Cross-language activation and semantic judgements of translation ambiguous words among Chinese–English bilinguals

Ambiguous word translation among bilinguals

Translation ambiguity is the one-to-many mapping of word forms and meanings between a source and target language (Schwieter & Prior, 2020), often including homographs, homophones, homonymy, and morphological ambiguities (Prior et al., 2011). Tokowicz et al. (2002) investigated the correspondence between 562 English words translated into Dutch and German and found that 25% of them had multiple meanings in Dutch and 40% of them had multiple meanings in German. Other studies reporting on the correspondence between English words translated into Hebrew and Spanish found that 55% of the words translated into Hebrew (Smith et al., 2012) and 60% translated into Spanish had multiple meanings (Prior et al., 2007). Lyons (1977) classifies translation ambiguities into homographs and polysemants according to the degree of semantic relatedness between the words in question. A homograph is a word that has two or more meanings that are not semantically related (e.g., the word ‘adapt’, translated into Chinese, has two unrelated meanings of ‘适应’ and ‘改编’). A polysemant is a word that has two or more meanings that are semantically related (e.g., the word ‘branch’ can have two different, yet similar semantic meanings of ‘树枝’ and ‘分支机构’) (Li & Wang, 2013). Degani and Tokowicz (2010) divided translational ambiguities into form-ambiguous words, meaning-ambiguous words, and unambiguous words. Form-ambiguous words are those that have only one meaning in a given language, but have two or more related meanings in another language with different orthographic forms. Meaning-ambiguous words refer to those which have more than one meaning in a language, and when translated into another language, they correspond to different meanings. And unambiguous words refer to words that have only one meaning in both languages (X.-L. Wang et al., 2019). Y. Wang and Zhang (2017) further divided meaning-ambiguous words into semantically related meaning-ambiguous words and SU meaning-ambiguous words.

Primary and secondary translations

Prior research has examined bilinguals’ performance when processing translation ambiguous words and has found that responses are slower and less accurate compared to words that have only one translation equivalent (Basnight-Brown et al., 2018; Laxén & Lavaur, 2010; Tokowicz & Kroll, 2007). These effects emerge regardless of whether the target language is the more dominant L1 or a weaker L2 (Boada et al., 2013). Moreover, some translation equivalents in a target language can be considered more dominant (i.e., more frequently used) than others (Eddington & Tokowicz, 2013; Laxén & Lavaur, 2010; Prior et al., 2013; Y. Wang & Zhang, 2013). Laxén and Lavaur found that translation performance was faster and more accurate when words were presented with their primary translations compared to when they were presented with their secondary translations. When examining the semantic relatedness between the various translation possibilities, the results further showed that translation recognition was faster when a target word’s (e.g., bateau in French) translations were related in meaning (e.g., boat and ship) compared to when the target’s (e.g., argent) translations were unrelated (e.g., money and silver). Taken together, this growing body of research suggests that translating words that have multiple lexical possibilities is not only slower and less accurate, but also sensitive to factors, such as semantic relatedness and whether the translations are primary or secondary equivalents. For Chinese–English bilinguals, an additional complication of translation ambiguous words is that they often do not have a one-to-one semantic correspondence. For example, the Chinese words corresponding to the concept of ‘interest’ include ‘兴趣’ and ‘利息’, and these same characters can also be translated into multiple English words. Conversely, the English words corresponding to ‘垃圾’ include ‘garbage’, ‘litter’, ‘trash’, and ‘rubbish’.

The extent to which the primary and secondary translations are activated when bilinguals process target words remains unclear. Some researchers have compared bilinguals’ performance when translating from English into Spanish, German, Dutch, or Hebrew, and have found that primary translation equivalents are indeed activated. However, the number of meanings that the primary translation has plays a modulating role (Degani et al., 2016). Zhou et al. (2019) administered translation recognition tasks to Chinese–English bilinguals to explore the impact of L2 proficiency and sentence context on the processing of translation ambiguous words. The results showed that participants were slower and less accurate when recognizing translation ambiguous words compared to unambiguous words. Moreover, the response of recognizing semantically related translation ambiguous words was faster and more accurate than recognizing SU translation ambiguous words. Crucially, this effect was more pronounced for secondary compared to primary translations. Similar findings were reported by Y. Wang and Zhang (2013) who used the cross-language masking priming paradigm to examine Chinese–English bilinguals’ recognition of translation ambiguous words. The results showed that translation recognition was faster and more accurate for primary compared to secondary translation equivalents.

