Abstract

A Stroop ink-colour naming task was used to examine interference and facilitation effects in 16 Cantonese-English and 16 Mandarin-English bilinguals. The critical question of this experiment is whether the Stroop effects will be the same for within- and cross-language conditions. Participants named ink-colours both in L1 Chinese (in one block) and in L2 English (in another block). The related words were congruent, incongruent from the response set, or incongruent not in the response set, and there were also matched control words. The words were either in Chinese or in English, and so there were both within-language and cross-language conditions in the experiment. If bilinguals are able to ‘restrict’ processing to the language to-be-produced, there should be no cross-language effects at all, but previous researches have shown cross-language Stroop effects. For Cantonese speakers, the experiment found greater within- than cross-language facilitation and interference, whereas for Mandarin speakers the pattern was more mixed, with some cases of greater cross- than within-language interference effects. The results are interpreted in terms of the Revised Hierarchical Model, and suggest that L2 proficiency levels may not be the only factor that determines the pattern of Stroop facilitation and interference effects, but that language experience of the bilinguals may also play an important role.

Stroop facilitation and interference effects in Chinese-English bilinguals:

Effects of language dominance and response set membership

Introduction

With ever increasing numbers of people becoming bilingual, the question of how bilinguals can control two languages has been the subject of much research. Bilinguals can speak two different languages and so have words in each language to express the same concept. Words in both their first language (L1) and second language (L2) will be activated when speaking and/or when listening and reading (Desmet & Duyck, 2007). Traditionally, it is believed that when a bilingual read a word in their first or second language, their conceptual system activates the lexical representations of words in both languages (Costa, Santesteban, & Ivanova, 2006). Therefore, this parallel activation from semantics can present competition between lexical representations (Wilshire, Keall, Stuart & O’Donnell, 2007), and bilinguals must ignore or inhibit the unwanted representations to make sure only the correct word is produced in the required language. If bilinguals are able to ‘restrict’ processing to the language to-be-produced, there should be no cross-language effects at all. But, previous researches have shown that there are cross-language Stroop effects (e.g. Chen & Ho, 1986; Goldfarb & Tzelgov, 2007; Lee & Chan, 2000; Okuniewska, 2007; Rosselli et al., 2002; Tzelgov, Henik & Leiser, 1990; Zied et al., 2004). Therefore, the critical question is whether Stroop facilitation and interference effects will be the same for within- and cross-language conditions and the present study is designed to examine, in the same experiment, whether there are larger within- than cross-language (a) facilitation effects, (b) interference effects from the same response set, and (c) interference effects from non-response set words.

The Stroop task is a popular method to study language processing of bilinguals (e.g. Chen & Ho, 1986; Goldfarb & Tzelgov, 2007; Lee & Chan, 2000; Okuniewska, 2007; Rosselli et al., 2002; Tzelgov et al., 1990; Zied et al., 2004). John Ridley Stoop developed this task in 1935 (Stroop, 1935). In the basic version of this task, participants are required to name the ink colour of printed words or neutral control words.

There are three key results from the Stroop task that form the basis for this study.  First, there is the classic Stroop interference effect.  Colour naming times are slower when the word is an incongruent colour name (e.g., the word red printed in blue). This interference effect has been replicated in a large number of studies.  Second, there is the Stroop facilitation effect. Colour naming times are faster when the word is congruent with the colour (e.g., the word blue printed in blue).  Typically, the Stroop interference effect is larger than the facilitation effect (MacLeod, 1991).  Third, there is a response set membership effect. When the incongruent colour words are from the same set of colours used as targets, the Stroop interference effect is larger than when they are colour names not in the response set.  For example, when naming only the four colours blue, green, red and yellow, the word blue is in the response set but the words brown and pink are not.  The word blue printed in green produces a larger interference effect than the word brown in green (e.g. Klein, 1964; Lamers, Roelofs & Rabeling-Keus, 2010). There is also an effect of response set size. It has been reported that the interference effect is larger when there are more response options (e.g. Ray, 1974; Lamers et al., 2010), but this will not be investigated in the present study as a major part of the interference observed in the colour-words Stroop task is specific to the membership effect (e.g. Glaser & Glaser, 1989; Klein, 1964; Proctor, 1978). Lamers et al. (2010) argued that the response set membership effect is related to the selective allocation of attention to eligible responses, as only these eligible responses compete for selection.

Theories of the Stroop effect.

