1. Introduction
Word frequency is one of the most important variables affecting lexical processing efficiency. Across languages and paradigms, high-frequency words are processed faster and more accurately than low-frequency words (for reviews, see Brysbaert et al., Reference Brysbaert, Mandera and Keuleers2018; Rayner, Reference Rayner1998). Notably, numerous studies have reported that word frequency effects are typically larger in second language (L2) than in first language (L1; e.g., Cop et al., Reference Cop, Keuleers, Drieghe and Duyck2015; Mor & Prior, Reference Mor and Prior2022). Although this pattern has been widely replicated, its underlying cognitive mechanisms remain debated. Two broad classes of accounts have been proposed to explain enlarged L2 word frequency effects. One attributes them to differences in lexical representations, whereas the other emphasizes that online co-activation of L1 and L2 lexical items introduces interference. Prior studies have mainly compared L1 and L2 within typologically similar alphabetic languages, making it difficult to disentangle these two accounts. Although a few studies have explored L2 word frequency effects in bilinguals with typologically distinct L1–L2 pairs (e.g., Mor & Prior, Reference Mor and Prior2020, Reference Mor and Prior2022; Sui et al., Reference Sui, Woumans, Duyck and Dirix2025), it remains unclear whether L1–L2 similarity modulates these effects to distinguish between the two accounts. The present study addresses this gap by examining whether L2 word frequency effects are modulated by L1–L2 similarity among L2 readers from alphabetic and Japanese L1 backgrounds.
1.1. Word frequency effect across native and L2 groups
Previous studies have shown that word frequency effects in L1 are modulated by language proficiency. Readers with lower language proficiency exhibit larger word frequency effects than skilled readers, as observed in both alphabetic languages (Ashby et al., Reference Ashby, Rayner and Clifton2005; Barnes et al., Reference Barnes, Kim, Tighe and Vorstius2017; Cop et al., Reference Cop, Keuleers, Drieghe and Duyck2015; Gordon et al., Reference Gordon, Moore, Choi, Hoedemaker and Lowder2020; Kuperman & Van Dyke, Reference Kuperman and Van Dyke2011, Reference Kuperman and Van Dyke2013; Lee et al., Reference Lee, Godwin, Blythe and Drieghe2025; Taylor & Perfetti, Reference Taylor and Perfetti2016) and Chinese (Guan & Fraundorf, Reference Guan and Fraundorf2020; Liu et al., Reference Liu, Wang, Yan, Paterson and Pagán2021; Zhang et al., Reference Zhang, Hu, Bian, Li, Zhang and Zang2022b). In alphabetic languages, word frequency effects are larger for younger children than older children (Ballot & Zesiger, Reference Ballot and Zesiger2024; Huestegge et al., Reference Huestegge, Radach, Corbic and Huestegge2009; Khelifi et al., Reference Khelifi, Sparrow and Casalis2019), in children compared to adults (Joseph et al., Reference Joseph, Nation and Liversedge2013; Khelifi et al., Reference Khelifi, Sparrow and Casalis2019; McArthur et al., Reference McArthur, Whitford and Joanisse2024; Tiffin-Richards & Schroeder, Reference Tiffin-Richards and Schroeder2015; Whitford & Joanisse, Reference Whitford and Joanisse2018), and in older adults compared to younger adults (Kliegl et al., Reference Kliegl, Grabner, Rolfs and Engbert2004; McGowan et al., Reference McGowan, White and Paterson2015; Rayner et al., Reference Rayner, Reichle, Stroud, Williams and Pollatsek2006; Warrington et al., Reference Warrington, McGowan, Paterson and White2018; Whitford & Titone, Reference Whitford and Titone2017, Reference Whitford and Titone2019; Zhang et al., Reference Zhang, Warrington, Li, Pagán, Paterson, White and McGowan2022a; but see Allen et al., Reference Allen, Bucur, Grabbe, Work and Madden2011; McGowan et al., Reference McGowan, White, Jordan and Paterson2014). Similar patterns have also been observed in Chinese (Liu et al., Reference Liu, Tong and Su2020; Wang et al., Reference Wang, Li, Li, Xie, Liversedge and Paterson2018b; Zhang et al., Reference Zhang, Warrington, Li, Pagán, Paterson, White and McGowan2022a; but see Wang et al., Reference Wang, Li, Li, Xie, Chang, Paterson, White and Mcgowan2018a; Zang et al., Reference Zang, Zhang, Bai, Yan, Paterson and Liversedge2016).
A growing number of studies have investigated word frequency effects in L2 reading using different tasks. Isolated word recognition studies using different tasks such as lexical decision have consistently shown larger word frequency effects in L2 than in L1 among bilinguals (Duyck et al., Reference Duyck, Vanderelst, Desmet and Hartsuiker2008; Mor & Prior, Reference Mor and Prior2020), as well as larger effects in bilinguals compared to monolinguals (Brysbaert et al., Reference Brysbaert, Lagrou and Stevens2017; Diependaele et al., Reference Diependaele, Lemhöfer and Brysbaert2013; Duyck et al., Reference Duyck, Vanderelst, Desmet and Hartsuiker2008; Gollan et al., Reference Gollan, Slattery, Goldenberg, Van Assche, Duyck and Rayner2011; Hamrick & Pandža, Reference Hamrick and Pandža2020; Lehtonen et al., Reference Lehtonen, Hultén, Rodríguez-Fornells, Cunillera, Tuomainen and Laine2012; Lemhöfer et al., Reference Lemhöfer, Dijkstra, Schriefers, Baayen, Grainger and Zwitserlood2008). Recent eye-tracking studies examining online lexical processing during natural reading have also reported converging evidence for larger word frequency effects in L2 than in L1. Studies of English–French, French–English and Dutch–English speakers consistently show amplified frequency effects in L2 across early and late eye movement measures in both adult and child bilinguals (Cop et al., Reference Cop, Keuleers, Drieghe and Duyck2015; Whitford & Joanisse, Reference Whitford and Joanisse2018; Whitford & Titone, Reference Whitford and Titone2012, Reference Whitford and Titone2017). Corpus-based analyses further demonstrate that L2 English learners from diverse L1 backgrounds exhibit larger frequency effects than native speakers during sentence reading (Berzak & Levy, Reference Berzak and Levy2023). Collectively, these findings indicate that word frequency effects are reliably larger in L2 than in L1 during isolated word recognition and natural reading.
However, most previous research has examined L1 and L2 within alphabetic language pairs (e.g., Cop et al., Reference Cop, Keuleers, Drieghe and Duyck2015; Whitford & Joanisse, Reference Whitford and Joanisse2018; Whitford & Titone, Reference Whitford and Titone2012), which may limit the ability to isolate the unique characteristics of L2 processing. In contrast, relatively few studies have explored word frequency effects in bilinguals with typologically distinct L1–L2 pairs, such as Hebrew–English (Mor & Prior, Reference Mor and Prior2020, Reference Mor and Prior2022), Chinese–English (Sui et al., Reference Sui, Woumans, Duyck and Dirix2025) and Chinese–Russian (Norkina et al., Reference Norkina, Chernova, Alexeeva and Harchevnik2025). Studies examining typologically distinct L1–L2 pairs have reported similar patterns. For example, Hebrew–English bilinguals exhibit larger word frequency effects in L2 than in L1 in both lexical decision (Mor & Prior, Reference Mor and Prior2020) and eye-tracking studies (Mor & Prior, Reference Mor and Prior2022). Corpus-based eye-tracking research further demonstrates amplified L2 frequency effects among Chinese–English bilinguals during paragraph reading (Sui et al., Reference Sui, Woumans, Duyck and Dirix2025), and in Chinese–Russian bilinguals compared to Russian monolinguals during sentence reading (Norkina et al., Reference Norkina, Chernova, Alexeeva and Harchevnik2025). Together, these findings suggest that larger L2 word frequency effects may generalize across script differences.