Implicit priming paradigm

The implicit priming paradigm has been employed to study non-target language activation during target language processing (Thierry & Wu, 2007). In this experimental design, materials are exclusively presented in one language, but there is an implicit connection to another language. Thierry and Wu asked Chinese–English bilinguals to perform semantic judgements about visually presented English word pairs. Half of these pairs had a Chinese translation pair containing a repeated character (e.g., ‘train–ham’ translated into Chinese is ‘火车–火腿’) and the other half of the word pairs’ translations did not have a repeated character (e.g., ‘apple–table’ is ‘苹果–桌子’). The results found that L2 words pairs whose translation in Chinese had a repeated first character elicited smaller N400 components, indicating that L1 translations were automatically and unconsciously activated when processing L2 words. In another study, Y.-J. Wu and Thierry (2010) examined whether Chinese–English bilinguals automatically activated sounds and spellings of L1 words during L2 word comprehension. The results showed that L1 sound, but not spelling, induced smaller N400s, suggesting that bilinguals automatically activate L1 sounds but not their orthographic representations, during L2 comprehension.

Most studies exploring cross-language activation focus on the degree of L1 activation during L2 processing (Hermans et al., 2011; Higby et al., 2019; Sunderman & Kroll, 2006). Fewer studies have addressed the reverse, and among these studies, there have been inconsistent findings: while some report that bilinguals activate L2 words when processing L1 words (Ang et al., 2016; R.-M. Wang et al., 2011), others find that L2 words are not activated when processing L1 words (Alvarez et al., 2003; Sebastian-Gallés et al., 2006). To our knowledge, there is no study to date which makes use of the implicit priming paradigm to investigate L2 activation during L1 word processing. Employing this experimental design is critical to our understanding of non-target language activation. Moreover, research in this area has mainly examined the activation of primary translation equivalents with little attention to secondary translations. Consequently, it is unclear as to whether secondary translations are automatically activated. The present study will explore these important issues.

Method

Materials and design

Experiment 1 has a single factor within-subject design in which the type of Chinese translation word condition (primary Chinese translation with a sound repetition vs. secondary Chinese translation with a sound repetition vs. SU) is the independent variable. The experiment also contains a semantically related condition as a control. The dependent variables are reaction times (RTs) and accuracy.

The experimental materials are 136 pairs of English words, with each pair including a prime and a target word (see Appendix 1). Among these words, there were 34 pairs in each of the following conditions: primary Chinese translation sound repetition (hereinafter referred to as PSR), secondary Chinese translation sound repetition (SSR), SU, and semantic-related (SR). In the PSR condition, the primary Chinese translation of the target English word and the Chinese translation of the prime English word had initial sound repetitions and were not semantically related. For example, the prime ‘prediction’ and the target word ‘bathroom’ in Chinese are ‘预测’ (‘yùcè’) and ‘浴室’ (‘yùshì’). In the SSR condition, the secondary Chinese translations of the target and primes had an initial sound repetition and again, were not semantically related. For example, the prime ‘history’ and the target word ‘interest’, when translated into Chinese is, respectively, ‘历史’, ‘利息’, the pronunciation of ‘历’ and ‘利’ are both ‘lì’. The primary translation of ‘interest’ in Chinese is ‘兴趣’ (‘xìngqù’) and the secondary translations are ‘历史’ (‘lìshǐ’) and ‘利息’ (‘lìxī’). In the SU condition, the prime and target word were SU and had no sound repetition when translated into Chinese. For example, the prime ‘stone’ and target word ‘clothes’ when translated into Chinese are ‘石头’ (‘shítou’) and ‘衣服’ (‘yīfú’), respectively. In the SR condition, which was used as a control condition, the prime and target words were semantically related and had no sound repetition when translated into Chinese. For example, the prime ‘winter’ and the target word ‘cold’ are ‘冬天’ (‘dōngtiān’) and ‘寒冷’ (‘hánlěng’), respectively.