A number of theories have been proposed to explain performance in the Stroop task and especially the Stroop interference effect.  The relative speed of processing theory claims that words are read faster than colours are named. The speed difference in word reading and colour naming is critical when two potential responses are competing to be the actual response. The interference effect is interpreted as being the result of response competition, which involves a time cost to be resolved. The facilitation effect because the response to the ink colour and the word are the same, so the response can be produced by whichever arrives first at the response stage.  The theory of automaticity of reading claims that naming the ink colour needs more attentional resources than word reading. Word reading is claimed to be automatic and colour naming was much less automatic. The more automatic processing of words therefore interferes with the less automatic process. So, when the printed word is different than the colour of the ink, a conceptual competition occurs which results in delayed responses. This competition produces the Stroop interference effect. On the other hand, when the printed word is the same with the colour of the ink, no conceptual competition occurs which results in facilitation.

Stroop Effects in Bilinguals

To study language processing in bilinguals, researchers have used the Stroop task to compare the pattern of within-language and cross-language interference effects (e.g. Chen & Ho, 1986; Goldfarb & Tzelgov, 2007; Lee & Chan, 2000; Rosselli et al., 2002; Tzelgov et al., 1990; Zied et al., 2004). In the within-language condition, the language of the stimulus word is the same as the language required to respond.In the cross-language condition, the language of the word is different than the language required to respond. A number of studies have shown that the Stroop interference effect is greater in within- than in cross-language conditions (e.g. Dyer, 1971; Fang, Tzeng & Alva, 1981; Kiyak, 1982; Preston & Lambert, 1969).However, the difference between within- and cross-language interference effects has not been systematically studied as a function of naming language (L1 vs. L2) or by comparing facilitation and interference effects and so these factors, along with the examination of response-set effects, were studied in the current experiment.

Theories of Stroop effects in bilinguals

The pattern of within- and cross- language Stroop interference effects have been interpreted in terms of how bilinguals access and store lexical information. Several models have been proposed to explain how bilinguals link words in their two languages.

The word association model (Potter, So, Eckardt & Feldman, 1984) proposes that L2 words are associated with their translation equivalents in their first language, so that L2 words can only access semantics via their corresponding L1 words. L2 words have a two-step process to access the meaning: first, they need to translate them and then access meaning via L1-to-semantic connections. Previous studies showing that L1 colour words produce more interference than L2 colour words in both response languages are consistent with this word association model (Chen & Ho, 1986; Mägiste, 1984; Potter et al., 1984). However, recent research has suggested that this model mainly applies to beginner or intermediate learners of L2, who, because in their low levels of L2 proficiency, need to rely more on their L1 to access semantics.

The concept mediation model (Potter et al., 1984) proposes that two parallel words are linked with each other in the semantic level instead of the translation equivalence. That means bilinguals can access the semantic representation by their L1 or L2 directly, instead of accessing their semantic representation via the translation equivalence in their L1. Previous researches show that this model is explaining how words in two languages are linking together in the memory of the people with relatively high proficiency of L2 (Chen and Ho, 1986; Costa et al., 2006; Zied et al., 2004). It is because they have more exposure to their L2 and they are more skilled in their L2, which means their proficiency in L2 is getting higher and thus, they do not need to rely on their L1 to understand L2. They can get access to the meaning of the words in L2 directly.

The revised hierarchical model (RHM), proposed by Kroll (1993) and Kroll and Stewart (1994), combines the word association and concept mediation models; see Figure 1.

Figure 1.  The Revised Hierarchical Model.  Solid lines represent stronger connections and dotted lines represent weaker connections.

The RHM claims that the link from the L1 lexicon to concepts (i.e., the semantic system) is stronger than from L2 to concepts. (Concepts also have stronger links to L2 than to L1.) It also claims that the link from L2-to-L1 is stronger than from L1-to-L2. With increasing proficiency in L2, the L2 lexicon becomes stronger, but during initial stage of L2 learning, learners are assumed to understand L2 words via their L1 translations. Research evidence has led to a consensus that in the early stage of L2 acquisition, learners are initially dependent on lexical transfer from L1 to access the meanings of L2 words, but with increasing exposure to L2, bilinguals may undergo a developmental shift from lexical to conceptual processing of L2 words.

The Bilingual Interactive Activation, or BIA+ model (van Heuven & Dijkstra, 2010) is a connectionist model of bilingual word recognition, in which words in a target language are selected by means of a higher-level control system. This model contains a word identification system and a task or decision system. In the word identification system, there are four steps. First, the visual input activates sub-lexical orthographic representations that then activate orthographic whole-word lexical and sub-lexical phonological representations simultaneously. Then, they activate the semantic representations and language nodes. Language nodes indicate the membership of a particular language. Representations from different languages are activated and the word identification system collects all the information. After that the word identification system sends the information to the task or decision system. The decision system selects the response by weighing the relative activation of language nodes from the identification system and making decision based on the specific task goal.

