Metaphorical language is not just a poetic expression but also a conceptual device to communicate abstract ideas. The majority of studies in the past decades have focused on nominal metaphors in the typical “A is B” structure (e.g., The rumor was a virus), but few have been conducted on predicate metaphors (e.g., The media bent the truth), despite their frequent occurrence in daily discourses (Cameron, 2008; Steen et al., 2010; Goatly, 2011). Compared with literal sentences that convey physical senses (e.g., The repairman bent the pipe), predicate metaphors use verbs figuratively via abstraction from concrete action terms. Recently, neural studies found sensorimotor activations across brain regions when English natives read predicate metaphors (Chen et al., 2008; Desai et al., 2013; Obert et al., 2014; Lai et al., 2019); however, it has not been well-understood how action-based metaphors go beyond embodiment to create abstract thoughts in the brains of L2 learners whose language proficiency levels could affect difficulties of comprehension and embodiment. Furthermore, predicate metaphors comprehension may be a process of indirect categorization with an intermediate entity (Utsumi and Sakamoto, 2011; Obert et al., 2018), but its processing mechanisms in native language and second language were not fully studied. The aims of the present study are thus to investigate whether and how the embodiment of verbs, via the activation of sensorimotor properties, influences two groups of L2 learners during their comprehension of conventional and novel predicate metaphors and to explore the issues of intermediate entity in their processing of predicate metaphors.

A predicate metaphor is a linguistic construction that exemplifies the embodied nature of cognition since verbs generally entail more action content. Embodied cognition (Lakoff and Johnson, 1980; Glenberg and Kaschak, 2002; Gibbs, 2006a; Zwaan and Taylor, 2006; Estes et al., 2008; Glenberg, 2010) holds that cognition is grounded in the physical body and sensorimotor system. Abstract concepts can be understood via metaphors, which link abstract, less familiar knowledge with concrete and familiar experience. For example, reading tactile metaphors (e.g., a rough day) and taste metaphors (e.g., a sweet compliment) can activate sensory regions responsive to touch and taste (Lacey et al., 2012; Citron and Goldberg, 2014). Reading action metaphors related to motion content (e.g., bent the truth) can activate regions involved in motor perception (Desai et al., 2011, 2013; Lauro et al., 2013). The stronger view of embodied accounts (Gibbs et al., 2004; Gibbs, 2006b) believes that metaphors are comprehended via sensorimotor simulation, just as literal language is comprehended; while the weaker view (Boroditsky and Ramscar, 2002; Cardillo et al., 2012; Jamrozik et al., 2016) argues that only novel metaphors draw on sensorimotor information from the source domain, but the representation of source domain becomes abstracted as metaphoric use increases. A predicate metaphor is a typical vehicle to address these mixed views as it allows a verb to be used literally to describe physical action and metaphorically to convey abstract ideas.

Recent neural studies have supported the embodied views of predicate metaphor processing. For example, a few secondary motor regions were found to be involved when participants read familiar predicate metaphors, but the primary sensory and motor regions were more activated when they read unfamiliar novel metaphors (Desai et al., 2011, 2013; Cardillo et al., 2012), suggesting that novel predicate metaphors relied more on sensorimotor information corresponding to the verbs. However, other studies showed inconsistent motor activations in the brain (Watson et al., 2013). It seems that novelty or conventionalization of metaphors plays a role in the mixed results on embodiment. According to the career of metaphor theory (Bowdle and Gentner, 2005), all metaphors start as novel expressions, then evolve into conventional expressions, and eventually become literal expressions that lose the figurative meaning. For predicate metaphors, the figurative meaning centers on the concrete motion aspect of verbs. When predicate metaphors (e.g., kick the habit) are conventionalized over time, the salient meanings with deep relational alignment will remain in the brain (e.g., give up something), but the surface sensorimotor properties of the base meaning will be shed (e.g., move the foot forward). In this sense, there may be a graded embodiment for metaphor processing relative to metaphor novelty (Desai et al., 2013; Jamrozik et al., 2016).

