Thirty-one English majors who had passed TEM-8 (Test for English Majors-Band 8) and thirty-two non-English majors who had scored between 430 and 480 on the CET-4 (College English Test Band 4) but had not passed the CET-6 (College English Test Band 6) were recruited as our participants. Of all the participants, the data of five participants (1 English majors and 4 non-English majors) were excluded from further analyses as their accuracy rate was below 60%. The data to enter into final analysis consisted of 30 English majors (age: M = 22 years, SD = 1.11 years, ranging from 20 years to 24 years; gender: 10 males, 20 females) and 28 non-English majors (age: M = 21 years, SD = 1.17 years, ranging from 19 years to 23 years; gender: 10 males, 18 females). Before the experiment, all the participants were required to complete an adapted version of language history questionnaire (Li et al., 2006) in which they needed to report their AoA (Age of Acquisition) of English and make a self-assessment on their English listening, speaking, reading and writing abilities (five points for full marks for each ability). Besides, they also needed to complete the Oxford Placement Test (QPT), a standardized test for English proficiency. Results of the questionnaire and QPT scores are given in Table 1.

English majors (M = 53.10, SD = 1.77) scored significantly higher than non-English majors (M = 36.18, SD = 1.68) in the QPT (t (56) = 37.312, p < 0.001). Two independent-sample t-test on participants’ total scores of four English abilities also showed that English majors (M = 13.05, SD = 1.86) scored significantly higher than non-English majors (M = 10.75, SD = 2.09; t (56) = −4.421, p < 0.001). At the same time, the self-rated score on English reading ability was positively correlated with the OPT score for both English majors (r (28) =0.817, p < 0.001) and non-English majors (r (26) = 0.899, p < 0.001). On account of this, we took English majors as HG learners and non-English majors as LG learners. Both HG and LG learners are late second language learners, with no significant difference in their AoA of English (HG: M = 9.75, SD = 1.97; LG: M = 8.88, SD = 2.01; t (56) = −1.651, p = 0.104). All participants were right-handed (Oldfield, 1971) native Chinese speakers with normal or corrected to normal vision and did not have any psychiatric or reading disorders. Each participant signed a formal inform consent before the experiment and were paid for their participation after the experiment. The experiment was approved by the Ethics Committee of Qufu Normal University.

Our stimuli consisted of 50 pairs of sentences as shown in (5). Of the 50 pairs of sentences, 24 pairs were adapted from the sentences in no-constraint conditions in Kush and Dillon (2021)’s Experiment 2, using familiar names and words to substitute for the corresponding unfamiliar counterparts, and the other 26 pairs were constructed by referring to the sentence structures in Kush and Dillon (2021). The 50 pairs of experimental sentences were selected from a total of 80 pairs of sentences. Thirty-four LG learners who did not participate in the formal experiment but were selected from non-English majors based on the same criteria as in formal experiment were recruited to rate the comprehensibility of all the sentences on a 5-point Likert scale (1 = totally incomprehensible, 5 = totally comprehensible). As a result, the sentences with over 3.75 rating score on average were chosen as the experimental stimuli (M = 4.33, SD = 0.31) so that all were equally understood by both HG and LG learners.

After a stranger bought him an iced coffee, Linda generously offered Victor the best seat at the meeting table.

As shown in (5), all the stimuli are made up of two parts, the subordinate clause (headed by while, before or after) as the first part and the main clause as the second part. The cataphoric pronoun occurs in the direct object position of the subordinate clause with an indefinite noun phrase (e.g., a stranger) or an indefinite pronoun (someone, anyone) as the subject. The main clause contains two names, one in subject position (followed by an adverb) and the other in object position. Adverbs following the matrix subject name were those with a weak pragmatic function, such as casually, generously, kindly and the like. All the adverbs were selected from the lexical syllabus for CET-4 and acquired an average rating score of more than 3.75 on a 5-ponit Likert scale (M = 4.01, SD = 0.58) in a familiarity pretest (1 = totally unfamiliar, 5 = totally familiar) by the same group of LG learners who participated in the sentence comprehensibility rating. The gender of the cataphoric pronoun was manipulated to either match or mismatch the gender of the matrix subject. The gender of the matrix object name was designed to match that of the cataphoric pronoun so that the pronoun could be resolved intra-sententially. All the sentences in gender-match conditions were counterbalanced so that the cataphoric pronouns him and her appeared evenly in the stimuli.

The experimental sentences were divided into two lists by a Latin-Square design, with 75 extra sentences as fillers which were similar to the target sentences in length and complexity, but did not contain cataphoric pronouns. To ensure that participants could attend to the reading task, 25 experimental sentences and 38 filler sentences were followed by a comprehension question for each. In case participants did not resolve cataphoric pronouns during reading, 7 out of the 25 comprehension questions following the experimental sentences required the interpretation of cataphoric pronouns (e.g., Who got an iced coffee?) and the remaining questions asked about argument roles in the main clause or subordinate clause (Who offered someone the best seat?/Who got the best seat?/Who bought someone an iced coffee?).

