To understand a text, individuals not only represent the text itself but also build mental representations, or situation models, of the scenarios described (Zwaan and Radvansky, 1998; Magliano et al., 1999). An embodied account of language comprehension argues that motor–perceptual representations are integral to constructing these mental representations (Barsalou, 1999; Zwaan, 1999). By activating and integrating motor–perceptual experiences, individuals can gather and elaborate on information about the events described in a text, leading to a high level of language comprehension (Zwaan, 2014; De Vega, 2015). That is, mental representations are multimodal, involving perceptual and motor representations, and are therefore embodied. Constructing mental representations can be challenging for less skilled language users, as it requires higher-order and deep-level processing (Zwaan and Taylor, 2006). Empirical evidence shows that L2 users often find tasks requiring such processing inefficient and effortful (Francis and Gutiérrez, 2012; Horiba, 1996; Pérez et al., 2019). Thus, building mental representations appears particularly challenging for L2 users. Notably, Zwaan and Brown (1996) found that English learners of French developed weak and reduced mental representations of French stories, making fewer inferences, while their mental representations of English stories were strong and comprehensive.

The compatibility effect, frequently observed in studies using a Sentence–Picture Verification Task (SPVT), offers behavioral evidence that individuals mentally create motor–perceptual representations to comprehend sentences in L1. For example, in Zwaan and Pecher's (2012) research on perceptual representation, a facilitating effect (i.e., a shorter reaction time, RT) was observed when the shape/orientation of an object implied in a sentence matched what was seen in a picture. Specifically, RTs were shorter when participants saw an image of a bird with outstretched wings after reading “There is a bird in the sky.” Conversely, an impeding effect, characterized by a longer RT, was observed when there was a mismatch between the sentence and the picture, such as when the sentence was “There is a bird in the nest.” Similarly, Holt and Beilock (2006) employed an SPVT to show that native English speakers mentally construct motor representations when reading action sentences in L1. In summary, the facilitating and impeding effects observed in the SPVT indicate the formation of motor or perceptual representations. Ferstl et al. (2017) investigated whether manipulating action performers' facial appearances and clothing affected viewers' judgments of differences and similarities between two ambiguous actions. Their results showed that visual representations of actions, along with the facial identities of the action performers, influenced viewers' ability to distinguish between the actions. Notably, both Ferstl et al. (2017) and Holt and Beilock (2006) used picture stimuli depicting only the action performers' hands in motion (i.e., representing affordances) without showing the actual objects involved. This similarity provides useful guidelines for displaying picture stimuli in motor embodiment studies.

The Indexical Hypothesis (IH) is an embodied approach to language comprehension. It emphasizes situated action and suggests that the meaning of a situation is determined by the series of actions individuals take to interact with the physical world. Accordingly, sentence comprehension relies on bodily action and involves three processes: indexing, extracting and inferring affordances, and meshing (Glenberg and Robertson, 1999; Kaschak and Glenberg, 2000). To determine whether a sentence is meaningful, individuals index or map words/phrases to their physical referents at the beginning of the process of understanding the sentence. Subsequently, they derive potential affordances related to these references. Affordances represent how individuals interact with referents and their applications in various situations. For example, to move luggage, travelers pull the collapsible handle. Finally, the grammatical structure of the sentence guides the meshing of these affordances to execute the intended action. Here is an example of how this process works: “He hangs the jacket on the collapsible handle of upright luggage.” The syntactic structure of this sentence guides readers to integrate specific actions implied in the sentence to achieve the goal of hanging up the jacket, such as extending the collapsible handle from the wheeled bag of the luggage and draping the jacket over it. However, meaninglessness arises when affordances are meshed into impossible actions (e.g., “he hangs the jacket on the upright bowl”). The ability to successfully mesh affordances into a coherent set of actions depends on the relationship between the affordances of the individual (i.e., he), the objects involved (i.e., jacket, luggage/bowl), and the goal specified in the sentence context.

