Event conflation in high stakes testing: a comparison of usage and relationship to writing scores by language types

1.1 The theory of event conflation

Talmy (1985) first noted that it is possible to conceptualize, and linguistically encode, two separate events as a single one, and named this phenomenon event conflation (also known as event integration). For example, though the sentence John hurried across the bridge seems to be a single event, there are actually two micro-events contained in the same sentence: John hurrying and John crossing the bridge. He considered one of the two micro-events, the physical or metaphorical path, to be the ‘main-event,’ and the other (the means, cause, or additional event) to be the ‘co-event.’ Talmy (1991) further suggested that a typology based on event conflation is possible. Specifically, he called expressions that encode the main-event in the main verb of a sentence and the co-event outside of it verb-framed. Alternatively, he dubbed expressions that encode the main-event in a ‘satellite,’ a linguistic element in a sister-relation to the main verb, and the co-event in the main verb, satellite-framed. Talmy (1985, 2000] suggested that there were five types of conflated events, as shown in Table 1, below. In each example, the (a) sentence represents a satellite-framed expression and the (b) sentence represents a verb-framed expression.

Despite there theoretically being five conflated event types, the lion’s share of related research has been dedicated to motion events. This is likely due to the fact that motion and change of state events are linguistically encoded more often than other event types and that motion events have fewer path types, making them easier to study than change of state events (Spring 2014; Spring and Ono 2024). However, there are studies that note the parallels between resultative constructions or change of state events and motion events including those that show similar tendencies in framing for both event types for particular groups of speakers (e.g., Aske 1989; Beavers et al. 2010; Goldberg and Jackendoff 2004; Ono 2004, 2009]). Furthermore, the Event Conflation Finder (ECF) program (Spring and Ono 2024) has made it easier to extract and categorize instances of event conflation in not only motion events, but also change of state events, and therefore this paper endeavors to include all five types of conflated events.

1.2 The linguistic typology of event conflation

Talmy (1991) also claimed that a typology of languages was possible based on whether or not speakers generally used verb-framing or satellite-framing. However, subsequent works have since suggested a number of changes to it. For example, Slobin (2004) noted that some languages encode both the co-event and main-event in linguistically equivalent forms, such as serial verbs, and that such encoding should be considered a different type, which he dubbed equipollent-framing. For example, (1a) and (1b) show that in Mandarin, both the main-event and co-event are encoded in verbs. Though Talmy (1991) originally considered the first verbs (zǒu, tī) to be the main verb of the sentence and the second verbs (jìn, kāi) to be a satellite, works such as Chen and Guo (2009) have also argued that these should be considered equivalent forms. However, Matsumoto (2003, 2017] argues that jìn in (1a) and kāi in (1b) should be considered a satellite because the verbs zǒu and tī are the semantic heads of their respective sentences. On the other hand, he suggests that Japanese compound verbs, such as kake-nobotta in (1c) and fumi-tsubushita in (1d), should be considered an equipollently-framed expression because it is a single, inseparable unit that acts as the semantic head of the sentence. Other works such as Beavers et al. (2010) and Croft et al. (2010) have also noted that whether or not one considers both serial verbs (e.g., 1a) and compound verbs (e.g., 1b) to be equipollent-framing or not, there are inherent differences in them that might require further nuanced study.

The linguistic typology proposed by Talmy (1991], 2000]) was not meant to be based solely on what linguistic expressions are available in any given language, but on which conflation pattern (i.e., verb-framing or satellite-framing) were more colloquially prevalent in a language. Of course, framing choices are also constrained by what linguistic resources are available (Beavers et al. 2010; Croft et al. 2010). For example, English speakers do not use serial verbs because they are not grammatically allowed in English. Similarly, most of the Romance languages, such as Spanish and French, typically do not allow for manner of motion verbs to be combined with path satellites in boundary crossing events (e.g., Hendriks and Hickmann 2015). Some studies have suggested that certain types of manner or path are more noticeable or salient than others (e.g., Akita and Matsumoto 2020; Spring 2021), and thus more likely to be included in a conflated event. Others have suggested that processing constraints may also explain some choices (e.g., Beavers et al. 2010). However, even when these differences are accounted for, there can be considerable variation within the speakers of a given language (Lewandowski 2021).