Before the experiment, we evaluated the semantic relatedness and translation accuracy of all experimental materials, along with the primary and secondary Chinese translations of the target words in the SSR condition. Word frequency and length were matched and can be seen in Table 1. According to the MCword frequency database and t-tests, word length and frequencies of the experimental words under the three conditions were matched, ps > .05. Twenty college students from the same research population with a score of 425 or more on the College English Test-6 (CET-6) were selected to determine the Chinese primary translation equivalents and secondary translation equivalents of 100 English words. Based on the Chinese primary and secondary translations that were consistently identified by more than half of the subjects and on a database of Chinese translation norms for English words (Wen & van Heuven, 2017), the target words for the SSR condition were determined. Twenty-two additional college students with a score of 425 or more on the CET-6 were selected to evaluate the semantic relatedness and translation accuracy of these materials. Semantic relatedness was assessed using a 5-point Likert-type scale in which ‘1’ meant ‘not related at all’ and ‘5’ meant ‘very related’. The average score of all word pairs in the SR condition was higher than 4 and the average scores of those in the PSR, SSR, and SU conditions were lower than 2.5. T-tests revealed that the difference between the scores of PSR and SU was not significant, p = .23; the scores of SSR and SU were significantly different, p < .001; and the scores of PSR and SSR were significantly different, p < .001. Translation accuracy scores were also based on a 5-point Likert-type scale in which ‘1’ meant ‘not accurate at all’ and ‘5’ meant ‘very accurate’. Translation accuracy of all words was rated higher than 4.

OPEN IN VIEWER

Table 1.

Descriptive information about the materials in Experiment 1.

Condition		Semantic relatedness	Word length	Word frequency
PSR	Prime	1.76 ± .24	6.44 ± 1.86	68.40 ± 67.96
	Target		5.65 ± 1.57	88.65 ± 102.79
SSR	Prime	2.03 ± .27	5.26 ± 1.33	56.62 ± 65.18
	Target		5.09 ± 1.16	112.36 ± 73.53
SU	Prime	1.63 ± .21	5.24 ± 1.60	72.51 ± 65.96
	Target		6.00 ± 2.49	104.78 ± 98.87
SR (filler)	Prime	4.67 ± .24	5.29 ± 1.49	73.76 ± 56.60
	Target		5.44 ± 2.09	89.75 ± 72.50

Note. PSR: primary Chinese translation sound repetition; SSR: secondary Chinese translation sound repetition; SU: semantically unrelated; SR: semantic-related.

Results

We first eliminated RT data that were less than 200 ms and ±3 SDs from the overall mean RT and accuracy data that were ±3 SDs from the overall mean accuracy. We also removed one participant from the analyses whose accuracy rate was less than 75%. The mean and SD of RTs and accuracy for PSR, SSR, and SU can be seen in Table 2.

OPEN IN VIEWER

Table 2.

Experiment 1 mean reaction times (in ms) and accuracy (in %).

Condition	Reaction time	Accuracy
PSR	864.34 ± 119.80	95.41 ± 3.54
SSR	856.04 ± 117.87	98.09 ± 2.37
SU	869.42 ± 107.35	97.68 ± 3.06

Note. PSR: primary Chinese translation sound repetition; SSR: secondary Chinese translation sound repetition; SU: semantically unrelated.

For RTs, the results of the repeated-measures analysis of variance (ANOVA) showed that the main effect of condition did not reach significance, F(2,42) = .965, p = .389, η _p ² = .044. For accuracy, there was a main effect of condition, F(2,42) = 7.229, p = .002, η _p ² = .256. Post hoc analyses found that the accuracy of the PSR condition was significantly lower than the SSR condition, p = .001, d = .757, the accuracy of PSR was lower than SU, p = .008, d = .641, and the accuracy of SSR was no different than that of SU, p = .623 (see Figures 1 and 2).

OPEN IN VIEWER

Figure 1.

Experiment 1 accuracy rates (in %) per condition.

OPEN IN VIEWER

Figure 2.

Violin graphs of Experiment 1 accuracy rates (in %) per condition.

