Conceptualisation of event roles in L1 and L2 by Japanese learners of English: a cross-linguistic comparison of perspectives of event construal

1 Introduction

Human beings live in a dynamic world filled with continuous streams of activities. To make sense of these activities, we need to segment the scenes in front of us into meaningful units with a beginning and an end at a given spatial location, which is called events (Kurby and Zacks 2008; Zacks and Tversky 2001). Although the ability to understand events is arguably one of the universal aspects of human cognition, a growing body of studies has demonstrated that speakers of different languages linguistically encode events differently and further conceptualise events differently (e.g., Flecken et al. 2014; Papafragou and Grigoroglou 2019; Pavlenko 2011, 2014; Slobin 1997, 2000, 2004).

In the cross-linguistic studies on event construal, perspective has been one of the research foci. Cognitive linguists argued that some languages, such as Japanese, tend to construe events from a more egocentric perspective than other languages (Ikegami 2005, 2008; Uehara 2006), known as subjective construal (Langacker 1990). For example, previous studies argued that English speakers tend to narrate stories from a neutral perspective, so their narrative viewpoints switch among different characters accordingly. Contrarily, Japanese speakers tend to project themselves onto the main character(s) in the stories and have a relatively fixed viewpoint throughout the whole narratives (Nakahama 2009, 2011; Nakahama and Kurihara 2006).

However, the cross-linguistic comparisons of event construal between Japanese and other languages have largely stopped at the theoretical level. The findings were mainly based on qualitative descriptive studies that used limited linguistic samples. Little quantitative work has explicitly tested whether the previously documented cross-linguistic differences are reflected in actual language use. Furthermore, if any empirical work shows that the documented differences in the perspective of event construal between Japanese and other languages are manifest, it would be meaningful to study how Japanese speakers learn to reconceptualise their perspective in a second language which construes events from a different perspective. To address the questions above, we focus on how thematic roles of events are mapped onto linguistic units in Japanese and English by comparing both L1 and L2 data. In the section below, we first briefly review the literature on typological contrasts in the perspective of event construal and in thematic roles of events.

2 Experiment

We conducted an image description experiment to test our hypotheses.

2.2 Materials

We prepared four different categories of images showing agent-patient action chains: “an animal chasing a human”, “a human chasing an animal”, “a human throwing an object”, and “a disaster threatening a human” (hereafter “animal → human”, “human → animal”, “human → object”, and “disaster → human”). The images were drawn by a professional illustrator. We adjusted the distance between agents and patients to be the same in each combination. Moreover, we also resized each individual entity in the images (four types: animals, humans, disasters, objects) to better reflect their actual sizes. We ended up with 448^[2] combinations of agent-patient action chains as the image pool for the present study.

To make sure that the prepared images can be recognised by participants as we intended, we conducted an image recognition experiment. We chose several images from each of the four categories, which, in total, included all the four types of individual entities that appear in the images. In total, 42 images were selected as the materials for the image recognition experiment.^[3] Then, we recruited 21 native speakers of Japanese and 21 native speakers of English to write down their descriptions of the images via Google Forms in their native languages. Based on the average accuracy rates, we selected five most successfully recognised individual entities from each type. The recognition accuracy rates for all the selected individual entities were above 90 %. In total, 100 images were singled out from the image pool. Finally, we flipped these images horizontally to create a counterbalanced set of images with the direction of agent-patient action chains reversed. We eventually finalised 200 images as the materials (see Figure 1 for sample images).

Figure 1:

The four categories of images used in this study. The direction in the parentheses indicates the direction of actions.

3 Results and discussion

Unfortunately, ten Japanese speakers participated in the experiment in the same language twice (either in L1 Japanese or in L2 English), so the whole data of their descriptions were removed. The remaining data of eighty-eight participants are publicly available and can be found at https://osf.io/2jvmk/.

We first numerically coded the responses of the remaining participants with respect to whether agents were assigned the grammatical function of subjects^[4] in the sentences and in the sentence fragments.