Prior studies have reported that word frequency effects in L2 are modulated by language proficiency. Bilinguals with lower L2 proficiency exhibit larger L2 word frequency effects (Berzak & Levy, Reference Berzak and Levy2023; Diependaele et al., Reference Diependaele, Lemhöfer and Brysbaert2013; Li et al., Reference Li, Li, Liu and Bai2017, Reference Li, Sui, Hyönä and Chen2026; Mor & Prior, Reference Mor and Prior2020, Reference Mor and Prior2022; Whitford & Titone, Reference Whitford and Titone2012; Zhao et al., Reference Zhao, Wei and Yao2024). When L2 proficiency is sufficiently high or is controlled for, the difference in word frequency effects between bilinguals and monolinguals disappears (Berzak & Levy, Reference Berzak and Levy2023; Brysbaert et al., Reference Brysbaert, Lagrou and Stevens2017; Diependaele et al., Reference Diependaele, Lemhöfer and Brysbaert2013). For example, Brysbaert et al. (Reference Brysbaert, Lagrou and Stevens2017) found larger word frequency effects in Dutch-English bilinguals compared to English monolinguals in a lexical decision task. Critically, when vocabulary size was considered, the differences between the word frequency effect in L1 and L2 became vanishingly small.
1.2. Theoretical accounts for larger L2 word frequency effect
Despite the robustness of larger L2 word frequency effects, their underlying cognitive mechanisms remain debated. Two prominent accounts have been advanced to explain this pattern (Diependaele et al., Reference Diependaele, Lemhöfer and Brysbaert2013): the lexical entrenchment hypothesis and the language competition hypothesis. The lexical entrenchment hypothesis attributes the word frequency effect to differences in the strength of lexical representations in memory. Words encountered more frequently develop more entrenched representations and are therefore processed more efficiently. Because bilinguals typically have lower cumulative exposure and proficiency in L2 than in L1, L2 lexical representations are assumed to be weaker, making lexical retrieval more effortful and leading to larger word frequency effects. Crucially, this account predicts that the magnitude of the L2 frequency effect should primarily reflect differences in representational strength and thus be relatively independent of the typological similarity between L1 and L2 (Diependaele et al., Reference Diependaele, Lemhöfer and Brysbaert2013).
In contrast, the language competition hypothesis emphasizes that bilingual lexical processing involves parallel activation of L1 and L2 representations and that larger L2 word frequency effects result from language competition. According to this view, L2 word recognition is influenced not only by the strength of L2 lexical representations but also by interference from co-activated L1 items. Bilinguals need to cope with greater competition among similar word form representations than monolinguals. Consequently, the magnitude of L2 word frequency effects may depend on the similarity between L2 and L1 lexical form representations, such that learners whose L1 is more similar to L2 may experience stronger interference and altered frequency effects. Importantly, unlike the lexical entrenchment hypothesis, this hypothesis predicts that the L2 word frequency effect increases as a function of L1–L2 similarity.
Although cross-language activation during word recognition is well established (e.g., cognate effects and cross-language neighborhood effects; de Groot et al., Reference de Groot, Borgwaldt, Bos and van den Eijnden2002; Gastmann et al., Reference Gastmann, Poarch and Schimke2025; Whitford & Joanisse, Reference Whitford and Joanisse2021; Whitford & Titone, Reference Whitford and Titone2019), and the effects of L1–L2 distance on eye movements in L2 reading have been documented (Kuperman, Reference Kuperman2025), few studies have directly tested whether L1–L2 similarity modulates L2 word frequency effects. Evidence from Chinese–English bilinguals (Sui et al., Reference Sui, Woumans, Duyck and Dirix2025) suggests that their observed patterns do not fully align with predictions from either the language competition or the lexical entrenchment hypotheses, particularly when L1 and L2 differ substantially in script. This indicates that the applicability of these accounts to cross-script bilinguals remains unclear.
In sum, the critical distinction between the lexical entrenchment and language competition hypotheses lies in whether L2 word frequency effects are driven by weaker lexical representations or by cross-language interference. Previous studies, however, have largely focused on alphabetic language pairs, where shared script may introduce close L1–L2 links through specific language exposure, lexical item overlap and transfer of underlying reading skills (Mor & Prior, Reference Mor and Prior2020, Reference Mor and Prior2022). These factors may mask processing differences that are purely attributable to L2 status. Although several studies have explored L2 word frequency effects in bilinguals with typologically distinct L1–L2 pairs (Mor & Prior, Reference Mor and Prior2020, Reference Mor and Prior2022; Norkina et al., Reference Norkina, Chernova, Alexeeva and Harchevnik2025; Sui et al., Reference Sui, Woumans, Duyck and Dirix2025), it remains unclear whether L1–L2 similarity modulates word frequency effects.
1.3. Word frequency effects in L2 Chinese reading
While existing theoretical accounts of L2 word frequency effects have largely been developed based on alphabetic languages, Chinese, as a logographic language, provides a unique opportunity to test the predictions of the lexical entrenchment and language competition hypotheses, particularly through comparisons between L2 learners whose L1 is typologically similar versus dissimilar to Chinese.
Chinese is a logographic language with fundamental differences from alphabetic languages. Chinese words are composed of characters and are written without spaces to mark word boundaries. These characteristics pose challenges for L2 learners who often struggle to segment continuous character strings into words during reading (Li et al., Reference Li, Ji, Li, Liu, Wang, Gunn and Paterson2025; Shen et al., Reference Shen, Liversedge, Tian, Zang, Cui, Bai, Yan and Rayner2012; Yao et al., Reference Yao, Jiang, Chen and Li2025a, Reference Yao, Li, Chen and Li2025b). Such segmentation difficulty may be especially pronounced for low-frequency words, whose boundaries are less predictable due to limited exposure, potentially resulting in larger word frequency effects in L2 learners compared to native readers.
To date, although many studies have examined word frequency effects in L2 Chinese (Jiang, Reference Jiang2005, Reference Jiang2006; Jiang et al., Reference Jiang, Zhao, Huang, Liu and Wang2006; Zhang, Reference Zhang2008), direct comparisons between L1 and L2 Chinese readers remain scarce. To our knowledge, only Hao et al. (Reference Hao, Wu and Duan2024) and Lai and Jiang (Reference Lai and Jiang2026) have included both groups. Hao et al. (Reference Hao, Wu and Duan2024) reported that Indonesian-speaking L2 learners exhibited word frequency effects in a lexical decision task, whereas native speakers did not. However, this lexical decision task may not reflect natural Chinese reading, where the lack of interword spaces demands simultaneous segmentation and identification (Li & Pollatsek, Reference Li and Pollatsek2020), a process that cannot be captured in isolated word recognition (Sui et al., Reference Sui, Woumans, Duyck and Dirix2025). Recently, Lai and Jiang (Reference Lai and Jiang2026) used eye-tracking and reported larger word frequency effects for alphabetic-L1 learners than for L1 readers during sentence reading. But with only one L2 group, the study leaves open the question of how L1–L2 similarity modulates the word frequency effect, specifically whether learners from typologically different L1 backgrounds show differential frequency effects. Answering this question is crucial because the lexical entrenchment hypothesis and the language-competition hypothesis make different predictions regarding the influence of L1–L2 similarity.