The inhibitory control model (ICM) proposed by Green (1998) is a theory of language control in speech production. The ICM claims that bilinguals experience a continuous competition between lexical representations of their L1 and L2, which is always activated during reading, speaking, and listening (e.g. Costa, 2005; Spivey & Marian, 1999; van Heuven, Dijkstra & Grainger, 1998). An inhibitory control system limits the attention to the intended language and inhibits the unwanted language to resolve competition. When the participant is required to name a picture in L2, this will activate semantic representations that activate the relevant schemas in both L1 and L2. Suppression must then occur via language tags attached to each lexical representation. Because of the intention to name in L2, the task schema ‘name in L2’ enables participant to inhibit all words with L1 tags. Task schemas therefore activate words in the intended language and to suppress words in the unintended language.

Factors affecting Stroop interference

Bilinguals in different stages of L2 proficiency show different patterns of interference. However, what are the factors affecting the level of interference? Previous studies reported that language similarity, language experience, and the proficiency level of second language are factors affecting the level of interference.

Language Similarity. A number of studies show that patterns of cross- or within-language interference depend on the language similarity (Chen & Ho, 1986; Fang et al., 1981; Lee & Chan, 2000; Preston & Lambert, 1969; Sumiya & Healy, 2004). Preston and Lambert (1969) showed that less similarity of the colour names in two languages resulted in greater within- than cross-language interference. It was argued that when two languages are quite different, participants can easily inhibit the competing stimuli in a second language and concentrate on the colour-naming response demanded by the task. However, Costa et al. (2006) and Lee and Chan (2000) showed that language similarity has only a limited effect on Stroop interference. In the present study, all the participants are Chinese-English bilinguals, with languages and writing systems that are fundamentally different (Chen & Ho, 1986). English belongs to an alphabet system, while Chinese belongs to a logographic system. Furthermore, Lee and Chan (2000) examined whether Chinese and English orthographies would bring about different Strop effect but their results showed no difference in Stroop interference between Chinese and English. This evidence shows that language similarity may not significantly influence the interference creation in Chines-English bilinguals.

Language experience. Previous studies also show that language experience can affect the inhibitory control of two languages (Coderre, 2012; Verreyt, Woumans, Vandelanotte, Szmalec, & Duyck, 2016). Language experience is influenced by linguistic environment (e.g. Baus, Costa, & Carreiras, 2013; Linck, Kroll, & Sunderman, 2009), frequency of using L2, and experience of frequent switching between L1 and L2. People who live and are immersed in the L2 environment have greater exposure to L2 and more frequent to use the L2. The more frequent use of L2, the stronger the inhibitory control to regular the two languages. Some studies show that the higher the frequency of switching between L1 and L2 results in greater ability of executive control (Verreyt et al., 2016). With reference to the inhibitory control model, people who have a higher capacity of executive control may respond faster in the Stroop task or show a smaller interference effect. This is because participants who need to switch between L1 and L2 frequently gain more experience and training in controlling and inhibiting the unwanted language, therefore, they can have better executive control. Also, Coderre (2012) found that participants who can control the unwanted language better show a better ability to ignore the distracting word. Therefore, the level of interference in the Stroop task can be affected by the language experience of the participants.

Language Proficiency. Apart from the influence of language similarity and cognitive control ability, previous studies found that second language proficiency also affects Stroop interference (Chen & Ho, 1986; Costa et al., 2006; Mägiste, 1984; Zied et al., 2004). Participants with higher proficiency level in L2 usually show larger within-language interference than the cross-language interference (e.g., Costa et al., 2006; Zied et al., 2004). On the other hand, participants with lower level of proficiency in L2 usually show larger cross-language interference than the within-language interference (Chen & Ho, 1986; Mägiste, 1984). Moreover, Costa et al. (2006) found that people with lower L2 proficiency showed larger task-switch costs when switching to the first language than to the second language. Language proficiency is related to the ability of a person to control over a language (Hernandez & Li, 2007).

According to the Language Proficiency Hypothesis (Mägiste, 1984), when bilinguals perform the Stroop colour naming task, the names of colour words in the dominant language should create greater interference than words in the less proficient language. But the interference effect of the dominant language will be diminished when the second language proficiency level of bilinguals increased. Moreover, the interference effect created by their L1 and L2 should become equivalent when individuals are balanced bilinguals. Several studies support the Language Proficiency Hypothesis (Dalrymple-Alford, 1968; Mägiste, 1984; Preston & Lambert, 1969). Mägiste (1984) tested a group of German-Swedish bilinguals with different degrees of L2 proficiency, and showed that the degree of within- or cross-language interference depended on language proficiency, with larger cross-language interference for participants with higher level of L2 proficiency (Chen & Ho, 1986; Okuniewska, 2007). Tzelgov et al. (1990) tested bilinguals with different proficiency and age of acquisition of the second language, and confirmed that L2 proficiency influenced the degree of Stroop interference. These studies show that language similarity, experience, and proficiency can influence the degree of cross-language Stroop interference. In the present study, two groups of bilinguals were tested with similar level of L2 English proficiency. They were students studying in English, and living in England, who had high subjective ratings and formal assessments of English proficiency.