Santana and De Vega (2013) illustrated the action-based sentence comprehension model. Their study found an enhanced N400 effect when participants read sentences describing a protagonist performing two actions simultaneously (e.g., while applying ointment to her wounded hand, she stretched a bandage). N400 is a brain signature sensitive to semantic or world knowledge inconsistencies. However, Santana and De Vega (2013) explained that the increased N400 observed in their study demonstrated a motor incongruency effect rather than a semantic violation, as their sentence stimuli did not contain semantic anomalies. Moreover, the two actions (i.e., stretching a bandage and applying ointment) were consistent with the scene described in the sentences, indicating that world knowledge violations did not contribute to the increased N400. This explanation was further supported by another experiment from the same study, which found that the N400 effect diminished when the two actions were performed consecutively. They explained further that the observed N400 reflected the unsuccessful formation of motor representations in the brain's motor regions due to a failure in the affordance meshing of the two actions. They also noted that the N400 effect peaked at the end of the sentence (i.e., bandage), not at the mention of the second verb (i.e., stretched). Based on these findings, Santana and de Vega suggested that motor representations are formed when sufficient contextual information is provided. This suggestion appears to conflict with the theory that action verbs alone activate motor embodiment (Hauk et al., 2004). However, Schuil et al. (2013) found that the motor region was activated when action verbs were embedded in literal sentences (e.g., Peter picks up the books after the examination) but less activated in nonliteral sentences (e.g., Peter picks up the pieces after the examination). This highlights the essential role of sentence context in activating and integrating motor experiential traces for developing motor representations.

Oculomotor recording has proven to be a valuable tool for uncovering the mental representations formed for sentence comprehension (Anderson and Spivey, 2009) and has been employed to explore embodied language comprehension. For example, dwell time and fixation counts on the depicted protagonist and destination were shorter and fewer for sentences describing fast actions compared to those describing slow actions (e.g., A man dashed/sauntered into the supermarket). In other words, the speed of a motion described in a sentence influenced the amount of gaze directed at the protagonist (i.e., the man) and the destination (i.e., the supermarket) relevant to the motion (Lindsay et al., 2013; Speed and Vigliocco, 2014). The changes in dwell time and fixation counts on the depicted protagonist and the destination indicate that indexing occurs in two directions. One approach maps words or phrases to their mental/internal representations (Glenberg and Robertson, 2000), while the other links external/physical elements to these mental/internal representations (Spivey and Richardson, 2009). The eye-tracking studies above showed that embodied indexing occurs in two directions simultaneously. Specifically, verbal inputs create mental representations, which are then linked/indexed to the externally and pictorially presented protagonist, destination, and event.

Pavlenko (2014) argued that L2 is disembodied because bilinguals are generally exposed to L2 in grammar-based, amodal symbol-oriented instructional contexts. Using an SPVT, Norman and Peleg (2022) found that Hebrew L2 English readers did not show signs of constructing perceptual representations when comprehending English. Similarly, Chen et al. (2020), employing a delayed SPVT, found no perceptual embodiment in L2 Chinese and L3 English. Conversely, Vukovic and Williams (2014) showed that perceptual, experiential traces were automatically activated when English learners of Dutch processed English sentences describing perceptual information during an SPVT. Similarly, Ahn and Jiang (2018), using an SPVT, found that late L2 Korean learners developed perceptual representations for Korean sentences related to shape or orientation. Of particular interest is Foroni (2015) study, which observed that Dutch L2 English learners' cheek muscles for smiling were stimulated when reading sentences expressing affirmative emotions (e.g., I am grinning). This finding was consistent with his previous L1 study of this kind. However, unlike L1 users in his previous study, L2 readers did not show inhibited cheek muscle activity when reading sentences expressing negative emotions (e.g., I am not grinning). Therefore, Foroni (2015) concluded that emotional language comprehension in L2 is less embodied than in L1 (see Zhang and Vanek, 2021 for a discussion of the processing cost of negation in L2). Regarding motor representations in L2 processing, Vukovic and Shtyrov (2014) observed that the motor experiential traces activated by action verbs were weaker for L2 (English) than for L1 (German). However, Tian et al. (2020) found the opposite. In their study, three types of English and Chinese verb phrases were produced as stimuli, with each conveying a different meaning: abstract, literal, and metaphorical. They found that motor activation strength increased from abstract to metaphorical to literal verb phrases in both languages. However, the motor activation for verb phrases was stronger in L2 (English) than in L1 (Chinese). They concluded that the strong motor responses observed in L2 verb phrases indicated that participants were processing less automatic language rather than solely reflecting motor representations. Therefore, they supported the view that L2 action-related language is processed in a disembodied manner.