1.4. The present study
In the present study, we examined whether L2 word frequency effects in Chinese sentence reading are modulated by L1–L2 similarity. Specifically, we compared frequency effects in L2 Chinese learners from alphabetic-language backgrounds and Japanese L1 backgrounds using eye-tracking during sentence reading. Native Chinese readers were also included as a baseline group to establish the magnitude of word frequency effects in skilled native reading and to examine whether L2 readers exhibit larger word frequency effects than native readers. Our primary research question, however, concerned whether frequency effects differed between the two L2 groups as a function of their writing-system backgrounds. Whereas alphabetic languages share minimal orthographic and lexical overlap with Chinese, Japanese shares substantial orthographic and lexical similarity with Chinese, including the use of Chinese characters and a larger number of compound words in similar form and meaning. By contrasting these two learner groups, the present study provides a direct test of whether L2 word frequency effects are shaped by cross-linguistic similarity between L1 and L2, thereby adjudicating between accounts that attribute larger L2 word frequency effects to lexical entrenchment versus those emphasizing cross-language competition.
If L2 word frequency effects primarily reflect lexical entrenchment within the L2 lexicon, the magnitude of frequency effects should not differ between the two learner groups, assuming comparable L2 proficiency. In contrast, if L2 word frequency effects are driven by cross-language competition arising from shared orthographic and lexical representations, we would expect larger word frequency effects in learners with a Japanese L1 background, due to greater lexical representational overlap with Chinese.
2. Method
2.1. Participants
The sample analyzed in the present study comprised 150 participants: 50 native Chinese speakers (henceforth NS; 42 females; age range: 17–28 years; mean age: 21.1), 50 L2 learners of Chinese from alphabetic L1 backgrounds (henceforth L2-AL; 44 females; age range: 19–33 years; mean age: 24.0) and 50 L2 learners of Chinese from Japanese L1 backgrounds (henceforth L2-JL; 37 females; age range: 19–30 years; mean age: 22.3). The NS and L2-AL data were drawn from the dataset reported in Lai and Jiang (Reference Lai and Jiang2026), which included 56 NS and 57 L2-AL participants. For the present study, 50 L2-AL participants were selected to match the newly recruited group of 50 L2-JL participants in Chinese proficiency. To maintain equal group sizes, 50 NS participants were randomly selected from the original NS sample. The L2-JL data were collected specifically for the present study using the same materials, equipment and experimental procedure as those used for the NS and L2-AL groups in Lai and Jiang (Reference Lai and Jiang2026).
The recruitment target for the newly collected L2-JL group was set at 50 participants on the basis of feasibility, given the time-intensive nature of eye-tracking data collection and the limited availability of eligible participants. Because no previous eye-tracking study has directly compared word frequency effects in L2 Chinese reading among learners from different L1 backgrounds, no directly applicable estimate of the focal group-by-word-frequency interaction was available for a prospective power analysis. Following prior applications of simulation-based sensitivity analysis in bilingualism research (Washington & Wiley, Reference Washington and Wiley2023; Yow & Li, Reference Yow and Li2026), we conducted a sensitivity analysis on the fitted mixed-effects model using the simr package (Green & MacLeod, Reference Green and MacLeod2016). A simulation-based sensitivity analysis estimates the smallest effect that a given design can detect with a prespecified level of power. For each candidate effect size, data are repeatedly generated from the fitted model with the target effect set to that candidate value. The model is then refitted to each simulated data set, and power is estimated as the proportion of simulations in which the target effect is statistically significant. The analysis targeted the critical interaction between word frequency and L2 group (L2-AL versus L2-JL), which directly operationalizes our main research question. Total reading time on the target words was used as the dependent variable because this cumulative measure encompasses both initial processing and subsequent rereading and has been shown to be sensitive to word frequency effects in L2 reading (Lai & Jiang, Reference Lai and Jiang2026; Mor & Prior, Reference Mor and Prior2022; Whitford & Titone, Reference Whitford and Titone2012, Reference Whitford and Titone2017). We first conducted a coarse scan across standardized interaction effects of d = 0.05, 0.10, 0.15 and 0.20 (step = 0.05), using 200 simulations per value. Given the fitted model’s residual standard deviation of 0.510, these corresponded to interaction coefficients of b = 0.026, 0.051, 0.077 and 0.102 on the log scale. We then refined the search at d = 0.15, 0.16 and 0.17 (step = 0.01; b = 0.077, 0.082 and 0.087), using 1000 simulations per value (α = .05). The analysis indicated that the current design had approximately 80% power to detect an interaction of b = 0.077 or larger on the log scale, corresponding to the L2-AL group’s ratio of low- to high-frequency reading times being approximately 8% larger than that of the L2-JL group.
The L2-AL group included learners from a range of alphabetic-language backgrounds (e.g., Vietnamese, Thai, Indonesian, Arabic, Russian, Korean, Spanish, French, Cape Verdean Creole, Mongolian and Burmese). Although some of these languages, particularly Vietnamese and Korean, contain Chinese loanwords or have historical contact with Chinese, their contemporary writing systems do not employ Chinese characters (logographs) as a primary orthographic component. In contrast, Japanese uses kanji as an integral component of its contemporary writing system and contains a substantial body of Sino-Japanese vocabulary. Thus, relative to the L2-JL group, the L2-AL group was operationally characterized as having lower average orthographic and lexical similarity to Chinese.
All L2 participants had passed HSK 5 or HSK 6 (Hanyu Shuiping Kaoshi, the standard Chinese language proficiency test for non-native speakers administered by the Ministry of Education of the People’s Republic of China), which corresponds approximately to the B2 level of the Common European Framework of Reference for Languages (CEFR; Luo, Reference Luo2017), indicating upper-intermediate proficiency in Chinese. All L2 participants completed a character recognition test (Hou & Jiang, Reference Hou and Jiang2022; Jiang & Liu, Reference Jiang and Liu2004) and a ClozCHI test (Cai et al., Reference Cai, Han and Jiang2025). ClozCHI is a cloze test comprising three passages with varying levels of difficulty, in which participants fill each blank with one appropriate character or its Pinyin based on context (Cai et al., Reference Cai, Han and Jiang2025). Compared with HSK level, which was obtained at varying time points prior to the experiment and may not accurately reflect participants’ current proficiency, these two tests provide more fine-grained and up-to-date measures of L2 Chinese proficiency. The character recognition test scores were matched between L2-AL (M = 0.77, SD = 0.15) and L2-JL (M = 0.77, SD = 0.14), t(98) = −0.29, p = .77. The ClozCHI scores were also matched between L2-AL (M = 0.74, SD = 0.08) and L2-JL (M = 0.72, SD = 0.10), t(98) = 1.06, p = .29.
All participants had normal or corrected-to-normal vision. Informed consent was obtained from all participants, and monetary compensation was provided for their participation.