Discussion

The results from Experiment 1 found no significant differences in RTs between the PSR and SU conditions. However, in terms of accuracy, responses in the PSR condition had more errors than in the SU condition, indicating that Chinese–English bilinguals activate sounds found in the L1 translations of L2 words when judging their semantic relatedness (see also Y.-J. Wu & Thierry, 2010). Further analyses demonstrated that these sounds were found in the primary, but not secondary, Chinese translations. These patterns support the reordered access model which holds that the meanings of ambiguous words are activated in sequence according to their relative frequency and that prior context will affect the lexical stage of reading (Duffy et al., 1988). Given that the frequency of use of primary translation words is generally higher than that of secondary translations, the corresponding L1 primary translations appear to be more readily activated and accessible.

The results of Experiment 1 also showed that the cross-linguistic activation of L1 sounds interfered with L2 semantic judgements as reflected by lower accuracy in the PSR condition compared to the SU condition. These findings are consistent with a study by Friesen and Jared (2012) in which cross-language semantic activation was examined among French–English and English–French bilinguals. In their study, the bilinguals performed a category recognition task on word pairs containing both French and English words. The participants’ responses were slower and less accurate when they judged homophones between languages, indicating that phonology was involved in the activation of cross-language semantics.

In addition to Friesen and Jared (2012), other studies have reported interference of behavioural performance by L1 activation during L2 processing. For instance, T. Zhang et al. (2011) found that Chinese–English bilinguals responded faster to English word pairs whose Chinese translations repeated the first morpheme compared to translations with no repeated morpheme. One possible explanation for this difference is that lexical decisions take into account orthographic information and, thus, a repeated morpheme likely facilitates target recognition. These findings from these studies are supported by our results from Experiment 1. In another study, W. Zhang et al. (2023) recorded event-related brain potentials among Chinese–English bilinguals who engaged in an implicit translation-priming task involving L2 (English) word pairs with negative or positive affective valence. The results showed that negative L2 words induced a refractory period during which cross-language lexical access was blocked. In other words, when the prime word was negative, the access to L1 translation of the target word was blocked and thus relatedness judgements were accelerated if the two words were not semantically related. We should note that a limitation of our study is that it did not account for the emotional valence of words, particularly prime words, across conditions. In light of findings from the work of W. Zhang et al. (2023), future work should control for emotional valence of primes and targets in the experimental design.

Due to material limitations, the cross-language activation from English to Chinese examined in Experiment 1 was conducted under semantically independent conditions, while cross-language activation under semantically relevant conditions was not investigated. Therefore, in Experiment 2, we explored the activation of non-target languages under semantically relevant conditions.

Method

Materials and design

Experiment 2 uses a single factor within-subject design in which the type of Chinese translation word condition (semantically related condition with the same Chinese translation [‘SSC’] vs. semantically related condition with different Chinese translations [‘SDC’] vs. SU condition with different Chinese translations [‘SU’]) is the independent variable. The dependent variables are RTs and accuracy.

The materials included 180 pairs of English words, with each pair consisting of a prime and a target word (see Appendix 2). Among the pairs, there were 45 pairs of words in the SSC condition, 45 pairs of words in the SDC condition, and 90 pairs of words in the SU condition (i.e., 45 pairs for experimental comparison and 45 pairs as filler items). The prime in the SSC condition was semantically related to the target word and the two words shared the same translation. For example, ‘disturb’ and ‘bother’ are both ‘打扰’ (‘dǎrǎo’). In the SDC condition, the prime and target word were semantically related and when translated into Chinese, they have two different translations. For example, ‘argue’ and ‘fight’ are translated into Chinese as ‘争吵’ (‘zhēngchǎo’) and ‘打架’ (‘dǎjià’), respectively. In the SU condition, the prime and target word were SU and when translated into Chinese, they are different words. For instance, ‘wife’ and ‘city’ are translated into Chinese as ‘妻子’ (‘qīzi’) and ‘城市’ (‘chéngshì’), respectively.