Previous studies of Chinese-English Bilinguals

Chinese and English have quite different writing systems. Chinese uses ideographic characters and English uses alphabetic words. There have been mixed results from studies of Stroop effects in Chinese-English bilinguals. Smith and Kirsner (1982) found that Chinese stimulus words produced less overall interference than English words, a result that is inconsistent with Language Proficiency Hypothesis. However, Chen (1999) found that there were no differences in Stroop interference between Chinese and English words. The majority of previous studies tested in Cantonese, but not Mandarin (Chen & Ho, 1986; Chen & Leung, 1989; Chen & Ng, 1989; Lee & Chan, 2000). Cantonese and Mandarin are the two most important dialects of Chinese. Cantonese is the official language of Hong Kong and Macau, and is also a common dialect in the southern part of mainland China. Mandarin is the official language of the People’s Republic of China and the Republic of China. Mandarin is also one of the four official languages of Singapore. Some of the vocabularies in Cantonese and Mandarin are the same or similar. However, these two types of Chinese are different fundamentally and these two dialects are mutually unintelligible (Zhang, 1998). In this present experiment, participants named the ink colour of words either in Cantonese or Mandarin, and the colour names differed in pronunciation between the two languages. The difference in pronunciation is like British English and American English for the same word ‘potato’. This difference in pronunciation may not significantly affect the language production process. Therefore, it is worth testing both Cantonese and Mandarin in this study to examine the generality of the results found.

Research Rationale

To maintain the homogeneity of participants, only native Cantonese speakers who come from Hong Kong were recruited. For the Mandarin speakers, only native Mandarin speakers who come from mainland China were included. Although mainland China and Hong Kong both belong to the People’s Republic of China, people in Hong Kong and mainland China have different language experience due to the historical reasons. Hong Kong is a Cantonese and English dominated society. Most of the people in Hong Kong communicate in Cantonese or English daily. Both Cantonese and English are used for instruction in primary schools in Hong Kong. When students enter secondary school, they use English more frequently, as more and more subjects are taught in English. When they get into the university, all the subjects are taught in English, except for Chinese-related subjects, such as Chinese and Chinese History. Therefore, people in Hong Kong will have had more exposure to English, and more experience switching between English and Chinese. Students from mainland China have communicated mostly in Mandarin. The instruction medium of most schools in China is Chinese, with English only taught during English lessons. Therefore, they will have had less frequent use of English than students in Hong Kong. As there are few places in the world that have such a special relationship as Hong Kong and China, it is worth seeing how those different language experiences affect Stroop interference.

There are only few studies that have examined both Stroop interference and facilitation effects in Chinese-English bilinguals, and the findings have been mixed. The present study seeks to examine within the same experiment, within-language and cross-language Stroop effects when both naming in L1 (Cantonese and Mandarin) and in L2 English. The present experiment examined naming in both L1 Chinese and L2 English. Many previous studies of cross- and within-language Stroop interference effects have tested only one naming language (generally L1).  It is possible that the differences between cross- and within-language interference is different for naming in L1 and in L2, especially if there are asymmetric connections between L1 and L2, as proposed by the revised hierarchical model (see Figure 1). The experiment is designed to examine whether there are larger within- than cross-language (a) facilitation effects, (b) interference effects from the same response set, and (c) interference effects from non-response set words. These effects will be compared unrelated control words that are matched to the colour names on word frequency, which is a methodological advance on studies that have used meaningless characters as the baseline.

The critical hypothesis tested is whether there will be larger within-language than cross-language Stroop facilitation and interference effects. This experiment will also examine interference effects from colour names that are not in the set of response words. In the cross-language conditions of previous studies, the stimulus words were necessarily not in the response set.  For example, when naming in Chinese, English stimulus words must not be spoken. In the present study, the use of incongruent colour names that are not in the response set will therefore enable an examination of the effects of response set for both naming in Chinese and English.

Method

Participants.  Thirty-two Chinese-English bilinguals participated in this study. They are native speakers of Cantonese or Mandarin and also fluent in English. Half had Cantonese as their first language, and half had Mandarin. The 16 Cantonese speakers were from Hong Kong and the 16 Mandarin speakers were from mainland China. There were 11 females and 5 males in each group. All participants were students of the University of Essex, studying in English, and living in England, and so it was assumed that their proficiency in English was reasonably high.