For the case of sentences, if the response was a simple sentence, we looked at whether the agent in the image was described as the subject of the sentence. If so, this is an agent-focused description, which we coded as “1” (e.g., The horse is running after the man; The tsunami is approaching the girl). Reversely, if the patient in the image was described as the subject of the sentence, this is a patient-focused description, which we coded as “0” (e.g., The man runs away from the tiger; The umbrella is thrown away by the boy). Some responses were hard to judge, which we coded as “−1”. For example, the thematic role of agent and patient was not clearly divided (e.g., A tiger and a lady are running together; He runs with a deer); the response was a description of a scene setting (e.g., The tsunami is so strong).

If the response was a matrix clause (i.e., a clause that contains a subordinate clause), the coding was based on the subject of the main clause. For example, The forest is on fire while the boy is running away was coded as “1”; The boy is escaping while the volcano erupts was coded as “0”; The boy is running away while the kangaroo jogs was coded as “0”.

If the response was a coordination sentence in which two clauses that respectively express the action of the agent and of the patient are linked together by a conjunction such as and, we coded this kind of responses as “−1” because the two conjuncts are usually symmetrical (Goodall 2017), making it hard to judge which conjunct is the main clause. For example, A kangaroo is running, and a girl is escaping. However, if the two clauses both expressed the action of the agent, we coded this kind of responses as “1”. For example, A man gets angry and is throwing a chair. If the two clauses all expressed the action of the patient, we coded this kind of responses as “0”. For example, A man is being chased by a tiger and trying to run away to the left.

Participants also used the sentence fragments to describe the images. There were two types of sentence fragments: fragments with overt subjects (e.g., A kangaroo chasing after a girl) and with covert subjects^[5] (e.g., Running away from the tsunami). The coding for the fragments with overt subjects was based on the subjects of the fragments, following the coding procedures for the sentences described above. For the fragments with covert subjects, the coding was based on the implicitly inferred subjects. If the omitted subject was an agent, it was coded as “1” (e.g., Galloping after a boy). If it was a patient, it was coded as “0” (e.g., Escaping from the tornado).

Based on these coding criteria, we randomly picked up 100 sentences from L2 English data and another 100 from L1 Japanese data for checking interrater reliability between two raters (the authors). The interrater reliability for both sets of data reached 99 % between the two raters. See Appendix A for more detailed information on the linguistic formats (syntactic structures and voice choices) of the participants’ responses.

Using R version 4.2.2 (R Core Team 2022), we fitted generalised linear mixed-effects models (GLMM) with a logit link function to the data. For the GLMM, the lme4 package was used (Baayen et al. 2008; Bates et al. 2015; Jaeger 2008). The lmerTest package (Kuznetsova et al. 2017) was used to calculate p-values. All the plots were generated using plotLMER.fnc in the languageR package (Baayen and Shafaei-Bajestan 2019). Before we fitted statistical models, we removed the codes “−1” because we were interested in the dichotomy of an agent-focused perspective and a patient-focused perspective in the construal of event roles. The removed data accounted for 4.16 % of the whole data.

We started with an overall analysis of data of L1 Japanese, L1 English and L2 English. To avoid the complexity of data analysis and achieve interpretive ease for results, we split the data into two subsets for analyses by category. The first subset is the categories of human → object and human → animal, and the second subset is the categories of animal → human and disaster → human. We labelled the first subset as HumanAgent and the second subset as HumanPatient. To set the proper random effects structure, we tested random slopes and random contrasts for all the predictors considered in the present study: category, language, and direction. We included language, category, the interaction between the two as well as direction as the fixed effects. The categorical predictors were dummy-coded; the reference level of language was L1 English, and the reference levels of category were human → object for HumanAgent categories and animal → human for HumanPatient categories.