As illustrated in (5), four regions (underlined parts) were selected for experimental analyses: Region 1 (e.g., Linda), the main clause subject (the first potential antecedent of the cataphoric pronoun); Region 2 (e.g., generously), the adverb following the matrix subject, serving as the spillover region of Region 1; Region 3 (e.g., Elsa/Victor), the direct object in the main clause (the other potential antecedent for the cataphoric pronoun); Region 4 (e.g., the best), the spillover region of Region 3, containing two words immediately following Region 3.

The whole experiment was carried out in a quiet laboratory room. Participants were seated ~75 cm away from the display screen and 60 cm away from the eye-tracker. The eye-movement data were recorded using the desk-mounted Eyelink 1000 plus eye-tracker (SR Research, Toronto, ON, Canada) with a sampling rate of 1,000 Hz and a refresh rate of 150 Hz. Right eye was recorded. All the experimental sentences were displayed in 19-point Consolas font across two lines on a 19-inch computer screen.

Each participant was randomly assigned to one of the two presentation lists in which sentences were presented in a pseudo-randomized order such that two sentences in the same condition would not appear more than two consecutive times. Before the formal experiment, a nine-point calibration procedure was performed, followed by a practice session of five sentences for each participant. Drift corrections were performed between trials. All participants were instructed to read the sentences in a natural speed. After finishing reading the sentences, they should press the space key to answer questions or move into next trial. Once a question sentence appeared, participants should press either “F” or “J” to make a response, with the two keys counterbalanced across participants. The whole experiment was comprised of two sessions with a short interval in between. The whole experiment lasted about 40–50 min for HG learners and 60–70 min for LG learners.

Following Drummer and Felser (2018), we reported four eye-movement measures in each of the four regions: First Fixation Duration (FFD), First Pass Reading Time (FPRT), Regression Path Duration (RPD) and Total Reading Time (TRT).

Before exporting the eye-movement data, fixations that were shorter than 80 ms but within one character space of the previous or next fixations were merged with their neighboring fixations. The remaining fixations that were shorter than 80 ms or longer than 800 ms were excluded because they could not reflect proper language processing (Rayner and Pollatsek, 1989; Rayner, 1998). For each eye-movement measure, the data above or below three standard deviations were removed from further analyses. For HG learners, this accounted for 5.76% of the data in the by-participant analysis and 6.12% of the data in the by-item analysis. For LG learners, this accounted for 4.91% of the data in the by-participant analysis and 4.57% of data in the by-item analysis.

Two-way mixed repeated measures ANOVAs with participants (F1) and items (F2) as random variables were carried out for each eye-movement measure with between-subject factor Group (HG vs. LG) and within-subject factor Gender Match (yes vs. no) using SPSS Statistics 26.0. Table 2 displays two participant groups’ mean reading times on the four regions per condition and Table 3 provides a summary of ANOVAs for each region.

Experiment 1 intended to explore whether Chinese EFL learners initiated an active searching process in cataphora resolution and whether learners’ proficiency level influenced this searching process. The eye-movement data revealed a GME effect (Gender Match Effect) for all participants, but the effect was observed earlier in HG learners than in LG learners.

Contrary to a GMME effect that has been consistently observed in previous studies on cataphora resolution (van Gompel and Liversedge, 2003; Kazanina et al., 2007; Kazanina and Phillips, 2010; Clackson and Clahsen, 2011; Yoshida et al., 2014; Pablos et al., 2015; Patterson and Felser, 2019), our study unexpectedly revealed a GME effect at region 1 (subject name) and region 2 (adverb), suggesting that a gender-matching potential antecedent elicited longer reading times than its gender-mismatching counterpart at the matrix subject position. GMME effect has been taken as an indication of readers’ employment of active search mechanism in cataphora resolution, so did the absence of GMME effect mean Chinese EFL learners did not resolve the cataphoric pronoun in an active way? It seems hasty to jump to such a strong conclusion. Since readers’ active search for the cataphor’s antecedent implies a link has already been established between the cataphoric pronoun and the first syntactic position that could contain a noun phrase even before the information (e.g., gender and number) about the noun phrase is accessed, the nature of the active search mechanism lies in whether readers predictively link a cataphoric pronoun to the first syntactic position that could contain a potential antecedent. Based on this rationale, we believe that Chinese EFL learners were also engaged in an active search for the antecedent in the cataphora resolution process, otherwise the gender features of the first potential antecedent would not affect their reading times (though in a different direction from English natives). However, this active searching process adopted by Chinese EFL learners was distinct from the one used by English natives in that only a gender-matching antecedent could prompt Chinese EFL learners to engage in a deep processing of this antecedent, contrasting with English natives’ deep processing of both gender-matching and gender-mismatching antecedents, consequently resulting in GME effect for Chinese EFL learners while GMME effect for English natives. As for the reason why Chinese EFL learners behaved differently from English native speakers when facing a gender-mismatching name at the matrix subject position, we will elaborate it in the General Discussion part.