Age and educational level may affect the performance of participants in the Stroop Task (e.g. Brink &McDowd, 1999; Comalli, Wapner, & Werner,1962; Das, 1970; MacLeod, 1991; West, 1999).  The participants are aged 18 to 31 with at least university level of education. Both groups of participants are matched for age and educational level. The mean age of the Cantonese participants was 20.5 years (SD = 3.06) and the mean age of the Mandarin participants was 24.25 years (SD = 2.41). All the participants had normal colour vision and normal or corrected to normal vision. All were right-handed, and none reported any significant neurological and psychiatric problems (such as depression or substance abuse) or any other serious medical conditions. None reported needing any medication that could affect their brain functioning in the past or at this moment.

The English proficiency of the participants was assessed in three ways: (a) they provided their IELTS scores; (b) they provided self-ratings of their English proficiency for reading, writing, speaking and listening, and overall proficiency; and (c) they completed the LexTALE test (Lemhofer & Broersma, 2012), which is a lexical decision (i.e., word vs. nonword discrimination) task that provides an objective measure of English vocabulary knowledge that is highly correlated with general English proficiency and the validity is higher than self-rating. The means scores for these assessments are reported in Table 1.

Table 1.  L2 English proficiency levels of the participants tested.

Note 1: LexTALE scores of 60%-80% are equal to B2 grade at CEF level.

Note 2: In self-rating, 10 means they are an expert user and 1 means they are a beginner.

Stimulus materials. Four ink-colours were used: red, yellow, blue and green. These basic colours are easy to name. The colour words used were the names of these colours (which formed the response set) and also the words purple, black and white (which were the non-response set words). The words were presented in both Chinese and English. For each word, in each language, an unrelated control word matched for word frequency was selected using the data provided in SUBTLEX-CH (Cai & Brysbaert, 2010) and SUBTLEX-US (Brysbaert & New, 2009) for Chinese and English respectively. Table 2 shows the mean frequency of the colour and control words. Although the Chinese words were more frequent than the English words (244.6 vs. 103.2), the colour names and their respective control words were very closely matched. Appendix lists the colour words and their matched control words.

Table 2.  Mean word frequencies of the colour and control words used in the experiment.

Design. Five independent variables were manipulated in the experiment. (1) Group: Cantonese-English vs. Mandarin-English bilinguals. (2) Naming language: L1 Chinese vs. L2 English. (3) Language of stimulus printed words: Chinese vs. English. (4) Condition of related words: congruent (e.g. red in red) vs. incongruent in response set (e.g. blue in red) vs. incongruent not in response set (e.g. purple in red). (5) Related vs. matched control words for each condition.

Participants completed two blocks of ink-colour naming trials, one in L1 Chinese (either in Cantonese or in Mandarin, depending on group) and one in L2 English. Order of presentation of the naming tasks was counterbalanced. In both naming tasks, both Chinese and English words from all conditions were presented in a randomized order.  Therefore there were both within- and cross-language conditions in the experiment.

Procedure. Informed consent was obtained from each participant before the experiment. Participants performed two naming tasks: naming the ink colours in Chinese (either Mandarin or Cantonese), and naming in English.  Half of the participants in each group named first in Chinese and then in English, and half had the reverse order, so the order of the naming language was counterbalanced. In each task, there were a total of 144 experimental trials. For each naming task, there was first a familiarisation session that showed the colours to be used in the experiment and practice naming these with the target names. Then there were 12 practice trials presenting words in these colours.  Then there were the main experimental trials, which were divided into three blocks of 48 trials (although trials from all conditions were presented in each block).  Participants were instructed to verbally name the ink colour and to attempt to ignore the printed words, and to do this as quickly and as accurately as possible. Also, participants were informed that the ink colours must be one of the respond set colours (i.e. red, yellow, blue, or green).

The stimulus items were presented one at a time on a computer running a SuperLab programme. This presented the stimuli in a randomized order for each participant, and recorded ink-colour naming latencies. The participants were being a short break after each block to prevent fatigue, and the entire experiment took approx. 20 minutes to complete.

Results

Responses that were incorrect or where there were voice key failures (e.g., being stopped by heavy breathing or outside noise) were excluded. Naming times less than 250 ms or longer than 2500 ms were also excluded. The mean of non-excluded naming times in each condition was calculated for each participant. Given the complexity of the experiment, which manipulated five independent variables, the data were analysed by three four-way Analyses of Variance (ANOVAs), one for the congruency condition, one for incongruent from the response set, and one for incongruent not in the response set. Each of these 2x2x2x2 mixed ANOVAs had the between participant factor of Group (Cantonese vs. Mandarin), and the within participant factors of Response language (Chinese vs. English), Stimulus word language (Chinese vs. English words) and Relationship (Related vs. Control words).

(1) The congruent conditions.  Table 3 shows the mean naming times in each condition for this analysis, along with the facilitation effects found.

Table 3.  Mean naming times (in ms) in each condition for the analysis of congruency effects.