2.2. Materials
A total of 640 two-character words were selected as target words, comprising 320 high-frequency and 320 low-frequency words (see Table S1 in supplementary material). Word frequency data were obtained from the Chinese Lexical Database (CLD; Sun et al., Reference Sun, Hendrix, Ma and Baayen2018). Paired-samples t-tests indicated that high-frequency words (M = 92.34 occurrences per million, SD = 95.64) and low-frequency words (M = 1.44 occurrences per million, SD = 1.02) differed significantly in frequency, t(319) = 16.98, p < .001, and were matched on the frequency and number of strokes of both the first and second characters, as well as the total number of strokes of the two characters (ps > .10). Familiarity ratings, collected after the eye-tracking experiment on a 7-point scale (1 = very unfamiliar, 7 = very familiar), showed that the mean rating across all target words was 6.05 (SD = 0.92), with only three target words receiving a mean rating below 3. These results suggest that the target words were generally familiar to our L2 participants. The characteristics of target words are summarized in Table 1.
Characteristics of the target words
Table 1. Long description
The table consists of three columns: Characteristic, High-frequency, and Low-frequency. Standard deviations are provided in parentheses following each mean value.
* Frequency (number of occurrences per million): High-frequency is 92.34 (95.64); Low-frequency is 1.44 (1.02).
* Frequency of first character: High-frequency is 833.61 (1081.37); Low-frequency is 758.04 (1049.20).
* Frequency of second character: High-frequency is 768.79 (986.61); Low-frequency is 807.71 (1239.90).
* Number of strokes: High-frequency is 16.01 (4.54); Low-frequency is 16.08 (4.46).
* Number of strokes of first character: High-frequency is 7.87 (3.04); Low-frequency is 8.01 (3.09).
* Number of strokes of second character: High-frequency is 8.14 (3.11); Low-frequency is 8.08 (3.19).
Note: Standard deviations are given in parentheses.
Target words were embedded in 160 sentence frames, each containing two high-frequency and two low-frequency words. Using two pairs of high- and low-frequency words within the same sentence frame can increase the number of target words and enhance the robustness and generalizability of the results by reducing item-specific bias. Each participant saw only one target word per frame, and each frame was presented only once per participant (i.e., no participant was exposed to repeated frames across trials). Target words never appeared within the first or last four characters of a sentence. Sentences ranged from 18 to 31 characters and were displayed on a single line. Sentence plausibility was rated by 40 NS who did not take part in the main eye-tracking experiment, using a 7-point scale ranging from 1 (very implausible) to 7 (very plausible). Paired-samples t-tests indicated that plausibility ratings were matched between sentences containing high-frequency (M = 6.10, SD = 0.35) and low-frequency words (M = 6.06, SD = 0.35), t(319) = 1.41, p = .16. The predictability of each target word was assessed using a cloze task completed by another 10 NS, who read the sentence up to the target word and provided the next acceptable word. Paired-samples t-tests indicated that predictability was matched between high-frequency (M = 0.0009, SD = 0.01) and low-frequency words (M = 0.0003, SD = 0.01), t(319) = 1.00, p = .32. Examples of sentences and target words are presented in Table 2.
Examples of sentences and target words
Table 2. Long description
The table is organized into two columns: Target word and Example.
* The first category is High-frequency target words.
* The first example for High-frequency is a Chinese sentence with the target word ‘gift’ in bold. The English translation is: Xiao Chen was very happy that she had received a gift.
* The second example for High-frequency is a Chinese sentence with the target word ‘medal’ in bold. The English translation is: Xiao Chen was very happy that she had received a medal.
* The second category is Low-frequency target words.
* The first example for Low-frequency is a Chinese sentence with the target word ‘certificate’ in bold. The English translation is: Xiao Chen was very happy that she had received a certificate.
* The second example for Low-frequency is a Chinese sentence with the target word ‘high salary’ in bold. The English translation is: Xiao Chen was very happy that she had received a high salary.
Note: Target words are indicated in bold.
The experimental sentences were divided into eight lists, each list containing 80 experimental sentences (including 40 high-frequency words and 40 low-frequency words) as well as 40 filler sentences. Filler sentences were followed by yes/no comprehension questions. The mean frequency of high-frequency words did not differ significantly across the eight lists, F(7, 312) = 0.64, p = .73. Similarly, the mean frequency of low-frequency words was matched across lists, F(7, 312) = 1.12, p = .35. Sentences were counterbalanced across lists using a Latin square design and presented in a randomized order.
2.3. Apparatus
Participants’ eye movements were recorded using an SR Research Eyelink 1000 plus eye tracker with a sampling rate of 1000 Hz. Stimuli were displayed on a 20-inch CRT monitor with a resolution of 1024 × 768 pixels and a refresh rate of 150 Hz. Sentences were presented in Song font in black on a white background, with each character subtending approximately 0.85° of visual angle. The first character of the sentence and the fixation point were positioned at 100 pixels from the left edge and 384 pixels from the top of the display. Participants were seated approximately 72 cm from the screen. Following a 3-point horizontal calibration and validation, the maximum gaze-position error was less than 0.5°. Viewing was binocular, but only the right eye was recorded.
2.4. Procedure
Participants were tested individually in the laboratory. The experiment started with a brief instruction and a standard horizontal 3-point grid calibration and validation. Each trial began with a fixation box presented at the left side of the screen, followed by the presentation of the target sentence. Participants were instructed to read sentences silently for comprehension and to answer occasional comprehension questions by pressing yes/no buttons. After reading, participants pressed a button to proceed. Prior to the formal experimental trials, participants completed four practice trials to familiarize themselves with the task and the eye-tracking equipment. The experiment lasted approximately 20 minutes for NS and 40 minutes for L2 Chinese learners.
Before the eye-tracking session, L2 Chinese learners completed the character recognition test and the ClozCHI test, which were not administered to NS.
2.5. Data analysis
Both global and local eye movement measures were analyzed. Specifically, sentence reading rate (SRR; characters per minute) was analyzed to reflect general reading efficiency (Rayner, Reference Rayner1998). Five eye movement measures on the target words were analyzed, including skipping rate (SR, the probability of a region being skipped during first-pass reading), first fixation duration (FFD, the duration of the first fixation in a region during first-pass reading, following a progressive saccade), gaze duration (GD, the sum of all first-pass fixation durations in a region before leaving it, following a progressive saccade), total reading time (TT, the sum of the durations of all fixations in a region), and regression path duration (RPD, the sum of the durations of all fixations in a region and any rereading to the left of it until it was passed to the right). While SR, FFD and GD reflect the early stages of processing, TT and RPD reflect the late stages of processing.