Before the experiment, we evaluated the semantic relatedness and translation accuracy of all experimental materials. Word frequency and length were matched and can be seen in Table 3. According to the MCword frequency database and t-tests, word length and frequencies of the experimental words under the three conditions were matched, ps > .05. Seventeen age-matched students from the same university with a score of 425 or more on the CET-6 were selected to evaluate the semantic relatedness and translation accuracy of all materials. Moreover, after Experiment 1, we realized that it may have been a limitation to not have examined participants’ familiarity with the materials, and as such, before Experiment 2, we asked the age-matched participants to also make these judgements. Semantic relatedness was assessed on a 5-point Likert-type scale in which ‘1’ meant ‘not related at all’ and ‘5’ meant ‘very related’. The scores of word pairs in the SSC and SDC conditions were rated higher than 3.5 points, and the scores of word pairs in the SU condition were rated lower than 2.5 points: The difference between the scores of SSC and SDC was not significant, p = .076, but the scores between the SSC and SU conditions were significantly different, p < .001, and the scores between the SDC and SU conditions were significantly different, p < .001. Familiarity scores were also rated on a 5-point Likert-type scale in which ‘1’ meant ‘not familiar at all’ and ‘5’ meant ‘very familiar’. The result showed that familiarity of all words was higher than 4 points and the difference in familiarity between the conditions was not significant, ps > .05. Translation accuracy ratings, also given on a 5-point Likert-type scale, showed that all words were rated higher than 4 points, and the difference in translation accuracy between the conditions was not significant, ps > .05.

OPEN IN VIEWER

Table 3.

Descriptive information about materials in Experiment 2.

		Semantic relatedness	Familiarity	Word length	Word frequency
SSC	Prime	4.64 ± .29	4.96 ± .11	6.49 ± 1.85	65.86 ± 96.81
	Target		4.97 ± .06	6.69 ± 1.83	53.22 ± 55.95
SDC	Prime	4.49 ± .33	4.96 ± .10	6.76 ± 1.98	44.62 ± 44.02
	Target		4.90 ± .20	6.42 ± 1.88	72.09 ± 77.87
SU	Prime	1.58 ± .33	4.89 ± .19	6.58 ± 2.08	55.47 ± 48.19
	Target		4.94 ± .13	6.62 ± 1.96	62.48 ± 79.92

Note. SSC: semantically related condition with the same Chinese translation; SDC: semantically related condition with different Chinese translations; SU: semantically unrelated.

Results

We first eliminated RT data that were less than 200 ms and ±3 SDs from the overall mean RT and accuracy data that were ±3 SDs from the overall mean accuracy. The mean and SD of RTs and accuracy for the SSC, SDC, and SU conditions can be seen in Table 4.

OPEN IN VIEWER

Table 4.

Experiment 2 mean reaction times (in ms) and accuracy (in %).

Condition	Reaction time	Accuracy
SSC	752.77 ± 119.22	95.93 ± 3.49
SDC	832.20 ± 124.45	89.87 ± 7.76
SU	917.27 ± 129.05	90.41 ± 7.39

Note. SSC: semantically related condition with the same Chinese translation; SDC: semantically related condition with different Chinese translations; SU: semantically unrelated.

For RTs, the results of the repeated-measures ANOVA showed a significant main effect of condition, F(2,44) = 66.11, p < .001, η _p ² = .75. Post hoc analyses demonstrated that: RTs in the SSC condition were significantly faster than the SDC condition, p < .001, d = –.65; RTs in SSC were significantly faster than SU, p < .001, d = –1.314; and RTs in SDC were significantly faster than SU, p < .001, d = –.66 (see Figures 3 and 4).

OPEN IN VIEWER

Figure 3.

Experiment 2 reaction times per condition.

OPEN IN VIEWER

Figure 4.

Violin graphs of Experiment 2 reaction times per condition.

For accuracy, the results revealed a significant main effect of condition, F(2,44) = 8.647, p = .001, η _p ² = .282. Post hoc analyses showed that accuracy in the SSC condition was significantly higher than both the SDC condition, p = .001, d = 1.01, and the SU condition, p = .002, d = .928. There was no significant difference in accuracy between SDC and SU, p = .76. These results can be seen in Figures 5 and 6.

OPEN IN VIEWER

Figure 5.

Experiment 2 accuracy rates (in %) per condition.

OPEN IN VIEWER

Figure 6.

Violin graphs of Experiment 2 accuracy rates (in %) per condition.