Consistent with our predictions, LG learners were found to behave differently from HG learners in the processing of cataphoric pronouns. HG learners exhibited a GME effect in first pass reading times, regression path durations and total reading times at region 1, while LG learners only showed a GME effect in total reading times at region 2, i.e., gender features of subject names did not affect LG learners’ processing time until region 2 (the spillover region of the subject name). This suggests that LG learners, relative to HG learners, delayed their active searching process in cataphora resolution.

Significantly longer reading times in gender-match than in gender-mismatch conditions were also observed in first pass reading times, regression path durations and total reading times at region 3 (object name region). However, the gender-match effect at region 3 was different from that at region 1 and region 2. The name at region 3 provides another antecedent candidate in gender-match conditions while it is the only suitable antecedent in gender-mismatch conditions. Therefore, the longer reading times observed in gender-match than in gender-mismatch conditions might due to the competition between two gender-matching names for acting as the antecedent of the cataphoric pronoun. In addition, it appeared that HG learners were earlier influenced by the competition between subject name and object name in gender-match conditions as indicated by longer first fixation durations in gender-match conditions observed only for HG learners.

In summary, the results of Experiment 1 suggest that Chinese EFL learners also adopt an active search mechanism to resolve cataphoric pronouns, yet along a processing path distinct from English natives’. Specifically, Chinese EFL learners predictively link a cataphoric pronoun to the first potential antecedent in the sentence but only a gender-matching antecedent can prompt them to engage in deep processing of the antecedent. Moreover, the learners’ proficiency level modulates the time-course of their active searching process, with HG learners initiating an active search for the antecedent earlier than LG learners.

Both HG and LG Chinese EFL learners initiate an active search for the antecedent after encountering a cataphoric pronoun. Whether this searching process is also constrained by Binding Principles in the same way as English natives has not been clear, the issue Experiment 2 was to explore.

The recruitment criteria for participants were the same as in Experiment 1. A total of 27 English majors and 30 non-English majors participated in this experiment. The data of three participants (all non-English majors) were discarded as their accuracy rate was below 60%. The data for final analysis comprised of 27 English majors (age: M = 22.17 years, SD = 1.26 years, ranging from 20 years to 26 years; gender: 9 males, 18 females) and 27 non-English majors (age: M = 20.88 years, SD = 1.19 years, ranging from 19 years to 23 years; gender: 10 males, 17 females). All participants completed an adapted version of language history questionnaire (Li et al., 2006) and the QPT before the experiment. The results of the language history questionnaire and QPT scores are provided in Table 4.

As in Experiment 1, English majors (M = 52.37, SD = 1.52) were more proficient than non-English majors (M = 35.81, SD = 1.75; t (52) = 37.023, p < 0.001), which was also confirmed by the results from participants’ total scores of four English abilities, showing that English majors (M = 13.41, SD = 2.02) scored significantly higher than non-English majors (M = 11.22, SD = 2.68; t(52) = −3.433, p = 0.001). Similarly, self-rated score on English reading ability was positively correlated with the OPT score for both English majors (r (25) = 0.749, p < 0.001) and non-English majors (r (25) = 0.863, p < 0.001). Therefore, we grouped English majors as HG learners and non-English majors as LG learners. Both two participant groups were late second language learners, with no significant difference in their AoA of English (HG: M = 8.80 years, SD = 1.90 years; LG: M = 9.38 years, SD = 1.74 years; t (52) =1.146, p = 0.257). All participants signed a written inform consent before the experiment and were paid for their participation after the experiment.

The materials consisted of 60 sets of sentences as shown in (6). Each sentence was made up of one main clause and one fronted participle clause headed by while, before or after. All the materials were constructed following a two-factors (Constraint and Gender-Match) within-subjects design. The factor Constraint manipulated whether the coreference between the cataphoric pronoun and the matrix subject was constrained by Binding Principle B. In constraint conditions, the cataphor was the direct object of the infinitival verb while in no-constraint conditions, the cataphor was a possessive pronoun embedded in the direct object noun phrase. Across constraint and no-constraint conditions, the gender of the cataphoric pronoun was manipulated to either match or mismatch the gender of the matrix subject. As in Experiment 1, all the adverbs following the matrix subjects were selected from the lexical syllabus for CET-4 and met the requirement of an average rating score of more than 3.75 on a 5-ponit Likert scale (M = 4.12, SD = 0.46) in a familiarity pretest (1 = totally unfamiliar, 5 = totally familiar) by an additional group of 60 LG learners who were selected from non-English majors based on the same criteria as in formal experiment. There was a gender-matching name at the matrix object position across four conditions to ensure that each cataphoric pronoun was to have an intra-sentential referent.