Linear mixed-effects models (LMMs) were employed to analyze log-transformed continuous variables (SRR, FFD, GD, TT, RPD), and generalized linear mixed-effects models (GLMMs) were used to analyze the binary variable (SR). Word frequency, group and their interaction were treated as fixed effects, and participants and items were treated as crossed random effects, including intercepts and slopes (Baayen et al., Reference Baayen, Davidson and Bates2008). Word frequency and group were treated as categorical variables. For the group condition, three successive contrasts were set up to analyze differences between L2-AL and NS (Group 1), differences between L2-JL and NS (Group 2) and differences between L2-AL and L2-JL (Group 3). Specifically, contrast coding was employed to examine these contrasts. Two sets of models were fitted with different reference levels: the first model used NS as the reference level, yielding contrasts for L2-AL versus NS and L2-JL versus NS; the second model used L2-JL as the reference level, yielding the contrast for L2-AL versus L2-JL. Following Bates et al. (Reference Bates, Kliegl, Vasishth and Baayen2015a), the random effects structure of the models was trimmed down from the maximal structure (i.e., random intercepts and slopes for all fixed effects, for both subjects and items) when the models failed to converge. Model comparisons were conducted using likelihood ratio tests and the Bayesian Information Criterion (BIC), and the more parsimonious model was selected when it did not significantly differ from the more complex model (p > .05). Data were analyzed in R (version 4.4.2; R Core Team, 2024) using the lme4 package (version 1.1.37; Bates et al., Reference Bates, Mächler, Bolker and Walker2015b) and lmerTest package (version 3.1.3; Kuznetsova et al., Reference Kuznetsova, Brockhoff and Christensen2017). Simple effects were tested using emmeans package (Lenth, Reference Lenth2018).
3. Results
Native speakers achieved a mean accuracy of 94% (range: 83%–100%). The L2-AL group achieved a mean accuracy of 85% (range: 70%–100%), and the L2-JL group achieved a mean accuracy of 86% (range: 70%–100%). The high comprehension accuracy rates indicate that participants attended to and successfully comprehended the sentences.
Fixations shorter than 80 ms or longer than 1000 ms (approximately 8.7% of all fixations) were excluded from the analysis. Trials were removed if (a) participants made more than three blinks or no fixations in the whole sentence (0.2% of trials); (b) participants blinked on the target words, pre-target words or post-target words (4.9% of trials) or (c) eye movement measures were above or below three standard deviations from the mean (1.4% of trials). This resulted in a loss of 6.5% of the trials. Excluding trials with blinks (a and b) is common practice in eye-tracking reading research (e.g., Inhoff & Rayner, Reference Inhoff and Rayner1986; Liu et al., Reference Liu, Li, Han and Li2014; Rayner et al., Reference Rayner, Yang, Schuett and Slattery2013; Yao et al., Reference Yao, Li, Chen and Li2025b), since valid fixation data cannot be obtained during blinks. The exclusion rate due to blinks was relatively low and blinks occurred randomly, making them unlikely to systematically affect the validity of our results.
Table 3 presents the means and standard errors for each eye movement measure. Its note summarizes the number of observations retained per condition across the outcome measures, while the exact counts for every measure and Group × Frequency condition are reported in Table S2 of the supplementary material. Table 4 shows the results of the linear mixed-effects models/generalized linear mixed-effects models. The residuals of the linear mixed-effects models were approximately normally distributed.
Means and standard errors for eye movement measures
Table 3. Long description
The table contains 10 columns and 7 rows. The header row identifies three main groups: N S (Native Chinese speakers), L 2 dash A L (L 2 Chinese learners from alphabetic L 1 backgrounds), and L 2 dash J L (L 2 learners from Japanese L 1 backgrounds). Each group is subdivided into H F (High-frequency), L F (Low-frequency), and F E (Frequency effect).
Row 1: S R R (Sentence reading rate). N S: H F 639.69, L F 634.45, F E minus 5.24. L 2 dash A L: H F 213.64, L F 205.81, F E minus 7.83. L 2 dash J L: H F 287.85, L F 280.65, F E minus 7.20.
Row 2: S R (Skipping rate). N S: H F 0.32, L F 0.29, F E minus 0.03. L 2 dash A L: H F 0.03, L F 0.03, F E 0. L 2 dash J L: H F 0.10, L F 0.08, F E minus 0.02.
Row 3: F F D (First fixation duration). N S: H F 236.87, L F 243.75, F E 6.88. L 2 dash A L: H F 293.88, L F 305.77, F E 11.89. L 2 dash J L: H F 288.01, L F 299.68, F E 11.67.
Row 4: G D (Gaze duration). N S: H F 256.06, L F 266.80, F E 10.74. L 2 dash A L: H F 491.83, L F 564.34, F E 72.51. L 2 dash J L: H F 400.19, L F 451.26, F E 51.07.
Row 5: T T (Total reading time). N S: H F 326.79, L F 346.48, F E 19.69. L 2 dash A L: H F 705.66, L F 849.48, F E 143.82. L 2 dash J L: H F 582.00, L F 691.27, F E 109.27.
Row 6: R P D (Regression path duration). N S: H F 341.09, L F 354.78, F E 13.69. L 2 dash A L: H F 614.29, L F 718.06, F E 103.77. L 2 dash J L: H F 499.88, L F 565.61, F E 65.73.
Notes: Standard errors are given in parentheses. FE = Frequency effect; FFD = First fixation duration; GD = Gaze duration; HF = High-frequency; L2-AL = L2 Chinese learners from alphabetic L1 backgrounds; L2-JL = L2 learners from Japanese L1 backgrounds; LF = Low-frequency; NS = Native Chinese speakers; RPD = Regression path duration; SR = Skipping rate; SRR = Sentence reading rate; TT = Total reading time. The number of observations per condition across all measures ranged from 1314 to 1938 (M = 1715.75, SD = 188.61).
Results of the LMMs/GLMMs
Table 4. Long description
The table contains six sections based on the measure being analyzed. Each section includes columns for Variable, b (coefficient), S E (standard error), t/z value, and p-value.
* S R R (Sentence Reading Rate): Intercept (5.73), Frequency (-0.03), Group 1 (-1.14), Group 2 (-0.85), and Group 3 (-0.29) all show p < .001. Interactions between Frequency and Groups 1, 2, and 3 are not significant (p > .05).
* S R (Skipping Rate): Intercept (-2.70, p < .001), Frequency (-0.22, p = .009), Group 1 (-3.29, p < .001), Group 2 (-1.97, p < .001), and Group 3 (-1.32, p < .001). Interactions are not significant.
* F F D (First Fixation Duration): Intercept (5.57, p < .001), Frequency (0.03, p < .001), Group 1 (0.22, p < .001), and Group 2 (0.20, p < .001). Group 3 and all interactions are not significant.
* G D (Gaze Duration): Intercept (5.84), Frequency (0.09), Group 1 (0.65), Group 2 (0.42), and Group 3 (0.23) all show p < .001. Interactions for Frequency times Group 1 (0.10) and Group 2 (0.08) are significant at p < .001.
* T T (Total Reading Time): Intercept (6.14), Frequency (0.15), Group 1 (0.83), Group 2 (0.58), and Group 3 (0.24) all show p < .001. Interactions for Frequency times Group 1 (0.15) and Group 2 (0.13) are significant at p < .001.
* R P D (Regression Path Duration): Intercept (6.04), Frequency (0.11), Group 1 (0.65), Group 2 (0.41), and Group 3 (0.25) all show p < .001. Interactions for Frequency times Group 1 (0.13) and Group 2 (0.09) are significant at p < .001.
Notes specify Group 1 is L 2-A L versus N S, Group 2 is L 2-J L versus N S, and Group 3 is L 2-A L versus L 2-J L.
Notes: FFD = First fixation duration; GD = Gaze duration; Group 1 = L2-AL versus NS; Group 2 = L2-JL versus NS; Group 3 = L2-AL versus L2-JL; L2-AL = L2 Chinese learners from alphabetic L1 backgrounds; L2-JL = L2 learners from Japanese L1 backgrounds; NS = Native Chinese speakers; RPD = Regression path duration; SR = Skipping rate; SRR = Sentence reading rate; TT = Total reading time; **p < .01; ***p < .001; Results for Group 3 and Frequency × Group 3 were obtained from a separate model with L2-JL as the reference level.