In the past decade, many behavioral studies have addressed the embodied qualities of verbs (Richardson et al., 2001, 2003; Wilson and Gibbs, 2007), yet few studies have studied predicate metaphors in a sentence context or discourses, and it remains unclear what type of sensorimotor activation (e.g., visual or auditory) could influence predicate metaphor comprehension. For the structure of word meaning, a major division lies between the conceptual structure and spatial structure, the latter of which involves perception and action (Jackendoff, 2002); therefore, when a picture denoting an action is presented, the sensorimotor elements in the spatial structure are likely to be activated before a particular verb meaning is accessed in the conceptual structure. The imagery of such an action in the spatial structure can be readily available without retrieving a verb because the processing systems of verbal stimuli (text) and non-verbal stimuli (image) are functionally and structurally independent (dual coding theory, Paivio, 2014). In this sense, if one relies on the sensorimotor system when reading predicate metaphors, the imagery (i.e., embodiment) may facilitate abstraction from concrete meaning and result in faster comprehension.

Relatively few accounts have been proposed to explain how predicate metaphors are comprehended. Based on the studies on nominal metaphors, the categorization theory (Glucksberg, 2001, 2003, 2008; Torreano et al., 2005) was proposed to account for predicate metaphor processing. In the case of “My job is a jail,” the words “job” and “jail” are both exemplars of a superordinate category “confining and unpleasant situation.” Similarly, in the predicate metaphor “The rumor flew through the office,” the verb “fly” is a prototypical exemplar of the superordinate category “fast travel.” This process is a direct categorization (Glucksberg, 2001, 2003; Torreano et al., 2005) in which metaphorical meanings are achieved through direct abstraction from a superordinate category. However, Utsumi and Sakamoto (2011) doubted this explanation because the semantic structures of verbs are qualitatively different from those of nouns. Specifically, verbs involve events and action, and semantic relations entail rich elaborations and thematic roles, such as the hyponymy “To V1 is to V2 in some particular manner” (e.g., to bend is to force a different shape or direction). Nouns, however, refer to objects, and the semantic relations between nouns are mostly class inclusion, such as the superordination “A N1 is a N2” (e.g., A canary is a bird). Verbs are less likely to rely on hierarchical relations (Chen et al., 2008); therefore, the meanings of predicate metaphors may not be created via the superordinate categorization of the events that are evoked by verbs.

An intermediate entity is assumed to be activated by verbs before a figurative category of actions is finalized (Utsumi and Sakamoto, 2011). In the example The rumor flew through the office, the intermediate entity such as things that fly (e.g., plane or bird) or an imaginary scene of flying, may be evoked before an abstract meaning “travel fast” was created. In this case, understanding a predicate metaphor is a two-stage process with indirect categorization. An N400-LPC component observed in an ERP study (Obert et al., 2018) showed that semantic integration occurred during metaphor processing, and this result appeared consistent with the indirect view of categorization. However, these studies failed to specify the intermediate entity. Will the intermediate entity be the specific sensorimotor association from verbs or the prototypical agent of verb? When reading novel expressions such as The taxes pinch the industry, does one think of the act of pressing something between a thumb and a finger, a painful feeling, or a pair of shoes (a prototypical entity that pinches)? When reading familiar expressions such as The media bent the truth, does one directly retrieve the abstract meaning “distort” from long-term memory or rely on the concrete act of “causing to curve”? In this study, we assumed that if predicate metaphors are processed via intermediate entities, the sensorimotor aspects of verbs, then the pictures describing verb-related action can facilitate the understanding of metaphors.