Research on the role of perceptual traces in L2 comprehension has yielded mixed results. While previous L2 studies on perceptual embodiment have focused on lexical and sentential levels, studies on motor embodiment in L2 have been limited to the lexical and phrasal levels, often reporting conflicting and indeterminate results. However, L1 studies (Santana and De Vega, 2013; Schuil et al., 2013) have suggested that sentence context is crucial for activating and integrating motor experiential traces. Moreover, these traces are essential for forming motor representations needed to comprehend action sentences. Therefore, it is important to explore whether motor representations are developed at the sentential level in L2, consequently motivating this study to explore the motor aspects of mental representations (i.e., motor representations) formed for L2 action sentence comprehension. This study addresses the following research questions (RQs):

Using the SPVT paradigm and the eye-tracking technique, this study compared L2 readers' eye gaze while viewing matched and mismatched sentence-picture pairs. Specifically, the picture depicted a protagonist performing an action that either corresponded with or differed from the action implied in the sentence. Readers' eye gaze was recorded as they assessed whether the picture matched their comprehension of the sentence stimuli. As mentioned in Section 1.4, fixations index the development of mental representations. Therefore, analyzing participants' eye gaze on the picture will reveal the motor representations they form to comprehend the sentence.

This study operationalized the formation of motor representations using IH. Specifically, the answers to RQ 1 were derived from analyzing how L2 participants gaze at and examine the action and the protagonist's body effectors (i.e., head, hands, and legs) in the picture stimuli. Because action performers' identity (i.e., facial appearance/facial identity) affects observers' ability to correctly identity the actions of the action performers (Ferstl et al., 2017), our first hypothesis (HP) is as follows:

Specifically, participants were expected to index/map the protagonist's head in the picture to the protagonist described in the action sentence. That is, they would index the visual representation of the protagonist to the mental referent of the protagonist formed for comprehending the action sentence. This indexing would allow participants to determine if the depicted protagonist matched the one described in the sentence.

We also examined the relationships between the depicted body effectors, the implied action, and the described object based on the reference status in the sentence of the body effectors. The following hypotheses address affordances and meshing.

In summary, RQ 1 can be answered by examining which parts of the picture stimulus participants focus on and inspect and how frequently they do so. Although this study primarily focused on motor representations of hand actions, the picture stimuli also included the protagonist's legs. It was anticipated that the legs, compared with the hands, would have weaker relationships with the hand action and the described object. Consequently, participants were expected to focus more on the hands than on the legs. When the leg was presented in the picture stimuli, it was used as a distractor to assess whether the participants correctly formed the motor representation of hand actions. Previous studies employing the SPVT paradigm suggest that the facilitating and impeding effects reflect that motor–perceptual representations were formed to comprehend the sentence stimuli before the picture stimuli were displayed (Holt and Beilock, 2006). Therefore, we expected to see the two effects in this study.

Additionally, this study examined the extent to which the L2 mental representations are experientially embodied in motor terms, particularly focusing on the timing effect (i.e., early/late eye movements). Specifically, it investigated how quickly or slowly L2 participants identified actions depicted in picture stimuli that either matched or mismatched the actions described in the sentence stimuli. L2 mental representations tend to be reduced and weaker because of the slow and demanding nature of higher-order and deep-level processing for L2 users (Horiba, 1996; Zwaan and Brown, 1996). Based on previous findings, we propose the following hypothesis:

Overall, 56 right-handed L1 Chinese-speaking students (21 men, 35 women; mean age = 20.4 years, SD = 1.1) with normal or corrected-to-normal vision were recruited from universities in northern Taiwan. To participate, they were required to have taken a reading test from a globally recognized English proficiency test (e.g., TOEFL or IELTS) within the past 2 years and provide a copy of their test score report.

All participants had English reading abilities at the C1 level on the CEFR scale¹. We initially recruited 79 students for the experiment, but 23 of them were excluded from further analyses due to the following reasons: minoring in English, high rates of incorrect responses on verification tasks and comprehension tests, invalid RTs, invalid fixation durations, and technical issues during data collection. Details on data cleaning and analysis are provided in Section 2.7. The remaining 56 participants were majoring in engineering or medicine.