3.1. Global measure
For SRR, there was a main effect of word frequency, indicating that SRR was slower for low-frequency words than high-frequency words. There was also a main effect of group, with both L2-AL and L2-JL exhibiting slower SRR than NS, and L2-AL exhibiting slower SRR than L2-JL. The interaction between word frequency and group was marginally significant for Group 1 (L2-AL versus NS) and Group 2 (L2-JL versus NS), but not for Group 3 (L2-AL versus L2-JL). Further simple effects analysis revealed that the word frequency effect was not significant for NS (t = 0.83, p = .41), but significant for both L2-AL (t = 3.50, p < .001) and L2-JL (t = 3.35, p < .001). These results indicate that the word frequency effect was larger for both L2 groups compared to NS, with no significant difference between L2-AL and L2-JL.
3.2. Early local measures
For SR, there was a main effect of word frequency, indicating that SR was lower for low-frequency words than high-frequency words. There was also a main effect of group, with both L2-AL and L2-JL exhibiting lower SR than NS, and L2-AL exhibiting lower SR than L2-JL. None of the other effects were significant.
For FFD, there was a main effect of word frequency, indicating that FFD was longer for low-frequency words than high-frequency words. There was also a main effect of group, with both L2-AL and L2-JL exhibiting longer FFD than NS. None of the other effects were significant.
For GD, there was a main effect of word frequency, indicating that GD was longer for low-frequency words than high-frequency words. There was also a main effect of group, with both L2-AL and L2-JL exhibiting longer GD than NS, and L2-AL exhibiting longer GD than L2-JL. The interaction between word frequency and group was significant for Group 1 (L2-AL versus NS) and Group 2 (L2-JL versus NS), but not for Group 3 (L2-AL versus L2-JL). Further simple effects analysis revealed that the word frequency effect was marginally significant for NS (t = −1.81, p = .07), but significant for both L2-AL (t = −7.99, p < .001) and L2-JL (t = −6.98, p < .001). These results indicate that the word frequency effect was larger for both L2 groups compared to NS, with no significant difference between L2-AL and L2-JL.
3.3. Late local measures
For TT, there was a main effect of word frequency, indicating that TT was longer for low-frequency words than high-frequency words. There was also a main effect of group, with both L2-AL and L2-JL exhibiting longer TT than NS, and L2-AL exhibiting longer TT than L2-JL. The interaction between word frequency and group was significant for Group 1 (L2-AL versus NS) and Group 2 (L2-JL versus NS), but not for Group 3 (L2-AL versus L2-JL). Further simple effects analysis revealed that the word frequency effect was significant for all groups (NS: t = −2.42, p = .02; L2-AL: t = −9.87, p < .001; L2-JL: t = −8.28, p < .001). These results indicate that the word frequency effect was larger for both L2 groups compared to NS, with no significant difference between L2-AL and L2-JL.
For RPD, there was a main effect of word frequency, indicating that RPD was longer for low-frequency words than high-frequency words. There was also a main effect of group, with both L2-AL and L2-JL exhibiting longer RPD than NS, and L2-AL exhibiting longer RPD than L2-JL. The interaction between word frequency and group was significant for Group 1 (L2-AL versus NS) and Group 2 (L2-JL versus NS), but not for Group 3 (L2-AL versus L2-JL). Further simple effects analysis revealed that the word frequency effect was marginally significant for NS (t = −1.87, p = .06), but significant for both L2-AL (t = −8.91, p < .001) and L2-JL (t = −6.69, p < .001). These results indicate that the word frequency effect was larger for both L2 groups compared to NS, with no significant difference between L2-AL and L2-JL.
It is worth noting that the interaction between word frequency and group was not significant for Group 3 (L2-AL versus L2-JL) across eye movement measures. Since Bayes factors can be used to determine whether non-significant results support a null hypothesis (Dienes, Reference Dienes2014), Bayes factor analyses were further conducted to assess the strength of evidence for this null effect, using the BayesFactor package in R (Morey et al., Reference Morey, Rouder and Jamil2015). BF10 values ranging from 0.01 to 0.12 across eye movement measures provided moderate to very strong evidence in favor of the null hypothesis, indicating that L2 word frequency effects did not differ between L2-AL and L2-JL across eye movement measures.
Additionally, to examine whether sensitivity to word frequency varied as a function of L2 proficiency, correlations were conducted between L2 proficiency (character recognition and cloze test scores) and word frequency effects within each L2 group. The correlations were generally weak and non-significant across eye movement measures (|r| = .01–.27, p = .06–.94).
To explore word frequency effects more broadly, correlation analyses were conducted between log-transformed word frequency and eye movement measures for all words in the sentences (excluding the first and last words of each sentence; Meng et al., Reference Meng, Lan, Yan, Marsh and Liversedge2020). Word frequency was significantly negatively correlated with reading times (FFD, GD, TT and RPD) and positively with SR (|r| = .10–.35, ps < .001). Fisher’s z-tests revealed that word frequency showed significantly stronger correlations with GD, TT and RPD in L2-AL (|r| = .15–.35) than L2-JL (|r| = .10–.28; ps < .001), and a significantly stronger correlation with SR (r = .28) in L2-JL than L2-AL (r = .25; p < .001).
To summarize the main findings, word frequency effects were larger in both L2-AL and L2-JL than in NS in SRR, GD, TT and RPD. However, there was no significant difference in L2 word frequency effects between L2-AL and L2-JL across eye movement measures examined (SRR, SR, FFD, GD, TT and RPD).
4. Discussion
The present study examined word frequency effects in Chinese sentence reading among native Chinese readers and L2 Chinese learners from different writing-system backgrounds. More importantly, it investigated whether the magnitude of these effects varies as a function of L1–L2 writing-system similarity. The results showed that both L2 groups exhibited larger frequency effects than native readers, whereas the two L2 groups showed comparable frequency effects despite differences in writing-system background. These findings are more readily explained by differences in lexical entrenchment than by explanations based on cross-language competition.
The results revealed larger word frequency effects for L2 readers than for native Chinese readers in both global and local eye movement measures. Specifically, this pattern was observed in SRR, as well as in GD, TT and RPD, suggesting that lexical frequency exerted a stronger influence on both early and later stages of word processing in L2 than in L1 Chinese reading. These findings replicate a robust pattern in previous studies comparing L1 and L2 within alphabetic language pairs (Brysbaert et al., Reference Brysbaert, Lagrou and Stevens2017; Cop et al., Reference Cop, Keuleers, Drieghe and Duyck2015; Diependaele et al., Reference Diependaele, Lemhöfer and Brysbaert2013; Duyck et al., Reference Duyck, Vanderelst, Desmet and Hartsuiker2008; Gollan et al., Reference Gollan, Slattery, Goldenberg, Van Assche, Duyck and Rayner2011; Lehtonen et al., Reference Lehtonen, Hultén, Rodríguez-Fornells, Cunillera, Tuomainen and Laine2012; Lemhöfer et al., Reference Lemhöfer, Dijkstra, Schriefers, Baayen, Grainger and Zwitserlood2008; Whitford & Joanisse, Reference Whitford and Joanisse2018; Whitford & Titone, Reference Whitford and Titone2012, Reference Whitford and Titone2017), and typologically distinct L1–L2 pairs (Mor & Prior, Reference Mor and Prior2020, Reference Mor and Prior2022; Norkina et al., Reference Norkina, Chernova, Alexeeva and Harchevnik2025; Sui et al., Reference Sui, Woumans, Duyck and Dirix2025). Furthermore, the present study extended Lai and Jiang’s (Reference Lai and Jiang2026) eye-tracking findings of larger L2 word frequency effects in Chinese sentence reading, also observed in lexical decision tasks (Hao et al., Reference Hao, Wu and Duan2024), to learners from typologically different L1 backgrounds. Thus, our results further demonstrated the robustness of larger L2 word frequency effects and their generalizability across L1 backgrounds.