For HumanAgent categories, we tested the interaction of language and category, which was not significant. After removing the non-significant parameters, the final model was glmer(code ∼ direction + language + category + (1 + language + category | participant), data = dat_HumanAgent). Table 2 summarises the list of the fixed effects for HumanAgent categories. The final model for HumanPatient categories was glmer(code ∼ direction + language * category + (1 + language + category | participant), data = dat_HumanPatient). Table 3 summarises the list of the fixed effects for HumanPatient categories. Figure 2 shows the probability of selecting an agent as a subject in the three language groups across the four categories of images.

Figure 2:

The probability of selecting agents as subjects in the three language groups across the four categories of images. The probability above “0.5” (the grey area) indicates that agents were more likely to be selected as subjects. The panel for HumanAgent categories was based on the model with the non-significant language:category interaction.

We first observed the effect of category: Participants were more likely to select agents as subjects in HumanAgent categories (human → animal and human → object) than in HumanPatient categories (animal → human and disaster → human), especially when compared to the category of disaster → human. This suggests that when participants described the event roles, they changed their perspectives based on the agency and animacy of the entities in the events.

Of particular interest to us was the interaction between language and category. We used the emmeans package (Lenth 2023) for pairwise comparisons among the three language groups across the four categories. We observed the effect of language in the HumanPatient categories but did not in the HumanAgent categories, which is in line with our hypothesis. In the category of human → object, all three groups tended to select human entities as subjects. No significant differences were found between L1 English and L1 Japanese (p = 0.836) and between L1 English and L2 English (p = 0.933). Likewise, in the category of human → animal, all three groups tended to select human entities as subjects. No cross-linguistic differences were found between L1 English and L1 Japanese (p = 0.325), and between L1 English and L2 English (p = 0.805).

The most outstanding cross-linguistic difference was seen in animal → human as predicted in our hypothesis. While animal entities were selected as subjects by more than half of English speakers, human entities were selected as subjects by more than half of Japanese speakers in both L1 and L2. The pairwise comparisons showed significant differences between L1 English and L1 Japanese (p < 0.001), between L1 English and L2 English (p < 0.001), but not between L2 English and L1 Japanese (p = 0.926). This indicates that the majority of English speakers took an agent-focused perspective when describing the images showing animals chasing humans, whereas the majority of Japanese participants took a patient-focused perspective in both L1 and L2. Moreover, it also indicates that reconceptualisation of event roles was hard to achieve in this category where the contrast of the conceptual representation of events roles between the two languages is large.

Against our hypothesis, no cross-linguistic difference was found in the category of disaster → human. All three groups showed a human-oriented perspective when describing the event roles, as evidenced by non-significant p-values between the groups (L1 English – L1 Japanese: p = 0.595; L1 English – L2 English: p = 0.943; L2 English – L1 Japanese: p = 0.668).

Additionally, we observed the effect of direction, which indicates a language-general perceptual bias in both English and Japanese participants. That is, when the action initiated by an agent flows from left to right, the perceiver is more likely to describe the event from an agent-focused perspective. This result demonstrates that besides conceptual factors, visuospatial properties of scenes can also influence the syntactic choice in sentence production (Esaulova et al. 2019, 2020).

We further conducted an analysis on L2 data (8,314 data points) with the purpose of examining how L2 proficiency and daily English use influenced the linguistic encodings of event roles. Since no cross-linguistic differences were found in the HumanAgent categories, we fitted the models only to the HumanPatient categories, with the special interest in the category of animal → human. To set the proper random-effects structure, we tested random slopes and random contrasts for the factors category and direction. In the analysis of L2 proficiency, we defined category, averageEngScore_s, the interaction between the two, and direction as the fixed effects. In the analysis of daily English use, we included category, dailyEngUse_s, the interaction between the two, and direction as the fixed effects.

The interaction of category and averageEngScore_s was not significant. Specifically speaking, with the L2 proficiency increasing, the performances of Japanese learners barely changed in the category of animal → human. Even the learners with relatively high English proficiency tended to have a human-oriented perspective when encoding the event roles in this category. This is also the case for dailyEngUse_s (daily English use). One reason for this might be the small variance in the participants’ self-reported amount of daily English use (range = 0: 5.33, SD = 1.32, see Table 1).