Of the 60 sets of target sentences, 24 sets were adapted from Kush and Dillon (2021)’s Experiment 1 and the other 26 sets were constructed following the same sentence patterns. The 60 sets of target sentences were selected from a total of 85 sets of sentences. The same group of LG learners who participated in the familiarity test on adverbs were also asked to rate the comprehensibility of those sentences on a 5-point Likert scale (1 = totally incomprehensible, 5 = totally comprehensible). Sentences with 3.75 rating score on average were selected as the experimental items (M = 4.34, SD = 0.28).

After buying her an iced coffee, Linda generously offered Elsa the best seat at the meeting table.

After buying him an iced coffee, Linda generously offered Victor the best seat at the meeting table.

After buying her colleague an iced coffee, Linda generously offered Elsa the best seat at the meeting table.

After buying his colleague an iced coffee, Linda generously offered Victor the best seat at the meeting table.

To ensure that each participant would receive a sufficient number of target sentences from each condition, the 60 sets of experimental sentences were divided into two lists in such a way that each participant could see two sentences from one set, one in constraint condition, and the other in no-constraint condition. Each list consisted of 120 experimental sentences (30 experimental sentences for each condition) and mixed with equal 120 fillers. As in Experiment 1, the fillers were of similar length and complexity to the experimental sentences and did not contain cataphoric pronouns. Similarly, half of the experimental sentences and filler sentences were followed by a comprehension question. In case participants did not resolve cataphoric pronouns during reading, 12 out of the 60 comprehension questions following the experimental sentences required the interpretation of cataphoric pronouns (e.g., Who got an iced coffee?/Whose colleague got an iced coffee?) and the remaining questions mainly targeted the interpretation of argument roles in the main clause (e.g., Who offered someone the best seat?/Who got the best seat?) or the interpretation of the implicit subject in the adjunct clause (e.g., Who bought someone an iced coffee?).

Also similar to Experiment 1, four regions were selected for further analyses. Region 1 was the matrix subject name; Region 2 was the adverb following the matrix subject; Region 3 was the matrix object name; Region 4 consisted of two words that came after Region 3.

The procedure was identical to Experiment 1. The formal part of the experiment was divided into three blocks with a short break in between. After each break, a nine-point calibration procedure was conducted again. All the stimuli in each presentation list were presented in a pseudo-randomized order such that two sentences in the same condition would not consecutively appear more than twice. The entire experiment lasted about 70–80 min for HG learners and 80–90 min for LG learners.

The same four eye-movement measures were reported as in Experiment 1: first fixation duration, first pass reading time, regression path duration and total reading time.

Before exporting the eye-movement data, fixations shorter than 80 ms but within one character space of the previous or next fixations were merged with their neighboring fixations. The fixations that were shorter than 80 ms or longer than 800 ms were excluded as well. For each eye-movement measure, the data that were above or below three standard deviations were discarded from further analyses. For HG learners, this accounted for 7.90% of data in the by-participant analysis and 5.66% of data in the by-item analysis. For LG learners, this accounted for 6.14% of the data in the by-participant analysis and 6.28% of data in the by-item analysis.

Three-way mixed repeated measures ANOVAs with participants (F1) and items (F2) as random variables were carried out for each eye-movement measure with between-subject factor Group (HG vs. LG) and within-subject factors Constraint (yes vs. no) and Gender Match (yes vs. no) using SPSS Statistics 26.0. Table 5 displays two participant groups’ mean reading times on the four regions per condition and Table 6 provides a summary of ANOVAs for each region.

Experiment 2 was conducted to further examine whether Chinese EFL learners’ cataphora resolution process was restricted by the syntactic constraint Binding Principle B and whether the processing pattern varied with the learners’ proficiency. Interpretations of the major results are presented in the following.¹

Main effects of Gender Match were found for early eye-movement measures (first fixation durations and first pass reading times) at region 2, the spillover region of region 1. A gender-mismatching subject name elicited significantly longer reading times than a gender-matching one in both constraint pair and no-constraint pair. This GMME effect unaffected by the factor Constraint suggested that Chinese EFL learners tried to establish coreference relations between the cataphoric pronoun and the matrix subject in both constraint and no-constraint conditions, even if the coreference in constraint conditions violated Binding Principle B. The expected interaction between Constraint and Gender Match, however, just occurred in the late eye-movement measure (regression path durations) at region 2. A gender-mismatching subject induced more regressions in no-constraint conditions than in constraint conditions. In other words, the gender-incongruency between the cataphoric pronoun and the matrix subject only induced greater processing difficulty when the coreference between the cataphoric pronoun and the matrix subject was free from the constraint of Binding Principle B, suggesting that syntactically-illicit subject in constraint conditions was no longer considered as the potential antecedent of the cataphoric pronoun. Therefore, the early and late eye-tracking data at region 2 together demonstrated that Binding Principle B did not constrain Chinese EFL learners’ early processing of cataphoric pronouns, but only played its role in the late stage of cataphora resolution. Meanwhile, Chinese EFL learners’ delayed application of Binding Principle B also came up in the different effects of Constraint on first fixation durations and total reading times at region 2. While the main effect of Constraint on first fixation durations indicated that learners had more difficulty in processing constraint sentences, the main effect of Constraint on total reading times at the same region showed that the learners had more difficulty with no-constraint sentences. This indicates that in early processing, Chinese EFL learners were considering a potential antecedent which should be ruled out by Binding Principle B, but in later processing, they only considered a syntactically-licit antecedent for the cataphoric pronoun, which was consistent with the conclusion drawn from the effects of Gender Match and its interaction with Constraint.