Note: *= Significant at 0.05 level; ns=Not significant at 0.05 level

The ANOVA produced significant main effects for naming language (F(1,30) = 6.12, p = 0.02) and relationship (F(1,30) = 19.64, p  < 0.0001).  Mean naming times were faster in Chinese than in English (696 vs. 729 ms) and in the related than the control conditions (693 vs. 731 ms), which showed a facilitation effect. There was no significant main effect of group; (F(1,30) = 1.36, p=0.25). The interaction between naming language and relationship was significant (F(1,30) = 19.74, p  < 0.0001), which showed that the facilitation effect was larger when naming in English than Chinese. The three-way interaction between naming language, stimulus words and relationship was significant (F(1,30) = 9.70, p < 0.005), which showed a larger within- than cross-language facilitation effect.

(2) The incongruent and in response set conditions.  Table 4 shows the mean naming times in each condition for this analysis, along with the interference effects found.

Table 4.  Mean naming times (in ms) in each condition for the analysis of interference effects for words in the response set.

Note: *= Significant at 0.05 level; ns=Not significant at 0.05 level

The ANOVA produced significant main effects for naming language (F(1,30) = 34.26, p < 0.00001) and relationship (F(1,30) = 45.52, p < 0.00001). Naming times were faster in Chinese than English (746 vs. 808 ms) and were slower with related than with control words (816 vs. 738 ms), showing a large overall Stroop interference effect (of 78 ms). The interaction between naming language and stimulus words was significant (F(1,30) = 5.42, p = 0.03). The four-way interaction was also significant (F(1,30) = 6.336, p = 0.02).

(3) The incongruent and not in response set conditions.  Table 5 shows the mean naming times in each condition for this analysis, along with the interference effects found.

Table 5.  Mean naming times (in ms) in each condition for the analysis of interference effects for words not in the response set.

Note: *= Significant at 0.05 level; ns=Not significant at 0.05 level

There were significant main effects for naming language (F(1,30) = 16.13, p < 0.001) and relationship (F(1,30) = 57.81, p < 0.00001). Mean naming times were faster in Chinese than English (729 vs. 782 ms), and slower in the related than the control conditions (777 vs. 733 ms), which showed a mean interference effect of 44ms. The interaction between naming language and stimuli words was significant (F(1,30) = 23.92, p < 0.0001).

Finally, the Stroop interference effects (calculated for each participant) were analysed to examine the effect of response set membership.  A 2x2x2x2 ANOVA was performed on the interference effects with the between participant factor of group, and the within participant factors of response language, language of stimulus words, and response set (in vs. not in response set).  Table 6 shows the Stroop interference effects in each condition.

Table 6.  The Stroop interference effects in each condition (in ms).

Note: *= Significant at 0.05 level; ns=Not significant at 0.05 level

There was a significant main effect of response set (F(1,30) = 8.24, p < 0.01); the mean Stroop interference effect was larger when the stimulus words were in the response set than when they were not in the response set (78 vs. 44 ms). The four-way interaction was also significant (F(1,30) = 5.92, p = 0.02).  This shows that the effect of response set membership on the Stroop interference effects was larger for the within- than the cross-language conditions for the Cantonese group, but showed the opposite pattern for the Mandarin group.

Discussion

The experiment replicated the three basic results from the Stroop ink-colour naming task: there was a facilitation effect from congruent words, a larger interference effect from incongruent words (e.g. MacLeod, 1991), and the interference effect was larger when the stimulus words were in the response set than when they were not (which is consistent with previous studies; e.g. Klein, 1964; Lamers et al., 2010).

The Stroop facilitation effect was larger for the within-language than for the cross-language conditions for both the Cantonese and Mandarin groups.  However, both groups also showed that the within-language facilitation effect was larger when they were naming English words in English than when they were naming Chinese words in Chinese. Congruency seemed to help naming in L2 more than when naming in the more dominant L1.

The Cantonese group showed larger within-language than cross-language interference effects from words in the response set.  When naming in Chinese, this was clear: there was a significant interference effect of 101 ms from Chinese words, but only a non-significant effect of 28 ms from English words. When naming in English, there was a significant interference effect from English words (137 ms) and a smaller but still significant effect from Chinese words (79ms). The Mandarin group showed a somewhat different pattern of effects: they showed interference effects from all conditions, which did not substantially differ for within- than cross-language conditions or for naming in Chinese and English (see Table 4).

The results from stimulus words not in the response set (where the interference effects were overall smaller) were somewhat more mixed. When both groups named in Chinese, they showed larger interference from Chinese than English words. However, when they named in English, neither group showed any consistent differences between within- and cross-language conditions.

The finding that, when naming in Cantonese, there was a greater within-language than cross-language interference effect is consistent with previous studies (e.g. Chen & Ho, 1986; Dyer, 1971; Fang et al., 1981; Kiyak, 1982; Preston & Lambert, 1969). In addition, they show a greater response-set effect in within-language than cross-language conditions. However, the group of Mandarin speakers sometimes showed greater cross-language interference than within-language interference. This is consistent with the study of Gerhand, Deregowski and McAllister (1995), but is inconsistent with other studies (and also with the results of the Cantonese speakers tested here).

