language as an attribute of memory. Dunbar, K., & MacLeod, C. M. (1984). A horse race of a different Canadian Journal of Psychology, 30, 238-248. color: Stroop interference patterns with transformed words. Journal Magiste, E. (1984). Stroop tasks and dichotic translation: The develof Experimental Psychology: Human Perception and Performance, opment of interference patterns in bilinguals. Journal of Experi10,622-639. mental Psychology: Learning, Memory and Cognition, 10, 304315. Dyer, F. N. (1971). Color-naming interference in monolinguals and bilinguals. Journal of Verbal Learning and Verbal Behavior, 10, Magiste, E. (1986). Selected issues in second and third language 297-302. learning. In J. Vaid (Ed.), Language processing in bilinguals: Psycholinguistic and Neurological perspectives. Hillsdale, NJ: Erlbaum. Ehri, L. C, & Ryan, E. L. (1980). Performance of bilinguals in a picture-word interference task. Journal of Psycholinguistic ReMacNamara, J., & Kushnir, S. L. (1971). Linguistic independence of search, 9, 285-302. bilinguals: The input switch. Journal of Verbal Learning and Verbal Fang, S. P., Tzeng, O. J. L., & Alva, L. (1981). Intralanguage vs. Behavior, 10, 480-487. interlanguage Stroop effects in two types of writing systems. MemMorton, J. (1969). Categories of interference: verbal mediation and ory & Cognition, 9, 609-617. conflict in card sorting. British Journal ofPsychology, 60, 329-346. Frith, C. D. (1981). Schizophrenia: An abnormality of consciousness? Neely, J. M. (1977). Semantic priming and retrieval from lexical In G. Underwood & R. Stevens (Eds.), Aspects of Consciousness, memory: Roles of inhibitionless spreading activation and limitedVol. 2. New York: Academic Press. capacity attention. Journal of Experimental Psychology: General, 106, 226-254. Gerard, L. D., & Scarborough, D. L. (1989). Language-specific lexical access of homographs by bilinguals. Journal of Experimental PsyNeill, W. T. (1977). Inhibitory and facilitatory processes in selective chology: Learning, Memory and Cognition, 15, 303-315. attention. Journal of Experimental Psychology: Human Perception and Performance, 3, 444-450. Grainger, J., & Beauvillain, C. (1989). Associative priming in bilinNeill, W. T., & Westberry, R. L. (1987). Selective attention and the guals: Some limits of interlingual facilitation effect. Canadian suppression of cognitive noise. Journal of Experimental Psychology: Journal of Psychology, 261-273. Learning, Memory and Cognition, 13,121-334. Groot, A. M. B., de, & Nas, G. (1989). The bilingual lexicon: Some within- and between-language connections between lexical represen- Posner, M. I. (1978). Chronometric Explorations of Mind. Hillsdale, NJ: Erlbaum. tations. Manuscript submitted for publication. Posner, M. I., & Snyder, C. R. R. (1975). Attention and cognitive Hasher, L., & Zacks, R. T. (1979). Automatic and effortful processes control. In R. L. Solso (Ed.), Information Processing and Cognition: in memory. Journal of Experimental Psychology: General, 108, The Loyola Symposium. Hillsdale, NJ: Erlbaum. 356-388. Hock, H. S., & Egeth, H. E. (1970). Verbal interference with encoding Potter, M. C, So, K., Von Eckardt, B., & Feldman, L. B. (1984). in a perceptual classification task. Journal of Experimental PsyLexical and conceptual representation in beginning and proficient chology, 83, 299-303. bilinguals. Journal of Verbal Learning and Verbal Behavior, 23, 23-38. Johnson, J. S., & Newport, E. L. (1989). Critical period effects in Preston, M. S., & Lambert, W. E. (1969). Interlingual interference in second language learning: The influence of maturational state on a bilingual version of the Stroop color-word task. Journal of Verbal the acquisition of English as a