Further evidence supporting Chinese EFL learners’ late application of Binding Principle B was also found in late eye-tracking data at region 3. In this region, the object elicited more regressions in no-constraint than in constraint conditions. In no-constraint conditions, both the subject and object could be the antecedent of the cataphoric pronoun in terms of their syntactic position. In constraint conditions, however, only the object was licensed to co-refer with the cataphoric pronoun. If Chinese EFL learners abided by Binding Principle B in the late processing stage, the matrix subject would not interfere with the late integration stage during cataphora resolution in constraint conditions, but would exert its influence in no-constraint conditions. As a result, more regressive eye-movements from the object would be made in no-constraint than in constraint conditions which was just the pattern we observed in regression path durations at region 3. Besides, this effect of Constraint kept constant at region 4 as indicated by longer regression path durations and total reading times in no-constraint than in constraint conditions. In spite of Chinese EFL learners’ late application of the binding constraint, however, the gender feature of the syntactically-illicit name (i.e., matrix subject in constraint conditions) could still exert its influence on the integration stage in cataphora resolution, which can be seen from the longer total reading times in constraint, gender-match conditions than in constraint, gender-mismatch conditions at region 3.

It is worth-noting that a significant Constraint by Gender Match interaction was also observed in first fixation durations at region 1 (matrix subject name). We did not take this interaction as evidence that Principle B was early applied by Chinese EFL learners to constrain the cataphora resolution process because this interaction was only found in first fixation durations at region 1 and was absent in other early eye-movement measures in this region and its spillover region (region 2). Considering the experimental materials used in our experiment were quite different from what were used in previous cataphora resolution studies, the interpretation of this seemingly odd interaction should start from the structure of our own materials. It might be related to the process in which matrix subject is first assigned as the subject of the adjunct clause. This may also explain why the effect of Gender Match manifested itself in the form of longer reading times in gender-mismatch than in gender-match conditions. A detailed explanation is given in the General Discussion part.

The experimental data also show that learners’ proficiency did not influence their processing pattern in cataphora resolution. Except for the overall longer reading times observed in LG learners, LG and HG learners behaved remarkably alike in terms of the time-course during which Binding Principle B was applied to constrain their cataphora resolution process, suggesting that learners’ proficiency did not affect when the syntactic constraint was applied in cataphora resolution. However, it seems that LG learners are less skilled in applying Binding Principle B than HG learners during late stages of cataphora resolution, as indexed by the Constraint and Group interactions observed for total reading times at both region 1 and region 3. Moreover, the Constraint and Group interaction observed for first pass reading times at region 2 seems to suggest that LG learners have more difficulty in retrieving the gender feature of the cataphoric pronoun when they started the cataphora resolution process just after finishing the interpretation of PRO. This is supposed to arise from the fact that the LG learners in our Experiment have lower working memory than HG learners. L2 learners’ working memory has been found to positively correlate with their proficiency (van den Noort et al., 2006).

In brief, Binding Principle B constrains Chinese EFL learners’ cataphora resolution process, but just works at late stages of cataphora resolution. As a whole, the time-course of learners’ application of Binding Principle B in English cataphora processing is not modulated by their proficiency.