4 General discussion

We now review our research questions and hypotheses based on the results above. In the first research question, we asked whether speakers of Japanese and English encode event roles in different ways. We first hypothesised that in general, Japanese speakers have a human-oriented perspective in encoding event roles. The result was consistent with our hypothesis. Regardless of the thematic roles that human entities are assigned with, L1 Japanese speakers demonstrated a greater tendency to describe human entities as subjects of the sentences compared to English counterparts, which suggests that when Japanese speakers construe an event, they tend to empathise with the most animate entity in the event and then describe the event from that entity’s perspective. The merge of the perspectives of event observers and event participants is an embodiment of the internal perspective in event construal, a subjective construal.

We also hypothetically put forward a hierarchy of degrees of cross-linguistic differences between Japanese and English as follows:

(large) animal agent – human patient

(medium) inanimate agent – human patient

(small) human agent – animal patient, human agent – inanimate patient

Our result is partially in line with the hypothesis. Based on the results, this hierarchy should be revised as follows:

(large) animal agent – human patient

(small) inanimate agent – human patient

   human agent – inanimate patient

   human agent – animal patient

The degree of the conceptual contrast in event roles between the two languages was the largest in the category of animal → human. In this category where the most animate entities are patients and the less animate entities are agents, animacy and agency compete for the selection of the subject, so the cross-linguistic differences in the preference for animacy and agency emerged in the most evident way. The Japanese language, which shows an animacy advantage over agency because of subjective construal, guides its speakers to prioritise human entities in the selection of subjects, whereas the English language, which emphasises agency because of objective construal, guides its speakers to prioritise animal agents. The result here directly questions the universality of the Thematic Hierarchy (Grimshaw 1990; Jackendoff 1990), stating that the agent is the most prominent role and often comes at the subject position.

Nonetheless, no cross-linguistic difference was observed in the category of disaster → human, which is inconsistent with our hypothesis. This result might indicate that the agency advantage over animacy in selecting event roles as subjects of sentences in English only appears when agents are animate entities, such as animals. The result seems to be inconsistent with the finding in the previous studies that inanimate subjects are more acceptable for English speakers than for Japanese speakers (Lebkuecher and Malt 2024; Otaki and Shirahata 2017). One possible reason for the different findings between the present study and the past ones is that the higher tolerance of inanimate subjects for English speakers does not necessarily imply that English speakers would more actively select inanimate entities as subjects in actual language use, which again suggests that the construal difference between Japanese and English is a continuum rather than two extremes, as argued by Langacker (1990). However, there is another theoretical possibility that can account for this result. Other types of inanimate entities might show different results, such as projectiles (e.g., arrow → man) and instruments (e.g., ball → man). Conceptualisers need to assume that there are other latent agents behind the projectiles and instruments, which all require energy from external sources, most often from human beings. In the present study, we used natural disasters for the inanimate category. For this type of entities, conceptualisers do not have to assume that there are other latent agents behind because natural disasters can generate energy themselves. By assuming human agents standing behind, English speakers might be motivated to choose projectiles and instruments as subjects of sentences because these types of entities might be perceived to be the extension of the latent agents. We leave this issue to future studies.

The results for human → animal and human → object are consistent with our hypotheses. The cross-linguistic differences were negligibly small in the two categories where the humans are the agents and the most animate entities. Therefore, both Japanese and English speakers showed a strong tendency to describe humans as subjects of sentences.

Taken together, these quantitative results elucidate that the Japanese language prefers subjective construal whereas the English language prefers objective construal, and the preferences for subjective/objective construal are a matter of degree. Our result adds a piece of quantitative evidence to the subjectivity theory in the cognitive linguistics literature. Japanese encourages its speakers to merge their viewpoints with those of human participants in events, with whom they can most easily empathise. This gives rise to an animacy advantage over agency in Japanese. In contrast to Japanese, English encourages its speakers to construe events from an external vantage point, thus placing more emphasis on agency and showing a greater tendency to encode event roles from the perspective of initiators of actions. However, at the same time, animacy also plays an important role for English speakers. The agency advantage over animacy appears only when agents are animate entities.