The findings of larger word frequency effects in L2 than in L1 can be readily explained with the lexical entrenchment hypothesis (Diependaele et al., Reference Diependaele, Lemhöfer and Brysbaert2013). Under this hypothesis, frequency effects index the strength and quality of lexical representations. L2 lexical memory representations are weaker than those in L1 due to reduced exposure, thereby amplifying sensitivity to frequency differences. This may be particularly pronounced in Chinese, where words are written without spaces and readers must segment continuous character strings into words (Li et al., Reference Li, Ji, Li, Liu, Wang, Gunn and Paterson2025; Shen et al., Reference Shen, Liversedge, Tian, Zang, Cui, Bai, Yan and Rayner2012; Yao et al., Reference Yao, Jiang, Chen and Li2025a, Reference Yao, Li, Chen and Li2025b). Weaker lexical representations make word boundaries less predictable and word identification more difficult, especially for low-frequency words, further amplifying word frequency effects in L2 readers. Conversely, for highly skilled native readers, extensive exposure to text leads to more robust lexical representations even for low-frequency words, thereby reducing processing differences between high- and low-frequency words.
However, stronger word frequency effects in L2 than in L1, by themselves, are insufficient to distinguish between the lexical entrenchment hypothesis and the language competition hypothesis (Diependaele et al., Reference Diependaele, Lemhöfer and Brysbaert2013). Importantly, this study showed that L2 word frequency effects did not differ between L2 learners from alphabetic-language and Japanese L1 backgrounds across eye movement measures. This absence of a reliable interaction provides a critical test of the competing accounts. Under the lexical entrenchment hypothesis, frequency effects primarily reflect the strength and automatization of lexical representations, which are determined by language experience and proficiency rather than by L1–L2 similarity (Diependaele et al., Reference Diependaele, Lemhöfer and Brysbaert2013). Because the two L2 groups were matched on Chinese proficiency, comparable frequency effects are what this account predicts. In contrast, the language competition hypothesis predicts that L2 learners from Japanese L1 backgrounds should exhibit larger L2 word frequency effects, given the substantial orthographic and lexical overlap between Japanese and Chinese (Diependaele et al., Reference Diependaele, Lemhöfer and Brysbaert2013). The absence of such modulation, supported both by non-significant interactions and by Bayesian evidence favoring the null model, suggests that competition-based accounts may not adequately explain L2 word frequency effects. In contrast, these findings are more readily accommodated by the lexical entrenchment hypothesis, which emphasizes the role of accumulated reading experience in shaping the strength of lexical representations.
Specifically, rather than reflecting dynamic cross-language interference, the larger L2 frequency effects observed here appear to index differences in lexical representational strength. This interpretation is consistent with converging theoretical accounts (e.g., Dijkstra & Van Heuven, Reference Dijkstra and Van Heuven2002; Gollan et al., Reference Gollan, Montoya, Cera and Sandoval2008, Reference Gollan, Slattery, Goldenberg, Van Assche, Duyck and Rayner2011; Perfetti, Reference Perfetti2007). The weaker links hypothesis (Gollan et al., Reference Gollan, Montoya, Cera and Sandoval2008, Reference Gollan, Slattery, Goldenberg, Van Assche, Duyck and Rayner2011) attributes larger L2 word frequency effects to reduced language exposure and divided language use, which weaken form–meaning connections. Similarly, the lexical quality hypothesis (Perfetti, Reference Perfetti2007) proposes that lexical representations vary in the precision and integration of orthographic, phonological and semantic information. Lower proficiency entails lower lexical quality, leading to slower and more frequency-sensitive word recognition. Moreover, the bilingual interactive activation Plus model (BIA+; Dijkstra & Van Heuven, Reference Dijkstra and Van Heuven2002) posits that L2 words generally have lower resting-level activations than do L1 words of the same corpus frequency, thus assuming larger word frequency effects in L2 than in L1. Relatedly, this interpretation is supported by converging evidence demonstrating the dominant role of language proficiency in word frequency effects, both in L1 (Ashby et al., Reference Ashby, Rayner and Clifton2005; Barnes et al., Reference Barnes, Kim, Tighe and Vorstius2017; Gordon et al., Reference Gordon, Moore, Choi, Hoedemaker and Lowder2020; Kuperman & Van Dyke, Reference Kuperman and Van Dyke2011, Reference Kuperman and Van Dyke2013; Lee et al., Reference Lee, Godwin, Blythe and Drieghe2025; Zhang et al., Reference Zhang, Hu, Bian, Li, Zhang and Zang2022b) and in L2 (Berzak & Levy, Reference Berzak and Levy2023; Li et al., Reference Li, Li, Liu and Bai2017, Reference Li, Sui, Hyönä and Chen2026; Mor & Prior, Reference Mor and Prior2020, Reference Mor and Prior2022; Whitford & Titone, Reference Whitford and Titone2012; Zhao et al., Reference Zhao, Wei and Yao2024). Furthermore, larger word frequency effects in L2 relative to L1 disappear when proficiency is controlled for (Brysbaert et al., Reference Brysbaert, Lagrou and Stevens2017) or when it reaches a sufficiently high level (Berzak & Levy, Reference Berzak and Levy2023), even for bilinguals from different L1 backgrounds (Diependaele et al., Reference Diependaele, Lemhöfer and Brysbaert2013). Collectively, these theoretical accounts and empirical findings are consistent with the lexical entrenchment hypothesis (Diependaele et al., Reference Diependaele, Lemhöfer and Brysbaert2013).
According to the lexical entrenchment hypothesis (Diependaele et al., Reference Diependaele, Lemhöfer and Brysbaert2013), one may also expect a negative correlation between language proficiency and frequency effects in each L2 group. To evaluate this prediction, we conducted correlation analyses between L2 proficiency and word frequency effects within each L2 group across the eye movement measures. The analyses revealed weak and non-significant correlations (|r| = .01–.27, p = .06–.94). One possible explanation is that the relatively restricted L2 proficiency range of our participants limited variability within the L2 groups, which may have reduced the likelihood of detecting correlations between L2 proficiency and frequency effects. In addition, by comparing words only at the extremes of the frequency continuum, our design may have overlooked the intermediate range, where group differences are likely to be most pronounced (Brysbaert et al., Reference Brysbaert, Lagrou and Stevens2017, Reference Brysbaert, Mandera and Keuleers2018). This interpretation is supported by Brysbaert et al. (Reference Brysbaert, Lagrou and Stevens2017), who reported a strong interaction between vocabulary size and word frequency in the middle frequency range, but not at the lower or higher ends of the frequency spectrum.