Just as English natives, Chinese EFL learners also trigger an active search for the antecedent upon encountering an English cataphoric pronoun. However, the specific searching process is quite different from that of English native speakers. Reading times of English natives are slowed down by a gender-mismatching name at the matrix subject position while reading times of Chinese EFL learners are slowed down by a gender-matching name at the same position. This unexpected gender-match effect contradicts the gender-mismatch effect that has been consistently found in previous studies except for Bertenshaw (2009). Bertenshaw (2009) ascribed the gender-match effect observed in Japanese EFL learners to the influence of learners’ first language in which disjointed coreference between the cataphoric pronoun and the first name was preferred in the context of experimental sentences. However, this explanation cannot account for the gender-match effect in our study for two reasons. The first reason is that experimental sentences in Bertenshaw’s study differed from what we used. In Bertenshaw’s (2009) study, a lead-in sentence containing one male name and one female name preceded the critical sentence containing the cataphoric pronoun, which made the pronoun not strictly cataphoric since it can access interpretation from the male or female name based on its gender co-indexation. Therefore, the existence of the lead-in sentence might be the real reason that Japanese EFL learners’ first language had an impact on their interpretation of the cataphoric pronoun. The second reason is that Chinese EFL learners do not possess prior knowledge regarding the resolution of cataphoric pronouns, for there is no intra-sentential cataphora in Chinese (Wang, 1994; Wang, 2000, 2006; Zhao and Shao, 2002). Then how to account for the gender-match effect in our study? One possibility is that Chinese EFL learners are affected by their awareness of first language in the option of processing strategies in cataphora resolution. The transfer of L1 processing strategies in L2 sentence processing has been confirmed by a number of studies (e.g., Gass, 1987; Harrington, 1987; Kilborn and Cooreman, 1987; Kilborn, 1989; Hernandez et al., 1994; Su, 2001). As Chinese is a semantics-driven language (Xu, 1999), Chinese native speakers are more concerned with semantic congruity than syntactic congruity when processing sentences. This processing preference might be transferred to their L2 sentence processing, which makes them less error-tolerant in semantic plausibility than English natives upon encountering a gender-mismatching potential antecedent in cataphora resolution. As mentioned above, before the gender feature of the matrix subject name becomes available, a link between the cataphoric pronoun and the name has already been established by both English native speakers and Chinese EFL learners. When the subject name mismatches the cataphoric pronoun in gender, Chinese EFL learners can quickly revoke the previously established coreference relation and stop taking the subject name as a potential antecedent. By contrast, when the information of that name matches the cataphoric pronoun in gender, they retain the link and engage in a deep processing of the subject name. As a result, more reading time was spent on a gender-matching subject name compared to a gender-mismatching name. However, a gender-mismatching subject name cannot make English native speakers abandon the established coreference relation in a short time as they are more error-tolerant than Chinese EFL learners in English sentence processing. Consequently, whether the subject name matches or mismatches the cataphoric pronoun in gender, English native speakers keep engaged in a deep processing of the name, yielding greater processing difficulty at the gender-mismatching relative to the gender-matching name region. From a cognitive perspective, the L1 transfer found in our study is conceptual transfer whose basic idea is that a person’s comprehension and production of one language is affected by the concepts and conceptualization patterns he acquires in another language (Jarvis, 2007, 2011). Experientialism argues that although people of different nations share the ability to conceptualize their experience, they have developed different conceptual systems due to differences in their specific life experiences, environments, and cultures (Lakoff, 1987). Therefore, Chinese natives differ from English natives in terms of their specific conceptual system and conceptual patterns, as revealed by their language typology (Chinese, a semantics-first language contrasts English, a syntax-first language). As a result, Chinese EFL learners’ processing of English sentences in our study is subject to the influence of their native language.

At the object name region (region 3), the regression path durations and total reading times were significantly longer in gender-match conditions than in gender-mismatch conditions. This can be explained by the Competition Model (McDonald and MacWhinney, 1995). According to the Model, processing should be much more difficult when there are two potential antecedents than when there is only one for the anaphoric pronoun, because the two potential antecedents might compete for the final interpretation of the anaphoric pronoun whereas there is no competition involved when only one potential antecedent is available. In our Experiment 1, the object name in gender-match conditions serves as another potential antecedent for the cataphoric pronoun, competing with the subject name for being chosen as the final interpretation of the cataphoric pronoun during later stages of cataphora resolution. In contrast, the object name in gender-mismatch conditions is the only appropriate antecedent for the cataphoric pronoun and no competition will arise between the object name and the subject name, for the gender feature of the subject name clashes with that of the cataphoric pronoun. Evidently, Competition Model can not only be used to reveal the processing mechanism behind ambiguous forwards anaphora resolution, but also holds explanatory power in ambiguous cataphora resolution as in our study.

Apart from the two above-mentioned findings, Experiment 1 also demonstrated a modulating effect of learners’ proficiency on the time-course of their active searching process. Specifically, LG learners delayed their active searching process relative to HG learners. Following the above-mentioned L1 transfer account, LG learners should be more affected by their L1, thereby showing earlier and stronger GME effect. However, what we found was just the opposite. HG learners exhibited earlier and stronger GME effect than LG learners, which means there was another force playing a greater role. This dominating force stems from learners’ employment of active search mechanism in cataphora resolution. As we have discussed in the preceding text, readers’ employment of an active search mechanism implies two steps: the first step is to predictively establish the coreference relationship between the cataphoric pronoun and the matrix subject position, and the second step is to access the gender feature of the matrix subject to confirm their analysis. Accordingly, two reasons might help to explain our results. One is related to LG learners’ reduced predictive ability compared to HG learners in L2 sentence comprehension, which has been demonstrated by several studies (e.g., Kaan, 2014; Peters et al., 2015). As a result of LG learners’ reduced predictive ability, the predictive process of linking the cataphoric pronoun to the matrix subject position (region 1) might be slower and more cognitive-demanding for LG learners than HG learners. The other reason for our results might be that LG learners are slower and less efficient than HG learners in accessing the gender features of the matrix subject. According to the revised hierarchical model (RHM), L2 and L1 words in the bilingual mental lexicon share one conceptual system while their corresponding forms are represented separately (Kroll and Stewart, 1994). Several scholars (Chen, 1990; Dufour and Kroll, 1995; Cheung and Chen, 1998) interpreted the RHM from a developmental perspective and held that L2 learners with lower proficiency access L2 word meaning through L1 lexicon while learners with higher proficiency access L2 word meaning directly via the shared conceptual system, the claim being confirmed by several studies (e.g., Jared and Kroll, 2001; Kroll et al., 2006, 2010). Therefore, it is rational to assume that LG learners in our study are slower than HG learners in accessing the gender information of the matrix subject. In a word, LG learner’s reduced predictive ability and slower access to the gender information of the matrix subject together contribute to their relatively delayed employment of the active search mechanism in cataphora resolution.