G^*Power software Version 3.1.9.7 (Faul et al., 2007) was used to conduct a priori power analysis to estimate the optimal sample size for the 3 (body effectors) × 2 (match–mismatch conditions) repeated measure design. A large effect size (f²) of 0.35 was used, with an alpha of 0.05, a power of 0.80, and five predictor variables. The results showed that a minimum sample size of 43 was needed to detect a large effect size with an actual power of 0.81. Therefore, our final sample size (N = 56) was adequate. Superpower 0.20 (Lakens and Caldwell, 2021) was used to conduct a simulation-based priori power analysis. We specified a sample size of 20 per cell, a common SD of 1, a common correlation of 0.70, and means of 1, 1.3, 1, 2.05, 0, and 0.1 for the six cells, respectively, with an alpha of 0.05. The results showed that this study would achieve an expected power of 0.90 with an effect size of f² = 0.34. Thus, our final sample size of 28 per cell was adequate.

This study employed the SPVT combined with eye-tracking techniques to investigate how the reference status of body effectors and the match–mismatch conditions in sentences influenced participants' eye gaze on the three body effectors depicted in the picture stimuli. The study featured two repeated measures as independent variables: body effectors and match–mismatch conditions. For body effectors, there were three levels: head, hand, and leg. Each body effector was categorized according to its reference status within the sentence, reflecting the relationships among the three body effectors, the implied action, and the described objects. For example, in the sentence “Mr. Bean is measuring the length of cars in a car park,” the name “Mr. Bean” explicitly identifies the protagonist and, by extension, the Head body effector. The Hand body effector, however, is implicitly involved, as the action “measuring” suggests the use of hands, though it is not directly mentioned in the sentence. The leg body effector, neither explicitly nor implicitly referenced, serves as a distractor. Therefore, the body effectors were categorized as follows:

For match–mismatch conditions, we manipulated whether the action implied in the sentence matched or mismatched the action depicted in the picture (i.e., action-matching sentence vs. action-mismatching sentence).

To answer RQ 1, a 3 (Body effectors: Head, Hand, Leg) × 2 (Conditions: Match, Mismatch) repeated measure design was applied to eye movement data: fixation counts². The independent variables were body effectors and match-mismatch conditions, with fixation counts as the dependent variable. Moreover, two 2 × 2 repeated measure designs were used for further analysis:

Comprised two repeated measures of independent variables: mentioned vs. not-mentioned body effectors and match vs. mismatch conditions, each with two levels. Analysis of (A) determined whether and how these variables influenced participants' fixation counts on the body effectors depicted in the picture stimuli. Similarly, (B) involved two repeated measures of independent variables, each with two levels. Analysis of (B) assessed how these variables influenced participants' fixation counts on the body effectors depicted in the picture stimuli. To address RQ 2, a one-way repeated measures design was used, with match–mismatch conditions serving as the independent variable and the eye movement data for temporal measures³ in the Hand region serving as the dependent variable.

A total of 24 critical sentence–picture item sets were created, each consisting of two sentences and one full-color picture drawn by a professional artist⁴. Each of the two sentences implied a different action (i.e., each item set had two implied actions). Both sentences contained the same verb, and the implied actions were performed by the same body effector of the same protagonist. However, the picture in each item set matched the action implied by only one of the sentences (i.e., action-matching sentence vs. action-mismatching sentence; see Table 1, Figure 1 for examples). During each trial, participants were shown only one sentence from a given critical sentence–picture item set.

The verbs were selected based on their definitions in the Cambridge and Oxford English Dictionaries, with the criterion that neither dictionary could explicitly indicate which body effector executes the action described by the verb. For example, the verb “catch” was excluded because its definition, “to stop and hold a moving object or person, especially in your hands,” implies the use of hands.

This study focused on making inferences about hand- and arm-related actions. While the way an action is performed and the body effector associated with the action were not explicitly stated, they could be inferred from the context. Each sentence presented to the participants explicitly mentioned a protagonist. Sentence length and the protagonist's position in the sentence were consistent across sentence stimuli in both the match and mismatch conditions. In summary, although both sentences in any given sentence–picture item set contained the same verb, the action depicted in the picture matched the action in only one of the sentences. Notably, the specific details of the action were not explicitly mentioned but needed to be inferred from the sentence context. Additionally, filler pairs were included to prevent participants from discerning the purpose of the study. These filler sentences described objects, locations, or general situations and were paired with pictures that were similar but irrelevant to the meaning of the sentences (Table 1, Figure 2). Moreover, yes/no comprehension tests were developed for critical and filler items to verify that participants understood the sentences used in the verification tasks. For instance, the comprehension test for the “boy” sentence–picture item set shown in Table 1 is: “A boy is opening a carton of milk in the metro car.”