Now, we turn to the second research question. We were interested in how L1-based conceptualisation of event roles influences the linguistic encodings of event roles in L2. We hypothesised that both the L1-based conceptual transfer and the restructuring of the L1-based conceptualisation of event roles would take place in L2. The result is not in line with our hypothesis. Influenced by the L1 conceptualisation of event roles, Japanese participants in L2 remained more likely to select human entities as subjects of sentences than English participants in the category of animal → human. This L1-entrenched preference for subjective construal posed a great challenge for Japanese learners of English in the acquisition of event roles, as evidenced by the result that the magnitude of conceptual restructuring in the category of animal → human was not mediated by L2 proficiency nor the amount of daily English use. Even the learners with relatively high English proficiency in our study could not reconceptualise the event roles. There are several reasons that may help explain the observed difficulty in the conceptual restructuring of event roles by Japanese learners of English.

The first reason is that the degree of the cross-linguistic difference in the category of animal → human is very large between English and Japanese. As shown in Figure 2, in animal → human, the vast majority of English speakers tended to select the animal agents as the subjects, but the vast majority of Japanese speakers tended to select human patients as the subjects. The larger the degree of contrast in conceptual representations between L1 and L2, the greater the learning difficulty. The second reason is the nature of the acquisition difficulty of English event roles for Japanese speakers. As already mentioned in Section 1.3, the cross-linguistic differences in the conceptualisation of event roles between English and Japanese are rather probabilistic. In other words, it is not the issue of the presence or absence of certain linguistic terms or structures but the issue of pragmatic preference. The linguistic feature of low noticeability in the target language is challenging for L2 acquisition (Ellis 2006). To notice the difference, Japanese learners would need enough input that directs their attention to how agency is constructed in English. The input may come from explicit pedagogical instructions or from implicit learning, such as daily exposure to English, the experience of living in English-speaking environments, etc. This leads us to the third reason, which is the participants’ linguistic background. In the present study, the recruited Japanese participants had the limited experience of living in English-speaking environments. Only 11 out of 44 participants had the experience of living in an English-speaking country (See the published data on OSF). Moreover, Japanese was the dominant language for daily communications for all participants, and the amount of daily English use was rather limited (see Table 1). Furthermore, we are not aware of any English teaching materials in Japan that explicitly teach English encodings of event roles or teach the construal difference between Japanese and English. Thus, the learning experience of Japanese participants lacked enough input that helps raise their awareness of the construal difference in perspective between Japanese and English. Future studies can examine the impact of these factors by testing Japanese participants with more diverse L2 experiences and proficiency.

Can L2 learning in this challenging area be facilitated? Although the implicit learning, such as being exposed to English-speaking environments and increasing the daily use of English, helps learners attune themselves to the conventional construal of the target language by facilitating the probabilistic processing (Ellis 2002), construal is argued to be an area where learners benefit more from explicit instructions than mere exposure to L2 (Littlemore 2009). This is especially practical for teaching the perspective of event construal. As demonstrated in the Introduction, the cross-linguistic differences in the perspective of event construal between Japanese and English are reflected in various linguistic features. Therefore, English teachers in Japan can take advantage of multiple linguistic resources to make Japanese learners of English aware of the objective construal in L2 English. For example, compared with Japanese, English has a smaller number of intransitive verbs and is more likely to use transitive verbs to describe accidental causal events (Bohnemeyer et al. 2010; Fausey et al. 2010). English teachers can particularly compare (in)transitive verbs in the two languages, which helps Japanese learners to be more aware of agency in event construal. Another example is the benefactive constructions. This is especially useful because the give-type benefactive is highly frequent and among the first to be learned in many languages (Bowerman 1973; Goldberg et al. 2004). The give-type benefactive in Japanese shows a degree of complexity that many languages lack (Newman 1996; Shibatani 2003). There are two verbs for the give-type benefactive in Japanese, which are ageru and kureru. They distinguish the directions of benefits. If the direction of benefits goes from oneself to others, ageru is used. If the direction of benefits goes from others to oneself, kureru is used. The absence of the semantic distinction between “ego” and “others” in English give-type benefactive can guide Japanese learners to view events from a neutral (objective) perspective. Raising Japanese learners’ awareness of the preferred perspective of event construal in English from different linguistic features may drive the learners to acquire a new “fashion of speaking” – objective construal.