It is well-known fact that understanding metaphors in a second or foreign language may impose great difficulty due to linguistic barrier. Thus, untangling how L2 learners process L2 metaphors could have important insights for those trying to teach a second language (Littlemore, 2009, 2019). There has been a great body of literature exploring relationships between metaphors processing and second language learning in terms of conceptual metaphor theory (Yasuda, 2010; Lu and Sun, 2017), metaphorical competence (Danesi, 1993; Littlemore and Low, 2006), metaphor awareness (Guo, 2007; Chen and Lai, 2012; Boers, 2013), imagery (Ifantidou and Hatzidaki, 2019) and individual differences (Johnson and Rosano, 1993; Wegner et al., 2020). Previous ERPs studies on bilingual metaphor processing found neural similarities and difference in metaphor processing between L1 and L2 (Dong, 2013; Park and Chung, 2013; Xue et al., 2014; Liu, 2016; Jankowiak et al., 2017, 2021; Wang, 2018). Although metaphoric meaning integration might be of similarity in L1 and L2, L2 learners require more intensive cognitive mechanism or continued effort in lexicon-semantic access and tended to adopt L1 neural pathway to process L2 metaphors. However, Citron et al. (2020) found L2 speakers were less affected by increasing metaphoricity than L1 speakers, suggesting metaphorical language is not necessarily engaging in L2 speakers. Yet this research did not take L2 proficiency into consideration since several neural studies found that L2 proficiency difference had an impact on neural activities in the brain areas responsible for novel metaphors (Dong, 2013; Liu, 2016).

Most literature of L2 metaphor comprehension focused on nominal metaphors and temporal-spatial metaphors, but lacked discussion upon predicate metaphors. Most ERPs studies described the processing mechanism at the neural level, but behavioral evidences that probe into the embodied features of L2 metaphor processing are insufficient. As predicate metaphors are cognitively challenging, how sensorimotor stimulation facilitates L2 predicate metaphor processing becomes an important concern. According to dual coding theory, verbal information (e.g., language) and non-verbal information (e.g., image) both contribute to cognition. From a bilingual point of view (Schnotz and Horz, 2010), lower-level L2 learners can perform better with two information sources (i.e., picture and text) while higher-level learners can perform well with only one source (i.e., text). In this sense, the embodiment effect of action-related pictures on predicate metaphor comprehension will be different among L2 learners of different levels.

In summary, it remains unknown whether and how predicate metaphor processing draws from the sensorimotor system, and the mechanism of intermediate entity during predicate metaphor processing has been unclear. The mixed findings in previous literature could be a result of difference in metaphor novelty, language proficiency and experimental paradigms. Given that English predicate metaphors are unfamiliar to Chinese L2 learners whose mother tongue has less of this construction, a second language perspective could help to reveal more about the embodied features of predicate metaphor processing. Taken together, we form the following questions: (1) Does the embodiment of verbs (i.e., the activation of sensorimotor aspects) influence L2 predicate metaphor processing? (2) How is the embodiment effect different between higher-level and lower-level Chinese L2 learners when they read English conventional predicate metaphors and novel metaphors? (3) Are L2 predicate metaphors comprehended via intermediate entity in an indirect process?

After excluding participants with an error rate >15%, the behavioral data of 92 participants were valid. The average comprehension accuracy for higher-level group and lower-level group is 91 and 89%, respectively, indicating that the participants were engaged and attentive in the experiment. The analysis of reaction time was performed on the correct trials.

Table 4 shows the mean reaction time and mean accuracy. Concerning the data of metaphors, we found that conventional metaphors and novel metaphors were processed faster in the related priming condition than the unrelated priming condition (CM: M_Related = 3073.8 ms < M_Unrelated = 3289.3 ms, p = 0.000; NM: M_Related = 3545.4 ms < M_Unrelated = 3955.4 ms, p = 0.000). Specifically, in the higher-level group, novel metaphors had a significant priming effect (M_Related= 3328.9 ms < M_Unrelated= 3661.1 ms, p = 0.002) on reaction time while conventional metaphors and literals did not; in the lower-level group, the strongest priming effect was observed in the novel metaphors (M_Related= 3712.0 ms < M_Unrelated= 4181.9 ms, p = 0.000,), followed by the conventional metaphors (M_Related= 3236.4 ms < M_Unrelated= 3480.1 ms, p = 0.004). Therefore, the results of priming size seems to present a gradual facilitation on reaction time (Diff _Lower−NM = 469.9 ms, Diff _Higher−NM = 332.2 ms, Diff _Lower-_CM= 243.7 ms) across metaphor novelty, sentence type and proficiency groups.