second language. Cognitive PsycholLearning and Verbal Behavior, 8, 295-301. ogy, 21, 60-99. Rozin, P. (1976). The evolution of intelligence and access to the Kiyak, H. A. (1982). Interlingual interference in naming color words. cognitive unconscious. In J. M. Sprague & A. N. Epstein (Eds.), Journal of Cross-Cultural Psychology, 13, 125-135. Progress in Psychobiology and Physiological Psychology. New Keele, S. W. (1973). Attention and Human Performance. Pacific York: Academic Press. Palisades, CA: Goodyear. Schadler, M., Thissen, D. M. (1981). The development of automatic Kirsner, K., Smith, M. C, Lockhart, R. S., King, M. L., & Jain, M. word recognition and reading skill. Memory & Cognition, 9, 132(1984). The bilingual lexicon: Language-specific units in an inte141. grated network. Journal of Verbal Learning and Verbal Behavior, Schiller, P. H. (1966). Developmental study of color-word interfer23, 519-539. ence. Journal of Experimental Psychology, 72, 105-108. Levelt, W. J. M. (1983). Monitoring and self-repair in speech. CogSchneider, W., Dumais, S. T., & Shiffrin, R. M. (1984). Automatic nition, 14,41-104. and control processing and attention. In R. Parasuraman & D. R. Logan, G. D. (1979). On the use of a concurrent memory load to Davies (Eds.), Varieties ofAttention. New York: Academic Press. measure attention and automaticity. Journal ofExperimental PsySeymour, P. H. K. (1977). Conceptual encoding and locus of the chology: Human Perception and Performance, 5, 189-207. Stroop effect. Quarterly Journal of Experimental Psychology, 29, Logan, G. D. (1980). Attention and automaticity in Stroop and 245-265. priming tasks: Theory and data. Cognitive Psychology, 12, 523Shallice, T. (1972). Dual functions of consciousness. Psychological 553. Review, 79, 383-393. Logan, G. D. (1982). On the ability to inhibit complex movements: Shiffrin, R. M., & Schneider, W. (1977). Controlled and automatic A stop-signal study of typewriting. Journal of Experimental Psyhuman information processing: II. Perceptual learning, automatic chology: Human Perception and Performance, 8, 778-792. attending, and a general theory. Psychological Review, 84, 127Logan, G. D. (1985). Skill and automaticity: Relations, implications, 190. and future directions. Canadian Journal of Psychology, 39, 367Stroop, J. R. (1935). Studies of interference in serial verbal reactions. 386. Journal of Experimental Psychology, 18, 643-662. Logan, G. D., & Zbrodoff, N. J. (1979). When it helps to be misled: Tipper, S. P. (1985). The negative priming effect: inhibitory effects of Facilitative effects of increasing the frequency of conflicting stimuli ignored primes. Quarterly Journal of Experimental Psychology, in a Stroop-like task. Memory & Cognition, 7, 166-174. 37A, 571-590. Logan, G. D., Zbrodoff, N. J., & Williamson, J. (1984). Strategies in

BILINGUAL STROOP CONTROL

771

Tipper, S. P., & Cranston, M. (1985). Selective attention and priming: presented at the 1st European Congress of Psychology, Amsterdam. inhibitory and facilitatory effects of ignored primes. Quarterly Zbrodoff, N. J., & Logan, G. D. (1986). On the autonomy of mental Journal ofExperimental Psychology, 37A, 591-611. processes: A case study of arithmetic. Journal of Experimental Tipper, S. P., & Driver, J. (1988). Negative priming between pictures Psychology: General, 115, 118-130. and words in a selective attention task: Evidence for semantic processing of ignored stimuli. Memory & Cognition, 16, 64-70. Received October 3, 1989 Tzelgov, J., & Henik, A. (1989). The insensitivity of the semantic Revision received November 20, 1989 relatedness effect to surface differences and its implications. Paper Accepted December 29, 1989

Das könnte Ihnen auch gefallen