A separate group of 10 advanced L2 readers participated in a pilot study. The recruitment criteria of these participants were similar to those for the main study. The pilot study aimed to confirm that none of the words in the sentences—whether critical items or fillers—were unfamiliar to the participants and that the correct action could be inferred from each sentence stimulus. In the pilot study, participants rated the compatibility of each sentence stimulus with the corresponding illustrated action on a 7-point Likert scale, ranging from “very compatible” to “very incompatible.” Sentence–picture pair rated below 5 was excluded. Additionally, items that consistently induced prolonged RTs or unusual eye gaze⁵ were excluded. Ultimately, 12 critical item sets and 24 fillers were retained for the experiment. The average compatibility rating of the critical items in the match condition was 6.5 (SD = 0.60).

In summary, two lists of SPVT item sets (List 1 and List 2) were created. Both lists contained the same filler items. The critical items and sentence conditions (match and mismatch) were counterbalanced across the two lists. This means that if a critical item set presented its match condition on one list, it showed its mismatch condition on the other. Specifically, each picture in a critical item set appeared once in each list, paired with the action-matching sentence on one list and the action-mismatching sentence on the other. Moreover, each list comprised one practice block and three experimental blocks, with an equal number of filler pairs and matching and mismatching pairs in each block. Within each block, the critical pairs and filler pairs were randomly displayed. The order of the three experimental blocks was also randomized. Comprehension tests were administered after each block. Participants were randomly assigned to one of the two lists: half viewed List 1, while the other half viewed List 2.

Participants began with practice trials to familiarize themselves with the task. A 9-point calibration was used to adjust the eye-tracking apparatus. During the experiment, drift correction was performed before each sentence–picture pair presentation. Each sentence was displayed on the screen one at a time, centered and left-justified. A black cross was displayed for 100 ms before each sentence stimulus to direct participants' attention. The black cross was positioned to the left of the center of the screen, on the first letter of each sentence. Once participants had read and understood the sentence, they pressed the space bar. A black cross then appeared in the center of the screen for 100 ms, followed by a blank screen for 250 ms. A picture stimulus was then displayed either to the left or right of the center of the screen. The second black cross was intended to focus participants' visual attention before presenting the picture stimuli.

Participants were instructed to quickly determine whether the picture depicted any elements (e.g., a protagonist, place, object) mentioned in the preceding sentence, and their RTs were recorded. They used the computer keyboard to respond, pressing the “P” key labeled “Yes” if they believed the image was mentioned in the sentence and the “Q” key for “No” if they did not. Participants were not explicitly required to determine whether or not the action depicted in the picture matched the action implied in the sentence. Instead, they were simply asked to respond to “whether any element in the picture had been mentioned in the immediately preceding sentence.” The expected response for all of the critical trials was “yes,” while a “no” response was expected for all the fillers. As noted in Section 2.3, participants were randomly assigned to one of the two SPVT item set lists. Each list included one practice block and three experimental blocks. The comprehension test items were presented at the end of each block. Figure 3 illustrates the SPVT flowchart.

The EyeLink 1000 Plus eye-tracking system (SR Research Ltd.) with a 1,000 Hz sampling rate was used to record eye movements. Only the dominant eye was monitored, although viewing was binocular. Pictures and sentences were displayed on a 21-inch LCD monitor with a 1,440 × 900 screen resolution and a 50 Hz refresh rate. Participants sat 69 cm from the screen, with their heads stabilized by a chin-and-forehead rest to minimize head movement. The picture stimuli subtended a visual angle of 33.27° × 39.85° (width × height). Finally, sentences were displayed in 24-point Courier New font, black on a white background.

This study used a standard SPVT approach, measuring the RT required for participants to determine if anything presented in the picture was mentioned in the immediately preceding sentence. This method helped avoid inadvertently revealing the research objective to the participants during the task. Only RTs for the critical items were included in the statistical analysis. Additionally, eye movement data were analyzed to assess the processing time as participants visually attended to predetermined areas of interest (AOIs) in the picture stimuli.