One potential concern is that the observed findings might be influenced by cognates, given the substantial lexical overlap between Japanese and Chinese. To address this issue, we conducted a supplementary analysis excluding cognates from the target word set. A total of 298 non-cognate words were included, comprising 123 high-frequency and 175 low-frequency items. The results remained largely consistent with the original analysis (see results on OSF for details), suggesting that the observed effects are unlikely to be explained by cognate status.
More broadly, by moving beyond comparisons within alphabetic writing systems to a cross-script contrast involving Chinese as a morphosyllabic writing system in L2 reading, the present study provides a more stringent test of whether L1–L2 similarity modulates L2 word frequency effects. However, while larger L2 word frequency effects in our study are consistent with prior studies comparing typologically distinct L1–L2 pairs (Mor & Prior, Reference Mor and Prior2020, Reference Mor and Prior2022; Norkina et al., Reference Norkina, Chernova, Alexeeva and Harchevnik2025; Sui et al., Reference Sui, Woumans, Duyck and Dirix2025), the finding of comparable L2 word frequency effects across L2 learners from alphabetic and Japanese L1 backgrounds differs from Sui et al. (Reference Sui, Woumans, Duyck and Dirix2025), who reported larger L2 word frequency effects in Chinese–English bilinguals with different L1–L2 scripts than Dutch–English bilinguals during paragraph reading. One possible explanation for this discrepancy lies in differences in the language pairs and writing systems involved. In Sui et al. (Reference Sui, Woumans, Duyck and Dirix2025), the comparison involved English as the target language, where lexical processing is strongly dependent on alphabetic orthographic decoding. In contrast, the present study focused on Chinese reading, where lexical processing relies more heavily on character-level visual-orthographic and morphosyllabic information, which may reduce the impact of L1 script distance on word frequency effects.
This discrepancy may also be attributable to methodological differences. The present study employed an experimental design with controlled manipulation of frequency of target words and matching of key lexical variables (e.g., stroke number and word predictability), whereas Sui et al. (Reference Sui, Woumans, Duyck and Dirix2025) adopted a corpus-based approach in which all words encountered during paragraph reading were analyzed, allowing for a broader range of lexical variability. We conducted a supplementary analysis including all words in the sentences, and the results provided support for this interpretation. Specifically, the result showed that word frequency showed stronger associations with reading times in L2-AL group than in L2-JL group, whereas the opposite pattern was observed for skipping rate (SR).
This dissociation across measures may reflect differences in online processing efficiency or reading strategies between the two L2 groups. Although the groups were matched on L2 Chinese proficiency based on character recognition and cloze test scores, these measures primarily assessed general L2 proficiency and may not have captured finer-grained differences in online processing efficiency during reading. Since Japanese shares greater orthographic and lexical similarity with Chinese than alphabetic languages, L2-JL readers may benefit from greater experience with logographic orthography, which could enhance online reading efficiency in reading Chinese, even when offline proficiency measures are comparable.
This interpretation is further supported by the group differences in our data: L2-AL readers showed longer reading times (SRR, GD, TT and RPD) and lower SR than L2-JL, a pattern consistent with Sui et al. (Reference Sui, Woumans, Duyck and Dirix2025), who likewise found slower reading speeds in Chinese–English bilinguals than in Dutch–English bilinguals. In the analysis of all words (i.e., without controlling for other variables), the correlations between word frequency and reading times were stronger in the L2-AL group than in the L2-JL group. In contrast, the correlations between word frequency and skipping rate were stronger in the L2-JL group than in the L2-AL group. Because these correlations were derived from all words rather than experimentally controlled target items, they were likely modulated by factors such as character frequency, word length, contextual predictability and individual reading strategies. From this perspective, the stronger frequency–reading time correlations observed in the L2-AL group may reflect a more direct reliance on lexical decoding, whereas L2-JL readers may have relied more on contextual prediction and more efficient oculomotor strategies, which attenuated the linear relationship between frequency and reading times.
It should be noted that the L2-AL group was linguistically heterogeneous. Some learners came from languages such as Vietnamese and Korean, which contain substantial Chinese-derived vocabulary or historical links to Chinese. Although these languages lack the extensive character-level orthographic overlap with Chinese found in Japanese, such linguistic connections may have reduced the contrast between the two L2 groups, thereby attenuating group differences in word frequency effects. Similarly, Japanese is not a purely logographic system but a mixed writing system combining kanji and kana, which may also have weakened the distinction between the two groups. These factors may partly account for the absence of significant differences in L2 word frequency effects between the two groups. Nevertheless, Japanese remains substantially closer to Chinese than alphabetic writing systems because of its use of kanji. Thus, the contrast between Japanese and alphabetic-language learners of Chinese remains a meaningful and informative test of L1–L2 writing-system similarity.
It is also worth noting that the word frequency effect for native speakers was only marginally significant in some eye movement measures (e.g., gaze duration and regression path duration). One possible explanation is that word frequency effects in natural reading may be less pronounced than those typically observed in isolated word recognition tasks (Kuperman et al., Reference Kuperman, Drieghe, Keuleers and Brysbaert2013; Schilling et al., Reference Schilling, Rayner and Chumbley1998; Xiong et al., Reference Xiong, Yu, Veldre, Reichle and Andrews2023). In addition, the pattern of word frequency effects varied across individual eye movement measures in native speakers. This variability likely reflects differences in sensitivity across eye movement measures and the relatively small effect size of the frequency manipulation in native speakers.
Despite its contributions, the present study has several limitations. First, the alphabetic L1 group comprised learners from diverse native language backgrounds. Future research could examine more homogeneous alphabetic L1 groups with minimal historical and lexical overlap with Chinese to further clarify the role of cross-language similarity. Second, the L2 learners in the present study were relatively high in Chinese proficiency. It remains possible that cross-language competition effects are more likely to emerge at lower proficiency levels, when lexical representations are less entrenched. Future research including learners across a broader proficiency range would allow a more direct test of whether proficiency and cross-language competition modulate L2 word frequency effects. Third, we adopted a categorical manipulation of word frequency (high versus low), which limits the ability to capture the full shape of the frequency effect, particularly in the mid-frequency range where group differences may be most informative (Brysbaert et al., Reference Brysbaert, Lagrou and Stevens2017, Reference Brysbaert, Mandera and Keuleers2018). Future research should include words across a broader frequency range and apply non-linear analyses to provide a more fine-grained account of frequency effects across groups.
5. Conclusion
In sum, the present study showed that L2 Chinese readers exhibited larger word frequency effects than native speakers, and critically, these effects did not differ between learners from alphabetic and Japanese L1 backgrounds. The absence of L1-related modulation suggests that the enlarged word frequency effects in L2 Chinese reading are more likely attributable to differences in lexical entrenchment rather than cross-language competition. These findings therefore highlight the important role of lexical representation strength in shaping L2 word recognition, even across typologically distinct L1 backgrounds.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S136672892610193X.
Data availability statement
The data that support the findings of this study are openly available at https://osf.io/yw9me/overview?view_only=5d0af8a6a774452eb4e1705eaeab4bcf.
Acknowledgments
We would like to thank the editor and anonymous reviewers for their insightful and encouraging feedback on earlier versions of this paper.
Competing interests
The authors declare no potential conflicts of interest.
Ethics approval
The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.
Disclosure of use of AI tools
No generative AI or AI-assisted tools were employed in the preparation of this manuscript.
Open access