Facilitation effects.

Participants generally named colours faster in the congruent than in the matched control condition. This facilitation effect was generally larger for within- than cross-language conditions, and is consistent with the theory of automaticity, where the semantic representation of the words and the name of the ink colour are simultaneously activated. It is because participants automatically recognise the printed word while naming the ink colours. In the congruent condition, the meaning of the word and the name of the word are the same and so there is no conflict between them, therefore, participants can respond faster as no inhibition of the unwanted name is required and naming can be based on whichever arrives first at the response stage (Bialystok, Craik & Luk, 2008; Marian, Blumenfeld, Mizrahi, Kania & Cordes, 2013).

For both groups, there was larger within- than cross-language facilitation when naming in Chinese.  When Cantonese speakers name in English, they showed only within-language facilitation; English words produced facilitation (71 ms) but Chinese words did not produce a significant effect (15 ms).  However, when Mandarin speakers named in L2 English, they showed a facilitation effect from both English words (89 ms) and Chinese words (74 ms).  This suggests that the Cantonese group were better able to maintain language-specific control. They appeared to be able to restrict processing only to the target language. It is possible that the Mandarin speakers had less experience switching between languages than the Cantonese speakers from the bilingual society of Hong Kong where greater language control would need to be exercised when speaking in one language only. This is consistent with the Inhibitory Control Model, as additional processing is required when stimulus and response language different (Marian et al., 2013).

The fact that participants in both groups were faster naming in Chinese than English is also consistent with the Inhibitory Control Model, which claims that suppression of L1 words is more difficult than suppression of L2 words, as L1 words have a higher resting-level of activation than L2 words. It is easier to activate L1 words but harder to inhibit them. This prediction is supported by a number of fMRI studies; processing a weaker language needs to activate additional brain regions or stronger activation in the same cortical networks than processing L1 (e.g. Briellmann et al., 2004; De Bleser et al., 2003; Hasegawa, Carpenter & Just, 2002; Hernandez & Meschyan, 2006; Rüschemeyer, Zysset & Friederici, 2006; Vingerhoets et al., 2003).

Stroop interference effects.

The experiment found that colour naming times are slower in incongruent than control conditions. This result is consistent with many previous studies of the Stroop interference effects (e.g. MacLeod, 1991; Wang, Fan, Liu & Cai, 2016). When participants name the ink colour of a word, they find it is difficult to ignore the incongruent colour word. The conflict between the word and the colour name produces a significant delay in naming time. With reference to the theory of automaticity, participants automatically recognise the printed word while naming the ink colours. When the printed word denotes a different colour from the colour of the ink, competition in both conceptual encoding and response selection occurs. They need to suppress the non-target answer and produce the correct answer. This process requires participants to resolve the conflict between the well-learned automatized reading response and the slower colour-naming response, which produces slower naming times.

The effect of response set membership on the size of the Stroop interference effect shows that a component of the interference effect is due to response selection (in addition to a contribution from conceptual conflict). Lamers et al. (2010) argue that the response set membership effect is related to selective allocation of attention to eligible responses. Participants in the present experiment were informed that the ink colours must be one of the respond set colour, namely red, yellow, blue, or green. Therefore, only these eligible words will compete for response selection. In the non-response set condition, smaller interference results due to the meaning of the words, and thus participants do not need to allocate attention to the meaning of words not in the response set. The finding that there is less interference from colour names in the non-response set has been found by a number of previous studies (e.g. Glaser & Glaser, 1989; Klein, 1964; Lamers et al., 2010; Proctor, 1978).

Cantonese vs. Mandarin speaking bilinguals

The experiment found that the two groups of Chinese-English bilinguals showed a different pattern of within- and cross- language interference effects. The Cantonese group generally showed larger within-language than cross-language interference effects. This result suggests that, both when naming in Chinese and in English, different-language words were somehow attended to less than same-language words. However, if these bilinguals had perfect control over which language to process, there would be no cross-language interference effects at all. Yet, when naming in English, the Cantonese group did show significant interference effects from Chinese words. It appeared that they were unable to prevent attention being applied to incongruent Chinese words, even though their orthography provided a clear cue that they were not target language words. It would therefore seem that they could not completely suppress the processing of the Chinese words, even though they produced an interference effect on target naming time.