Discussion

The results of Experiment 2 showed that the SSC condition elicited faster RTs and more accurate responses compared to the SU condition. This suggests that under semantically related conditions, the implicit repetition of Chinese translations improves L2 judgements and that L1 translation equivalents are automatically activated when processing L2 words. This result supports the hypothesis that non-target language activation occurs during target language processing (Degani et al., 2018; Sunderman & Kroll, 2006; Talamas et al., 1999; R.-M. Wang et al., 2011). Meyer and Schvaneveldt (1971) first proposed the concept of the semantic priming effect, an observation in which semantically related words can be understood more quickly and accurately. This effect appears not only within a language but also between languages. The results of Experiment 2 show that responses in the SSC and SDC conditions were significantly faster than in the SU condition, reflecting a semantic priming effect that is consistent with previous studies (Neely, 1991; Sperber et al., 1979).

The results from Experiment 2 also found that under semantically related conditions, implicit L1 translation repetition facilitated the judgement of L2 semantic relatedness. The reader will recall in Experiment 1 that under SU conditions, the accuracy of bilinguals’ judgements was significantly lower when Chinese translation equivalents had repeated initial sounds compared to when Chinese translations had no repeated initial sounds. Taken together, the findings from Experiments 1 and 2 suggest that Chinese–English bilinguals will automatically activate Chinese when processing English, and that semantic relatedness has a moderating effect on this process: in SU conditions, the implicit repetition of Chinese translation equivalents interferes with semantic relatedness judgements of English word pairs, whereas under semantically related conditions, the implicit repetition of Chinese translation equivalents assists such judgements. The question that remains to be addressed is whether these patterns also emerge when L2 is the non-target language. Experiment 3 will examine this possibility.

Method

Materials and design

Experiment 3 uses a 2 × 2 within-subject design in which the independent variables are semantic relatedness (semantically related vs. SU) and English translation repetition type (English translation repeated vs. English translation non-repeated). Thus, the four conditions are the semantically related English translation repeated condition (‘SR’), the semantically related English translation non-repeated condition (‘SN’), the SU English translation repeated condition (‘UR’), and the SU English translation non-repeated condition (‘UN’). The dependent variables are RT and accuracy.

The experimental materials were 120 pairs of Chinese words consisting of 30 pairs of words in SR, 30 pairs for the SN condition, 30 pairs in the UR condition, and 30 pairs in the UN condition (see Appendix 3). In SR, the prime and target words were semantically related and when translated into English, they both had the same equivalent (e.g., the prime ‘号码’ and the target word ‘数字’ are both ‘number’ in English). In SN, the prime and target words were also semantically related but when translated into English, they have different translations (e.g., the prime ‘科学’ [‘science’] and the target word ‘调查’ [‘research’]). In UR, the primes and target words were SU, but when translated into English, they have the same translation equivalent (e.g., the prime ‘火车’ and the target word ‘训练’ are both ‘train’ in English). In UN, the prime and target words were SU, but when translated into English, they have different translations (e.g., the prime ‘世界’ (‘world’) and the target word ‘比较’ (‘compare’).

Before the experiment, we evaluated all the experimental materials for word frequency. As shown in Table 5, according to the MCword frequency database and t-tests, there was no significant difference in word frequency for priming and target words in the various conditions, ps > .05. We also assessed the materials in terms of semantic relatedness and familiarity of the English translated words among a group of age-matched students whose CET-6 scores were 425 or more. These 22 participants provided their ratings on 5-point Likert-type scales. For semantic relatedness (‘1’ = ‘not related at all’, ‘5’ = ‘very related’), the scores of all word pairs in the two semantically related conditions were higher than 3.5 points, and the scores of all word pairs in the two SU conditions were lower than 2.5 points. The difference between the SR and SN conditions was not significant, p > .05. The scores between UR and UN, between SR and UR, and between SN and UN were all significantly different ps < .001. For familiarity (‘1’ = not familiar at all, ‘5’ = very familiar), the results showed that the familiarity of all words was higher than 4 points and the difference in familiarity between the conditions was not significant, ps > .05. Moreover, we added a concreteness judgement to the materials of Experiment 3. For the concreteness of Chinese words (‘1’ = not concrete at all, ‘5’ = very concrete), the results showed no differences between the various conditions, ps > .05.

OPEN IN VIEWER

Table 5.

Description information about materials in Experiment 3.