A four-way mixed ANOVA of prime (2) × sentence type (2) × novelty (2) × proficiency (2) is conducted on reaction times and accuracy by participants (F₁) and items (F₂). As shown in Table 5, significant effects on RTs were observed for all within-subjects variables: prime [F₁ (1,90) = 46.789, p = 0.000, ηp2 = 0.342; F₂ (1,46) = 65.436, p = 0.000, ηp2 = 0.587], sentence type [F₁ (1,90) = 213.260, p = 0.000, ηp2 = 0.703; F₂ (1,46) = 66.387, p = 0.001, ηp2 = 0.591], and novelty [F₁ (1,90) = 126.12, p = 0.000, ηp2 = 0.584; F₂ (1,46) = 12.529, p = 0.001, ηp2 = 0.214]. The between-subjects effect of proficiency was also significant [F₁ (1,90) = 4.358, p = 0.040, ηp2 = 0.046; F₂ (1,46) = 5.676, p = 0.021, ηp2 = 0.110]. Pairwise comparison showed that participants spent less time processing predicate metaphors in the action-related pictures condition than in the unrelated condition (M_Related = 3022.7 ms < M_Unrelated = 3240.4 ms, p = 0.000); literal sentences were processed much faster than metaphorical sentences (M_L = 2825.1 ms < M_M = 3438.0 ms, p = 0.000); all target sentences in the conventional conditions were processed faster than in the novel condition (M_CM = 2971.4 ms < M_NM = 3291.8 ms, p = 0.000). Higher-level group spent less time in reading target sentences than lower-level group did (M_High = 2941.3 ms < M_Low = 3321.9 ms, p = 0.040).

The two-factor interactions were significant between sentence type and novelty (F₁ = 56.732, p = 0.000, ηp2 = 0.387; F₂ = 17.843, p = 0.000, ηp2 = 0.279), between prime and sentence type (F₁ = 6.032, p = 0.016, ηp2 = 0.063, F₂ < 1), and between prime and novelty (F₁ = 5.208, p = 0.025, ηp2 = 0.055, F₂ < 1). However, no interaction was observed with proficiency. Specifically, it took participants more time to read novel metaphors than conventional ones (M_CM = 3181.6 ms < M_NM = 3750.4 ms, t (1,183) = −13.719, p = 0.000), but no significant difference was found in the literal targets (M_L−CM = 2809.5 ms < M_L−NM = 2884.1 ms, t (1,183) = −1.796, p > 0.05). In the related priming conditions, literals were processed much faster than metaphor (M_M = 3309.6 ms < M_L = 2783.1 ms, t (1,183) = 10.496, p = 0.000) and in the unrelated conditions, the differences between sentences types were a bit greater (M_M = 3622.4 ms < M_L = 2910.5 ms, t (1,183) = 13.081, p = 0.000). Similarly, conventional targets were read faster than novel targets when they were primed with the pictures of actions (M_C = 2918.7 ms < M_N = 3072.4 ms, t (1,183) = −3.248, p = 0.001), but when the pictures were not action-related, the differences were slightly greater (M_C = 3174.0 ms < M_N = 3460.5 ms, t (1,183) = −6.983, p = 0.000). Figure 2 showed that the time gap between conventional targets and novel targets was smaller in the related-picture condition than in the unrelated conditions, suggesting the embodied priming effect was influenced by metaphor novelty.

Accuracy data presented a main effect for prime only with item analysis (F₂ = 4.697, p = 0.035, ηp2 = 0.093) but a stronger effect for novelty (F₁ = 37.303, p = 0.000, ηp2 = 0.239; F₂ = 7.835, p = 0.007, ηp2 = 0.146). Proficiency had little effect on accuracy (F₁ = 3.032, p > 0.05; F₂ < 1), suggesting that both groups of L2 learners did not differ greatly in the correct comprehension of predicate metaphors. No interaction effects were observed except one between sentence type and novelty, only with subject analysis (F₁ = 9.283, p = 0.003, ηp2 = 0.094; F₂ = 1.272, p = 0.265, ηp2 = 0.027).