The measurements are detailed in Table 2. Each body effector (the Hands, Head, and Legs) was assigned a specific AOI. Additionally, an All-Inclusive AOI covered the protagonist in the image (the whole picture) (Figure 4). The four AOIs, shown in the picture of the older woman in Table 1, enabled both general and specific inferences of the data. Each of the six eye movement measures detailed in Table 2 corresponds to a different time event (Godfroid, 2019). The Hand AOI was used to assess data on indicators (a)–(f), the All-Inclusive AOI was used for (e) and (f), and the Head and Leg AOIs for (f). This study focused exclusively on motor representations of hand- and arm-related actions. Therefore, the results for RQ 2, indicated by temporal eye movement measures, reflect the degree of motor embodiment of the Hand action developed by the participants. Conversely, RQ 1 results, shown by fixation counts, revealed which parts of the picture participants examined. Overall, RQ 1 was addressed through fixation counts (f) in the Head, Hand, and Leg AOIs, while RQ 2 was answered through temporal measures (a)–(e) in the Hand AOI.

The accuracy of the responses to the reading comprehension tests and the verification task were recorded. The reading comprehension test was designed to elicit a simple yes/no response (Table 1). The test results reflected the participants' attention to and understanding of each sentence stimulus. The comprehension tests were included based on the idea that people form comprehensive mental representations of what they read only when they read for understanding or learning (Zwaan, 2014). Accordingly, data were removed if the accuracy of either the reading comprehension tests or the SPVT was lower than 80%⁶.

This study used a repeated measures design. All data analyses were conducted using R software, Version 3.6.1 (R Core Team, 2019). Criteria for valid fixation and RT were strictly enforced for data cleaning. If any fixation or RT from a participant did not meet these criteria, all data collected from that participant were removed from the analyses⁷. The data are accessible at this website: https://osf.io/zeufn/s.

The pairwise t-test was conducted to determine whether match-mismatch conditions affected RTs, which required participants to judge compatibility after viewing the picture stimulus. On average, RTs in the mismatch condition (M = 2,179.91, SD = 1,021.51) were significantly longer than in the match condition (M = 1,360.08, SD = 528.46, t(335) = −12.98, p < 0.001, d = 0.98, 95% CI [−944.04, −695.62]). This finding indicates a facilitating effect in the match condition and an impeding effect in the mismatch condition.

When the entire picture was set as a single AOI (whole-picture AOI), the pairwise t-test showed that this region received more fixations in the mismatch condition (M = 5.02, SD = 2.02; M = 3.82, SD = 1.74) than in the match condition (M = 3.82, SD = 1.74; t(335) = −9.33, p < 0.001, d = 0.60, 95% CI [−1.08, −1.67]). Similarly, the paired samples t-test also demonstrated that total viewing time was significantly longer in the mismatch condition (M = 1,802.92, SD = 981.08) than in the match condition (M = 1,061.07, SD = 528.18; t(335) = −12.13, p < 0.001, d = 0.90, 95% CI [−862.10, −621.59]). This indicates an impeding effect in the mismatch condition.

RQ 1 explored which parts of the picture stimulus participants looked at and inspected, and how often they looked at those parts. Fixation count data collected for RQ 1 were analyzed using a linear mixed model approach. The contrasts built for the analyses were match vs. mismatch conditions, mentioned vs. not-mentioned body effectors, and explicitly vs. implicitly mentioned body effectors. Four models were also constructed: the baseline model, body effectors model, match-mismatch conditions model, and interaction model (see Sections 2.2 and 2.7 for details on contrasts and models). To identify the best-fitting model, the four models were compared using the anova() function. The results showed that the body effectors model fit the data better than the baseline model [χ²(2) = 482.81, p < 0.001], and the match–mismatch conditions model fit better than the body effectors model [χ²(1) = 53.45, p < 0.001]. However, the best fit was the interaction model [χ²(2) = 79.05, p < 0.001], as confirmed by the Akaike Information Criterion: the baseline model = 5,747, the body effectors model = 5,268, the match–mismatch conditions model = 5,217, and the interaction model = 5,142. This indicates that the fixation counts were significantly affected by the match-mismatch conditions and the type of body effectors (Match: Hand: M = 1.74, SD = 0.89; Head: M = 1.81, SD = 0.97; Leg: M = 0.06, SD = 0.27; Mismatch: Hand: M = 2.82, SD = 1.19; Head: M = 2.09, SD = 1.07; Leg: M = 0.15, SD = 0.44). Figure 5 shows the interaction graph, illustrating the inspection pattern concerning fixation counts across the three body effectors, focusing on the non-parallel lines. The line representing the match condition illustrates that fixation counts were slightly higher on the Head than on the Hand. However, the line representing the mismatch condition showed a different trend: fixation counts on the Hand were higher than on the Head. This match and mismatch conditions seemed not to affect the Leg. The higher position of the mismatch condition compared to the match condition line indicated that the mismatch condition resulted in higher fixation counts on the Hand and the Head. Further analyses of the interaction are provided in the following sections.