		Semantic relatedness	Familiarity	Concreteness	Word frequency (%)
SR	Prime	4.43 ± .22	4.69 ± .24	3.26 ± .78	.014 ± .026
	Target		4.69 ± .24	3.15 ± .74	.012 ± .017
SN	Prime	4.43 ± .23	4.79 ± .21	3.26 ± .67	.014 ± .019
	Target		4.79 ± .23	3.13 ± .83	.012 ± .013
UR	Prime	2.34 ± .37	4.70 ± .36	3.36 ± 1.00	.013 ± .032
	Target		4.70 ± .36	3.07 ± .98	.012 ± .021
UN	Prime	1.93 ± .25	4.74 ± .22	3.41 ± .03	.015 ± .024
	Target		4.68 ± .23	3.06 ± .97	.014 ± .016

Note. SR: semantic-related; SN: semantically related English translation non-repeated condition; UR: semantically unrelated English translation repeated condition; UN: semantically unrelated English translation non-repeated condition.

Procedure

The experiment was carried out on computers using E-Prime 2.0 software and included 20 practice trials and 120 formal trials. The experimental materials were presented randomly. In each trial, a fixation point (+) appeared for 500 ms in the centre of the screen followed by a blank screen of 500 ms. A prime in Chinese then appeared for 500 ms followed by another blank screen of 500 ms. A target word in Chinese appeared and participants were required to quickly and accurately judge whether the prime and the target word were semantically related by pressing the ‘f’ key for ‘related’ and the ‘j’ key for ‘unrelated’. If a participant did not respond within 2500 ms, the screen went blank, and the next trial began. Response keys were counterbalanced across participants and the entire experiment lasted around 15 minutes.

Results

We first eliminated RT data that were less than 200 ms and ±3 SDs from the overall mean RT and accuracy data that were ±3 SDs from the overall mean accuracy. We also removed one participant from the analyses whose accuracy rate was less than 65%. The mean and SD of RTs and accuracy rates for the SR, SN, UR, and UN conditions can be seen in Table 6.

OPEN IN VIEWER

Table 6.

Experiment 3 mean reaction times (in ms) and accuracy (in %).

Condition	Reaction time	Accuracy
SR	723.03 ± 154.00	92.72 ± 7.42
SN	729.11 ± 154.38	89.25 ± 7.19
UR	816.59 ± 201.59	82.06 ± 14.17
UN	793.46 ± 186.37	90.84 ± 9.95

Note. SR: semantic-related; SN: semantically related English translation non-repeated condition; UR: semantically unrelated English translation repeated condition; UN: semantically unrelated English translation non-repeated condition.

For RTs, the results of the ANOVA showed that the main effect of semantic relatedness was significant, F(1,31) = 21.058, p < .001, η _p ² = .405, such that, RTs in semantically related conditions were faster than in SU conditions. The main effect of English translation repetition versus non-repetition was not significant, F(1,31) = 1.701, p = .202, η _p ² = .052, nor was its interaction with semantic relatedness, F(1,31) = 2.773, p = .106, η _p ² = .082.

For accuracy, there was a main effect of semantic relatedness, F(1,31) = 4.688, p = .038, η _p ² = .131. Specifically, the accuracy rate in the semantically related condition was higher than in the SU condition. There was also a significant main effect of English translation repetition, F(1,31) = 5.121, p = .031, η _p ² = .142, such that the accuracy of the English translation repetition condition was lower than that of the non-repetition condition. Moreover, the interaction between semantic relatedness and English translation repetition was significant, F(1,31) = 14.054, p = .001, η _p ² = .312. Post hoc analyses showed that in semantically related conditions, accuracy in SR was marginally higher than in SN, F(1,31) = 3.64, p = .066. For SU conditions, UN was significantly more accurate than UR, F(1,31) = 16.11, p < .001. In repetition conditions, accuracy in SR was significantly higher than in UR, F(1,31) = 12.98, p = .001. In non-repetition conditions, there was no significant difference in accuracy between SN and UN, p = .496. These findings are shown in Figures 7 and 8.

OPEN IN VIEWER

Figure 7.

Experiment 3 accuracy (in %) per condition.

OPEN IN VIEWER

Figure 8.

Violin graphs of Experiment 3 accuracy (in %) per condition.