Chinese EFL learners’ cataphora resolution process was constrained by Principle B. However, unlike English native’s early application of Principle B, Chinese EFL learners, irrespective of their proficiency levels, resorted to Principle B to resolve cataphora at a later stage. Before further exploring the underlying reasons for the different processing patterns between Chinese EFL learners and English native speakers, we first take a look at the seemingly odd Gender Match and Constraint interaction observed in first fixation durations at region 1. As we have mentioned in 3.6, the interpretation of this interaction should be based on the structure of the experimental materials used. For better illustration, the exemplar sentences in Experiment 2 are repeated here as (7).

After PRO buying her/him an iced coffee, Linda generously offered Elsa/Victor the best seat at the meeting table.

After PRO buying her/his colleague an iced coffee, Linda generously offered Elsa/Victor the best seat at the meeting table.

The experimental sentences involve two types of cataphoric relations, one of which is between PRO and the matrix subject and the other is between the cataphoric pronoun and its antecedent. Using the same materials, Kush and Dillon (2021) demonstrated that English native speakers could process the two types of cataphoric relations simultaneously by incrementally integrating syntactic and pragmatic information. However, our experimental data suggest that Chinese EFL learners could only process the two cataphoric relations one by one. Though a couple of studies claim that cataphoric pronouns are not used in Chinese (Wang, 1994; Wang, 2000, 2006; Zhao and Shao, 2002), some scholars take a different view in this issue (Xu and He, 2007; Gao, 2010; Yu, 2011). For instance, Gao (2010) held that cataphoric pronouns existed in Chinese in an invisible and inaudible way, i.e., zero cataphoric pronouns are present in Chinese. Gao (2010) conducted a corpus-based contrastive study of cataphora in Chinese and English, and found that cataphora in Chinese manifested itself only in the form of zero pronouns while cataphora in English existed in the form of both zero and overt pronouns. At the same time, zero cataphoric pronouns in Chinese and English are used in a similar way in which the invisible pronoun appears in a preposed modifier and the antecedent appears in the matrix subject position or as a modifier of the matrix subject. The PRO in our experimental sentence is just the kind of zero cataphoric pronouns Gao (2010) has identified. Since this kind of zero cataphoric pronouns exists in both Chinese and English and is used in a similar way across two languages, Chinese EFL learners might prioritize the processing of PRO (i.e., zero cataphoric pronoun) over the processing of the overt cataphoric pronoun which is rarely used in Chinese. Following this assumption, we can well explain the seemingly odd Constraint and Gender Match interaction observed in first fixation durations at region 1. Specifically, during the early processing of region 1, Chinese EFL learners only completed the interpretation of PRO, assigning it to the matrix subject (region 1) while the resolution of the cataphoric pronoun had not yet started at this time. After the matrix subject was interpreted as the subject of the subordinate clause, the meaning of the subordinate clause in no-constraint, gender-match conditions was accessed more easily than the meaning of the adjunct clause in no-constraint, gender-mismatch conditions due to the gender-incongruency between the matrix subject and the cataphoric pronoun, thereby resulting in significantly longer first fixations in no-constraint, gender-mismatch conditions than in no-constraint, gender-match conditions. At the same time, the process of assigning the matrix subject name to PRO made the gender feature of the matrix subject available to the parser before the resolution of cataphoric pronouns started, consequently leading to longer reading times being spent on a gender-mismatching subject name during cataphora resolution, contrastive to the GME effect observed in Experiment 1 where coreference between the cataphoric pronoun and the matrix subject was already established even before the gender information of the subject name becomes available to the parser. In addition to a Constraint and Gender Match interaction, there came up a significant Gender Match and Group interaction for first fixation durations at region 1. Further analysis on this interaction indicated that LG learners were more easily disrupted than HG learners by the gender feature of the cataphoric pronoun stored in their working memory when assigning the matrix subject to PRO.