To answer RQ 2, we conducted a series of paired t-tests. The analysis revealed no differences between the match and mismatch conditions in the first fixation duration on the Hand (match: M = 240.71, SD = 151.15; mismatch: M = 255.27, SD = 117.54, t(335) = −1.59, p = 0.11, 95% CI [−32.50, 3.39]). Conversely, the first-pass dwell time on the Hand was notably shorter in the match condition than in the mismatch condition (match: M = 293.54, SD = 157.54; mismatch: M = 395.97, SD = 198.39, t(335) = −7.51, p < 0.001, d = 0.50, 95% CI [−129.24, −75.60]).

When comparing total dwell time on the Hand between the match and mismatch conditions, it was observed that participants fixated on the Hand significantly longer in the mismatch condition than in the match condition (Match: M = 400.51, SD = 192.43; Mismatch: M = 633.76, SD = 310.41, t(335) = −11.52, p < 0.001, d = 0.88, 95% CI [−273.04, −193.44]). Additionally, a square-root transformation of the second-pass dwell time data showed that dwell time was significantly shorter in the match condition than in the mismatch condition (match: M = 7.00, SD = 7.62; Mismatch: M = 14.24, SD = 5.92, t(335) = −13.28, p < 0.001, d = 1.15, 95% CI [−8.35, −6.19]). Finally, there was a significant difference between the match and mismatch conditions for the regression-in-count measure. The participants revisited the Hand more frequently in the mismatch condition than in the match condition (match: M = 0.50, SD = 0.50; Mismatch: M = 1.29, SD = 0.71, t(335) = −16.71, p < 0.001, d = 1.49, 95% CI [−0.89, −0.70]). Ultimately, this indicates a significant impeding effect in the mismatch condition, primarily evident in late temporal eye movement measures.

Previous studies on L1 embodiment using the SPVT paradigm have shown that RT facilitation and impediment in the match and mismatch conditions, respectively, indicate that motor–perceptual representations were mentally formed for sentence comprehension (Zwaan and Pecher, 2012; De Koning et al., 2017). Similarly, the present study on L2 using the SPVT found both facilitating and impeding effects. Notably, these effects were also found in global eye movement measures, such as total dwell time on the whole-picture AOI. This indicates that L2 readers form embodied aspects of mental representations for L2 action sentence comprehension before the picture stimuli are presented, as evidenced by their eye gaze when inspecting the picture stimuli.

RQ 1 investigated whether and what motor experiential traces are involved in mental representations formed for L2 action sentence comprehension. To answer this, we examined which parts of the picture stimulus participants focused on and inspected. HP 3 proposed that the match–mismatch conditions of the sentences and the reference status in the sentence of the body effectors would jointly influence participants' eye gaze on the picture stimuli. The results supported HP 3, as the interaction model best fits the data on fixation counts. The findings related to fixation counts on each body effector, along with HP 1 and HP 2 for RQ1, are discussed in the following sections.

RQ 2 explored to what extent these L2 mental representations are grounded in motor experiential traces. Specifically, HP 4 examined the speed at which participants determine whether the action depicted in a picture matches or mismatches the action implied in a sentence. The results revealed significant differences in fixation on the Hand between the match and mismatch conditions. These differences were found in both early temporal indicators (first-pass dwell time) and late temporal indicators (second-pass dwell time, regression-in-count, and total dwell time). Recall that first-pass dwell time indexes initial processing, with a longer duration indicating the detection of an improbable object (Henderson et al., 1999). Therefore, the significant differences found on the first-pass dwell time indicated that during the first-pass dwell time, participants in the mismatch condition began to notice that the action depicted in the picture mismatched the action implied in the sentence. This raises the question of why these advanced L2 readers detected the mismatch at the second earliest temporal measure (first-pass dwell time) rather than the earliest (first fixation duration).