Discussion

Experiment 3 explored cross-language activation of corresponding L2 translation equivalents during L1 word processing in semantically related and unrelated contexts. The implicit priming paradigm was used to create conditions with repeated and non-repeated English translations. The results revealed an expected semantic priming effect: judging L1 word pairs that were semantically related was faster and more accurate than unrelated pairs. Moreover, in semantically related conditions, the participants were more accurate judging L1 word pairs when their translation equivalents were the same. However, in SU conditions, participants were less accurate in judging word pairs when their translations were the same. This implies that non-target language translation repetition promotes the processing of semantically related L1 word pairs but interferes with the processing of SU word pairs. It also suggests that when performing exclusively in their L1, bilinguals automatically activate L2 translation equivalents.

Cross-language activation of primary and secondary translations

In line with previous findings (Y.-J. Wu & Thierry, 2010), Experiment 1 found that when processing L2 words, Chinese–English bilinguals activated their L1 primary translation equivalents but not their secondary translations. According to the DCFM, word meanings are divided across multiple nodes that are shared with translation equivalents. Word translation is faster when more nodes are shared between translation equivalents (Laxén & Lavaur, 2010). In the context of the model, the results of Experiment 1 demonstrate the integrated nature of semantic nodes between L2 target words and their primary L1 translations. Moreover, the detection of cross-language activation of primary translations was based on implicit sound repetition. Similar findings were reported in an eyetracking study by Shook and Marian (2019) in which Spanish–English bilinguals heard words in their L2 English and were asked to click on the corresponding picture from a set of pictures. The results of their study showed longer and more frequent eye fixations on pictures whose translations had phonological similarity with the L2 target word’s translation. Although Shook and Marian’s study explored cross-language activation arising when processing auditorily presented words, the present study examined visual word processing and provides additional evidence that words processed in single-language contexts activate their translation equivalents, with further spreading activation to phonological competitors not implicitly seen.

Effects of semantic relatedness on non-target language activation

Among the findings from Experiment 2, was the result that that semantic relatedness regulated the influence of non-target language (L1) activation on target language (L2) processing. Although the role of semantic relatedness in cross-language activation has been given some attention in previous studies, a study by Y. Zhang et al. (2020) investigated the effects of semantic relatedness on L2 word learning. In the study, Chinese–English bilinguals learned three types of English pseudowords: the first type had two semantically related Chinese translation equivalents; the second type had two SU Chinese translation equivalents; and the third type had only one Chinese translation equivalent. Analyses on electrophysiological data revealed that for pseudowords with two semantically related L1 translations, the second translation equivalent learned induced larger N400 components compared to learning the first translation equivalent. Contrarily, for pseudowords with two SU L1 translations, the second translation learned was associated with larger Late Positive Component (LPCs) compared to the first. Zhang et al.’s findings indicate that learning a secondary translation equivalent is facilitated when it is semantically related to the primary translation equivalent and hampered when it is unrelated. In the present study, our results from Experiment 2 showed that when prime and target words were semantically related, the implicit repetition of the non-target language translation equivalents assisted the processing of the target language, and under SU conditions, such implicit repetition interfered with target language processing.

The bidirectional nature of cross-language activation among bilinguals

To our knowledge, the results of the present study are the first to show that when processing translation ambiguous words in a monolingual context, cross-language activation is bidirectional. Bilinguals’ language systems, and their semantic, phonetic, and orthographic representations overlap in distinct ways. Due to the connective, dynamic nature between the two languages, when bilinguals process the target language, various representations of the non-target language are also activated (Zhao & Mo, 2010). Our results suggest that regardless of whether the non-target language is the more-dominant L1 or the weaker L2, cross-language activation of translation equivalents will occur. These findings align with a recent speech production study by Liu et al. (2020). The results from their modified picture naming priming experiment found that activation of non-target language phonology occurred regardless of whether production was in the L1 or L2. The current study offers similar evidence for the bidirectional nature of cross-language activation in language comprehension, and the findings that facilitation and interference effects of translation repetition in semantically related and unrelated conditions, respectively, are consistent with previous studies of L1 activation during L2 processing (Qu, 2019; H. Zhang et al., 2012). Finally, we found a stable semantic priming effect in both L1 and L2 monolingual contexts such that performance in semantically related conditions was significantly faster and more accurate than in unrelated conditions.