Chinese EFL learners behave themselves differently from English natives in using Binding Principle B from the perspective of time course regarding cataphora resolution. Concerning the time-course of syntactic constraints in cataphora resolution, there are two major hypotheses. One is the early filter hypothesis, claiming that syntactic constraints play an early role in the parser’s active searching process so that only positions licensed by the grammatical constraints can be predictively considered as holding potential antecedents. The other one is the delayed filter hypothesis, holding that syntactic constraints are applied at a later processing stage to filter out ungrammatical interpretations, which implies noun phrases at positions not licensed by syntactic constraints are initially considered as potential antecedents. So to speak, the performance of English native speakers in Kush and Dillon (2021)’s study is consistent with the early filter hypothesis while the performance of Chinese EFL learners in our experiment 2 is in line with the delayed filter hypothesis. Why does Principle B function as an early filter in English natives’ cataphora resolution but a delayed filter in Chinese EFL learners’ cataphora resolution? The difference between L1 and L2 sentence processing might be responsible for the discrepancy. According to Shallow Structure Hypothesis (Clahsen and Felser, 2006a,b), L2 learners prioritize semantic and pragmatic cues over syntactic cues to guide their sentence processing, which leads to shallow and less detailed syntactic representations being computed by L2 learners, as compared to native speakers. Given that our experimental sentences involve two types of cataphoric relations, it is rational to assume that Chinese EFL learners have not established the required syntactic representations to correctly resolve the cataphoric pronoun when first encountering the matrix subject. As a result, the EFL learners made the ungrammatical interpretation that is based solely on the gender-match between the cataphoric pronoun and the matrix subject, without taking the syntactic constraint into account.

Previous forwards anaphora resolution studies show that L2 learners tend to delay the application of binding constraints (Felser et al., 2009; Felser and Cunnings, 2012). Targeting at L2 English learners’ online processing of reflexives, both Felser et al. (2009) and Felser and Cunnings (2012) found that L2 learners were initially affected by a discourse-salient but binding-inaccessible antecedent while English native speakers could immediately apply Binding Principle A to resolve reflexives, suggesting that L2 learners’ early processing of reflexives was not restricted by syntactic constraints. In our study, Chinese EFL learners were also found to have delayed their application of the binding constraint in cataphora resolution. Therefore, the delayed application of binding constraints by L2 learners in both forwards anaphora and cataphora resolution adds evidence to the claim that L2 learners have more difficulty than native speakers in establishing nonadjacent syntactic dependencies in real time (Clahsen and Felser, 2006a).

Both Binding Principle B and Binding Principle C are observed by EFL learners in English cataphora resolution. Unlike the delayed application of Binding Principle B by learners in our study, previous cataphora resolution studies demonstrated that native speakers and L2 learners patterned alike in terms of the time-course of their application of Binding Principle C when resolving cataphoric pronouns (Rodríguez, 2008; Bertenshaw, 2009; Drummer and Felser, 2018). This implies that Binding Principle B and Binding Principle C are treated differently by L2 learners. To be more specific, it seems that Binding Principle C was more easily applied than Binding Principle B by L2 learners in their online resolution of cataphoric pronouns. Comparing the materials used in previous L2 cataphora resolution studies and those used in our study, we find that the coreference relations subject to Binding Principle C was more easily detected than those subject to Binding Principle B because the structure of sentences involving Principle C constraint was much simpler than those involving Principle B constraint. On this account, it appears safe to argue that L2 learners’ sensitivity to different binding constraints in cataphora resolution is also dependent on the complexity of sentences involved. Despite L2 learners’ divergent sensitivity to Binding Principle B and Binding Principle C, all L2 learners can follow these two syntactic constraints in the corresponding sentences to exclude ungrammatical antecedents when resolving cataphoric pronouns online.

Chinese EFL learners’ proficiency did not modulate the time-course of their application of Binding Principle B in cataphora resolution. Specifically, both HG and LG learners delayed their application of the binding constraint when resolving cataphoric pronouns. Nevertheless, the data in Experiment 1 showed that learners’ proficiency affected when they initiated an active search for the antecedent. What is dedicated to the absence of the modulation effect of learners’ proficiency in Experiment 2? We resort to Shallow Structure Hypothesis to explain the absence. The key tenet of Shallow Structure Hypothesis is that L2 sentence processing is fundamentally different from L1 sentence processing and even highly proficient learners cannot achieve native-like level in terms of processing complex syntactic structures like ambiguous relative clauses and filler-gap dependencies while native-like processing of simple syntactic structures like subject-predicate agreement can be attained as the learners’ proficiency increases. As mentioned in the preceding part, the sentences in Experiment 2 involve two types of cataphoric relations and the correct resolution of cataphoric pronouns necessitates the successful representation of these two types of cataphoric relations. In contrast, the sentences in our Experiment 1 are relatively simple and the successful resolution of cataphoric pronouns relies more on whether the name matches the cataphoric pronoun in gender. That is, since the sentences were more complex in Experiment 2 than in Experiment 1, learners’ proficiency modulated their online performance in Experiment 1 but became invisible in Experiment 2.