There are several limitations in the present study. The first limitation of the present study is that the verbs considered in the experimental materials lacked diversity. The images only demonstrated three types of actions: chase, threaten, and throw. Since different verbs have various degrees of forces and durations, speakers may conceptualise event roles in different ways depending on the properties of verbs. For example, speakers may empathise with patients under strong forces more easily than patients under weak forces. Therefore, future studies should compare the cross-linguistic differences between Japanese and English in conceptualising event roles for different types of verbs. Second, we did not take into consideration construal differences that potentially exist among several syntactic patterns in the participants’ responses. For example, we coded the intransitive sentence A man is running away from a tiger to be the same as the passive sentence A man is being chased after by a tiger and to be the same as the subjectless sentence fragment Running away from a tiger, because all of them arguably took a patient-focused perspective. However, an intransitive sentence is argued to be less agentive than a passive sentence (Ikegami 1991), and a subjectless intransitive sentence might be even less agentive than an intransitive counterpart. The degree of agentivity reflects the degree of subjectivity in the event construal. Future studies should focus on these potential construal differences in a more fine-grained manner. Third, participants might have developed response strategies throughout the experiment, which might have enlarged or shrunken the cross-linguistic differences found in the current study. Unfortunately, because we used Google Forms to collect the data, the trial counts (i.e., the number of preceding trials) were not available. In addition, on Google Forms, participants could go back to the previous trials and correct their responses. Moreover, we used Amazon Mechanical Turk (MTurk) to recruit English participants as done in many past studies (e.g., Enochson and Culbertson 2015; Schmidtke and Kuperman 2024; Schnoebelen and Kuperman 2010). However, it should be noted that MTurk is unique in that all participants are anonymised, and they might not be motivated to take tasks seriously (Peer et al. 2022). Future studies should take these methodological issues into consideration. Finally, future studies should also investigate whether different linguistic encodings of event roles between Japanese and English influence the perception and memorisation of event roles. There have been some recent studies on the interface between linguistic expressions of event roles and cognition (Isasi-Isasmendi et al. 2023; Ünal et al. 2021). They all claimed that despite the cross-linguistic differences in linguistic encodings of event roles, the underlying cognitive mechanism for processing event roles is similar across languages, which is largely consistent with the Thematic Hierarchy. Nonetheless, we doubt this is the case for Japanese because event roles reflect the subjective construal in Japanese, which may be more pervasive in the perception of event roles than previously studied languages.

5 Conclusions

Using event roles as a window into the understanding of the interface of language and event cognition, this study quantitatively compared how Japanese and English speakers linguistically encode event roles differently. Our results demonstrated that Japanese prioritises animacy over agency, especially human entities, in selecting subjects of the sentences because Japanese speakers can easily empathise with the most animate entities in events and then describe the events from the perspectives of the most animate entities. In contrast to Japanese, agency plays a more important role for English, because English speakers are more likely to construe events from the perspective of initiators of actions. At the same time, animacy also plays a crucial role in English, demonstrated by the finding that the agency advantage over animacy arises only when agents are animate entities. Our results are in line with the cognitive linguistics theory that the Japanese language prefers subjective construal while the English language prefers objective construal, and the preference for subjective/objective construal is a matter of degree. At the same time, our results question the universality of the agent-first preference in the Thematic Hierarchy, which calls for paying more attention to subjectivity-prominent languages, such as Japanese.

Moreover, the current study explored how L1 conceptualisation of event roles influences L2 learning. We found that Japanese learners of English had difficulty in reconceptualising event roles in L2.