We conducted two separate analyses to further explore embodiment effect on L2 predicate metaphor comprehension as two proficiency groups has demonstrated significant difference.

For higher-level learners, the three-way ANOVA of prime (2) × sentence type (2) × novelty (2) showed significant main effects of prime [F₁ (1,39) = 17.356, p = 0.000, ηp2 = 0.308; F₂ (1,23) = 33.831, p = 0.000, ηp2 = 0.595], sentence type [F₁ (1,39) = 110.995, p = 0.000, ηp2 = 0.740; F₂ (1,23) = 35.055, p = 0.000, ηp2 = 0.604] and novelty [F₁ (1,39) = 46.184, p = 0.000, ηp2 = 0.542; F₂ (1,23) = 8.892, p = 0.007, ηp2 = 0.279]. Yet different to the four-way ANOVA, only one interaction effect was found between sentence type and novelty [F₁ (1,39) = 24.619, p = 0.000, ηp2 = 0.387; F₂ (1,23) = 11.921, p = 0.002, ηp2 = 0.341]. The reaction time of conventional metaphors and novel metaphors differed greatly [M_CM = 2952 ms < M_NM = 3495 ms, t (1,79) = −8.409, p = 0.000], while that of L-CMs and L-NMs did not (M_L−CM = 2621 ms < M_L−NM = 2697 ms, p > 0.05).

Although a three-factor interaction was not observed, the reaction times of novel metaphors were greatly reduced when primed with related pictures, but those of conventional metaphors were not affected [see Table 4, NM: M_Related = 3328.9 ms < M_Unrelated= 3661.1 ms, t (1,79) = −3.272, p = 0.002; CM: M_Related= 3328.9 ms < M_Unrelated = 3661.1 ms, t (1,79) = −2.018, p > 0.05]. This result may be due to participants' higher L2 proficiency and familiarity with conventional metaphors. Higher-level L2 learners draw on the sensorimotor information of verbs when they comprehend unfamiliar predicate metaphors while the concrete aspects of verbs are not essential when they comprehend familiar metaphors. This seems to be consistent with the weaker version of embodied cognition and the graded feature of L2 embodiment. In addition, it suggests that novel predicate metaphors could be processed via the intermediate entity (i.e., concrete, sensorimotor aspects of verbs) in an indirect manner, whereas conventional predicate metaphors could be processed through direct abstraction (i.e. direct categorization), without sensorimotor entities emerging.

Accuracy data revealed a main effect of prime [F₁ (1, 39) = 6.624, p = 0.014, ηp2 = 0.145; F₂ (1,39) = 2.483, p = 0.129], with more correct answers in the related priming condition. The effect of novelty was also present [F₁ (1, 39) = 16.784, p = 0.000, ηp2 = 0.301; F₂ (1, 23) = 2.569, p = 0.123], but these effects were only found with subject analysis. There were more correct answers in the related priming condition (M_Related= 91.6% < M_Unrelated= 89.4%, p = 0.013), and less accuracy in the novel groups (M_CM = 92.1% < M_NM = 88.9 %, p = 0.000). Neither the main effect of sentence type nor any interaction effects were found in comprehension accuracy.