The results from the Mandarin group were more variable.  First, the within-language interference effects they showed were generally smaller than those of the Cantonese group. Second, for incongruent words in the response set (which produce the ‘classic’ Stroop interference effect), when naming in Chinese, the Mandarin group showed significant within- and cross-language facilitation of a similar amount (56 vs. 76 ms). The Cantonese group showed only significant within-language interference when naming in Chinese. Third, for incongruent words in the response set, when naming in English, the Mandarin group showed significant cross-language interference (of 88 ms) and a smaller (and non-significant) within-language interference effect (of 59 ms). This result appears to support the word-association model rather than the concept-mediation model, as it seems that they are accessing L1 translation equivalents of stimulus L2 words. They may be translating L2 words into L1 in order to access their meaning, and so their L1 influenced their naming in L2 which produced greater cross- than within-language interference.

Taken together, the results of this study fit the Revised Hierarchical Model. This is effectively a combination of the word association and concept mediation models (De Groot, 1992, 1993, 1995). The relative influence of word-association or concept mediation processing varies at different stages of the development of L2 proficiency. The direct concept mediation of second language is acquired gradually. This may happen earlier for more familiar words or concepts (Dufour & Kroll, 1995). As L2 proficiency develops, there is a more from word-association to greater concept mediation, which may explain the results from the group of Mandarin speakers.

Although the Mandarin group have a relatively high proficiency in English, their results appear to be more consistent with the word-association model rather than concept-mediation model. This is inconsistent with those previous studies showing that the word association model is more applicable to people with low L2 proficiency. The Revised Hierarchical Model is more able to explain why the Mandarin participants, with relatively high proficiency in English, show larger cross- than the within-language interference. The model claims that connections between L1 and L2 are asymmetrical. The L2 word is very likely to activate the parallel L1 word, however, the activation of the L1 word is not automatically followed by an activation of the parallel L2 word (Kroll & Stewart, 1994; Tzelgov & Eben-Ezra, 1992). The results of the experiment show that the Cantonese group appears to be able to ‘ignore’ L2 words when they are naming in L1, but that they are unable to ‘ignore’ the more dominant L1 words when they are naming in L2.

When Mandarin participants named in Chinese, both Chinese and English words produced interference. When they named in English, Chinese words produced more interference than English words. These results may be interpreted with the Revised Hierarchical Model as their dominant L1 language is harder to suppress. When they are naming in L2, it is harder to suppress the activation of L1 words.

Mandarin participants show an inconsistent pattern with previous studies claiming that L2 proficiency level is the main factor that determines the connections between their L1, L2 and semantic representations. The results of the present experiment suggest that proficiency may not be the only important factor, and that the language experience of the participants may also be important. The Cantonese and Mandarin speakers had equivalently high levels of L2 English proficiency (see Table 1). For example, the Mandarin group had a mean IELT score of 6.38, which is in the B2 level in the Common European Framework of Reference for Languages (CEFR). However, the two groups learned English in quite different social and educational contexts. For the speakers of Mandarin in the mainland China, English was learned more as a ‘foreign’ language, taught explicitly in a classroom context, in a general cultural context that did not include wide exposure to English in society (although this may be changing).  In contrast, the Cantonese speakers learned English in Hong Kong, where both languages are prevalent in society. When growing up in Hong Kong, people will have been exposed to both languages from an early age in a wide variety of contexts (e.g. in general life, at school, and on television, radio and social media). As suggested earlier, in the Introduction when discussing language experience, it is possible that growing up immersed in a social environment with the experience of frequent exposure to L2 and frequent switching between L1 and L2, may have led to more efficient executive control over language use and to greater experience of interacting between two languages. These ideas are also consistent with the Inhibitory Control Model. Verreyt et al. (2016) found that bilinguals who switch between their L1 and L2 frequently have better executive control. Speakers with efficient control over their two languages may have a better ability to ignore distracting words. The Cantonese participants in this study could be considered to be efficient language users, and so show smaller cross-language Stroop effects, especially when they are naming in L1, as they can segregate two languages effectively. Previous research suggests that language experience may influence the language control of bilinguals, such that they have enhanced executive functioning (Bialystok, 2009; Bialystok et al., 2008).

Limitations, future directions of research, and conclusions of the study

The Cantonese participants tested were all Hong Kong, and no Cantonese speakers from other regions of China or from other countries were tested. This may be a limitation of the study.  Future research is needed to extend the generality of the results. A valuable avenue of future research will be to compare groups of Cantonese speakers who acquire language immersed in bilingual communities (such as in Hong Kong), where there is common exposure to both Cantonese and English, and those who have more limited exposure to English. It will also be useful to examine Chinese speakers with different second languages, in order to access the role of language (and especially orthographic) similarity.

In conclusion, this study found both within- and cross-language Stroop effects of facilitation, interference, and response set membership. The complex pattern of results found may be interpreted in terms of the Revised Hierarchical Model. The discussion of the results suggests that L2 proficiency levels alone may not be the only factor that determines the pattern of Stroop facilitation and interference effects, but that language experience of the bilinguals may also play an important role.