There are two reasons for this. First, as noted by De Graef et al. (1990), participants initially identified the object (Hand) in the picture, regardless of whether the condition matched or mismatched with the implied action. In other words, without first identifying the depicted limb or hand, participants would not have been able to determine whether the depicted action matched or mismatched the implied action.

Alternatively, the earliest time event rarely yields a significant effect in L2 studies compared to L1 due to processing speed differences between L1 and L2 (Godfroid, 2019). L2 processing speed is generally slower and more arduous than L1. It is well-documented that L2 higher-order and deep-level processing is inefficient and effortful (Francis and Gutiérrez, 2012; Horiba, 1996; Pérez et al., 2019). According to Zwaan and Brown (1996), mental representations in L2 are typically weaker and less detailed than those in L1, often due to fewer inferences being made. This supports the idea that participants in this study developed mental representations with a limited amount of motor experiential traces. In other words, their motor representations were less detailed and less clear regarding the actions implied in the sentences. The clarity of motor representations was sufficient for participants to recognize the action depicted in the picture in the match condition, but it was inadequate in the mismatch condition. This explains why all the late temporal eye-movement indicators, such as total dwell time and regression-in-count, on the Hand in this study, showed significant differences with large effect sizes between the match and mismatch conditions. Participants doubted whether the mismatched action in the picture truly differed from the action implied in the sentence due to the weak and unclear motor representations that they had formed. This uncertainty made them reinspect the conflicting part of the picture more frequently, increasing the regression-in-count and total dwell time on the Hand in the mismatch condition. Therefore, similar to the findings of Vukovic and Shtyrov (2014) and Foroni (2015), which suggest that L2 processing is less embodied in the perceptual aspect, this study indicates that L2 processing is also less embodied in the motor aspect.

Overall, this study shows that advanced L2 readers mentally form sufficient motor embodiment for L2 action sentence comprehension to identify mismatched actions, though not as quickly. Exploring how L2 proficiency differences impact temporal eye movement measures and the strength of motor representations is beyond the scope of this study.

This study shows that advanced L2 learners mentally form motor representations to comprehend action sentences. These representations include adequate motor experiential traces related to the action implied in the sentence, such as specific hand actions. These findings contradict the conclusions drawn by Pavlenko (2014) and Tian et al. (2020), who argued that L2 comprehension is disembodied. Although both Tian et al. (2020) and this study involved Chinese-speaking learners of English, Tian et al. employed lexical and phrasal stimuli. According to Borghi (2012) and Glenberg and Kaschak (2002), sentence context provides situational clues that guide the implementation of actions and goals. In other words, without sentence contexts, individuals struggle to mesh affordances. The conflicting results between Tian et al. and this study indicate that situational context is crucial for activating motor experiential traces, which promote motor representations formed for L2 action language comprehension. Thus, this study corroborates the findings of Santana and De Vega (2013) and Schuil et al. (2013), highlighting the importance of sentence context in activating motor experiential traces.

Contrary to the amodal symbols (e.g., translation practices) and grammatical structures commonly emphasized in L2 classrooms, recent L2 embodied pedagogy studies focus on improving vocabulary learning through activities such as mime, gestures, and multimodal inputs (Ulbricht, 2020; García-Gámez et al., 2021). While these activities aim to promote bodily experiences of learning, they often overlook the importance of situated context in embodied language comprehension. According to the IH, action should occur within a situated context. Therefore, effective embodied language pedagogy should incorporate activities that engage learners in scenarios requiring specific actions, thereby promoting motor–perceptual experiences relevant and meaningful within the learning context. Since adult L2 learners have already developed a range of situated embodied experiences through their L1, L2 embodied reading pedagogy should aim to connect and enhance these motor–perceptual, experiential traces to L2. For instance, group discussions about visually representing text content can help adult EFL learners with low proficiency tap into their existing motor–perceptual experiences (Shiang, 2018).