For lower-level group, the three-way ANOVA of reaction time revealed significant main effects of prime [F₁ (1, 51) = 31.632, p = 0.000, ηp2 = 0.383; F₂ (1, 23) = 32.988, p = 0.000, ηp2 = 0.589], novelty [F₁ (1, 51) = 85.891, p = 0.000, ηp2 = 0.627; F₂ (1, 23) = 4.453, p = 0.046, ηp2 = 0.162] and sentence type [F₁ (1, 51) = 117.68, p = 0.000, ηp2 = 0.698; F₂ (1, 23) = 31.558, p = 0.000, ηp2 = 0.578]. Although the results of main effects were similar between two groups, two significant interactions were found between sentence type and novelty [F₁(1, 51) = 33.838, p = 0.000, ηp2 = 0.399; F₂ (1, 23) = 6.093, p = 0.021, ηp2 = 0.209] and prime and sentence type [F₁ (1, 51) = 6.060, p = 0.017, ηp2 = 0.106; F₂ (1, 23) = 1.195, p = 0.286]. Based on the data of metaphors, the interaction between prime and novelty was also significant [F₁ (1, 51) =6.624, p = 0.013, ηp2 = 0.115; F₂ < 1]. The significant priming was found for both conventional and novel metaphors among lower-level learners (see Table 4, CM: M_Related = 3236.4 ms < M_Unrelated = 3480.1 ms, p = 0.004; NM: M_Related = 3712.0 ms < M_Unrelated = 4181.9 ms, p = 0.000), with greater facilitation for novel metaphors [Diff _CM =322.2 ms, Diff _NM= 469.9 ms, t (1,51) = −2.574, p = 0.013]; whereas this effect was only significant in the novel metaphors among higher-level learners.

As for accuracy data, no main effect of prime was found (F₁ < 1, F₂ = 2.221, p > 0.05), which was different to the results of higher-level group. The main effect of novelty was great (F₁ = 21.909, p = 0.000, ηp2 = 0.300; F₂ = 6.451, p = 0.018, ηp2 = 0.219) as targets in the novel conditions are less familiar, but neither the effect of sentence type nor any interaction effects were observed.

Our findings reveal that the embodiment effect via picture priming was positive: L2 learners of both groups could comprehend predicate metaphors much faster when primed with action-related pictures. Metaphor novelty and L2 proficiency played significant roles during the process, with novel metaphors receiving more facilitation than conventional metaphors, and lower-level L2 learners gaining greater embodiment benefits than higher-level L2 leaners. In the following, we will further discuss how the embodiment of verbs influences L2 predicate metaphor processing when novelty and L2 proficiency are taken into consideration, and explore the possibility of intermediate entities as a process of indirect categorization.

Given the difficulties L2 learners experience with metaphoric language comprehension and the paucity of research on predicate metaphors, this study first explored how embodiment influences L2 learners' processing of predicate metaphors. The sensorimotor elements of action-related pictures provided a graded priming effect, which was modulated by L2 proficiency and metaphor novelty. The present findings not only have important theoretical implications for embodied cognition and categorization model for metaphor processing, but also offers practical suggestions for L2 acquisition and teaching.

However, there are limitations in the study. First, the familiarity rating for the verbs in the novel metaphors was relatively lower, although not at a significant level. Second, pictures may carry literal sense of the verb and the present study did not manage to exclude it from picture priming due to design difficulty. Third, the other types of intermediate entity were not addressed here. If the hypothesis about indirect categorization was correct, then it would be expected that the auditory aspects of verbs may serve as the intermediate entity to facilitate auditory predicate metaphor processing. Despite a body of fMRI studies on predicate metaphors, more ERP and eye-tracking studies are needed to investigate the time course, attention and other qualitative features. For example, participants tend to focus on subject-verb phrases and rate higher metaphoricity for subject-noun phrase sentences over object-noun phrase sets (Torreano et al., 2005). This indicated that more attention was given to the first half of the sentence. In summation, future studies should consider verb familiarity, conventionality, different types of intermediate entities and different experimental methods.

The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author/s.

Ethical review and approval was not required for the study on human participants in accordance with the local legislation and institutional requirements. The patients/participants provided their written informed consent to participate in this study.

YF contributes greatly to this study, responsible for conceptualization, methodology, software, experiment, formal analysis, and writing (original draft and editing). RZ was responsible for supervision, conceptualization, and writing (review and editing). Both authors contributed to the article and approved the submitted version.

This study was supported by the Guangdong Provincial Philosophy and Social Science Foundation, China [Grant No. GD20CWY12].

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.