To this end, many studies have been conducted to observe which patterns are most commonly used by native speakers of various languages in order to categorize them. For example, Slobin (2004) used story-telling data to show that while the speakers of some languages do indeed tend to favor either satellite- or verb-framing, others show a tendency to use both in almost equal amounts. He noted that specifically, those languages which often use serial verbs, e.g., Mandarin, often use both ‘equipollent-framing’ (or satellite-framing by Matsumoto’s standard) and verb framing approximately equally, leading him to push for the inclusion of a third type of language: equipollently-framed languages. This notion has been supported by other theoretical and experimental studies that have also shown an intermittent amount of both types of framing by native speakers of Mandarin (e.g., Chen and Guo 2009; Spring and Horie 2013). However, Slobin (2004) also suggested that rather than adding a third language type, it might be more appropriate to reconsider Talmy’s (1991]) two-tier typology as a cline, based on how often manner is typically encoded in various linguistic structures or at all. Studies such as Ibarretxe-Antuñano (2009) also suggest that a cline might be more appropriate than a complete two-tier typology but suggest that the cline could be based on how often path (i.e., the main event) is encoded in the main verb of the sentence rather than based on manner (i.e., the co-event).

Though the aforementioned research is all based on motion-event research, similar studies have also been conducted with change-of-state expressions. For example, studies such as Washio (1997) and Mateu (2012, 2014] argue that Japanese and many Romance languages are restricted in their use of satellite-framed expressions in change-of-state events. Ono (2004, 2009] observed novels written in English and Japanese and their translations into the respective languages and found that English novels and translations consistently favor the use of more satellite-framed expressions both for motion events and change of state events, whereas Japanese favors the use of verb-framed expressions for both event types. Furthermore, Spring (2014) conducted video description tasks with change-of-state events and noted the same trend; English speakers used far more satellite-framed expressions than verb-framed expressions and Japanese speakers used far more verb-framed expressions than satellite-framed ones. Spring and Ono (2024) also found that English novels tended to contain more satellite-framed change-of-state expressions than verb-framed expressions when calibrating their event conflation finding tool. Therefore, though there is much less research on change-of-state events than motion events, it seems that similar typological tendencies can be observed in both event types.

Thus, though there is some consensus regarding the fact that event conflation exists and that some languages tend to use certain framing patterns more than others, the studies above show that there are still several unsolved matters. Specifically, there is still some disagreement about the exact classification of some languages, and more studies are still being conducted to examine how strong tendencies towards verb-framing is within a particular language and what factors might affect framing choices intralingually. Furthermore, more studies are still required to determine how congruent framing patterns are for languages across event types, although preliminary findings from studies comparing motion events and change-of-state events seem to suggest the same tendencies exist for both event types in English and Japanese. Therefore, it is important to further investigate not only what tendencies exist within general language types, but also how much variation exists between them, and how much variation exists across event types. This paper hopes to shine further light onto these issues by observing the framing tendencies of speakers of several languages on a typological level and language specific level across several event types.

2.1 Expressing motion events in a second language

Some studies find that L1 learners can be confused by L2 patterns that are either not common or disallowed in their L1. For example, Inagaki (2001, 2002] found that manner of motion verbs with prepositional phrases, such as in (2a), were particularly difficult for L1 Japanese learners of English because this pattern allows two meanings in English (i.e., one of location, as in 2b, and one of motion, as in 2c), whereas it only takes a location interpretation in Japanese. Therefore, L1 Japanese learners generally misinterpreted such motion expressions in English as location expressions.

However, much more attention in the literature has been paid to how L2 speakers choose to encode events, as there is often some overlap in L1 and L2 mappings, and the choices that learners make can potentially be indicative of underlying preferences for one framing pattern over another due to the cognitive focus of the L1. Here, a large number of studies agree that L2 learners tend to mimic their L1 patterns in the L2 until very high levels of proficiency (e.g., Brown and Chen 2013; Cadierno 2008, 2010]; Mano et al. 2019; Navarro and Nicholadis 2005; Spring 2014; Spring and Horie 2013; Yoshinari et al. 2021, 2023]). For example, Cadierno (2010) examined both verb-framed and satellite-framed learners of L2 Danish (a satellite-framed language) and found that learners with a satellite-framed L1 were much more likely to use satellite-framed expressions in their L2 than those with a verb-framed L1. Similarly, several studies have suggested that L1 Mandarin learners (either equipollent-framed following Slobin 2004 or satellite-framed following Matsumoto 2003) exhibit greater use of and variety of satellite-framed expressions in L2 English at earlier stages than their L1 Japanese counterparts (verb-framed) (e.g., Brown and Chen 2013; Spring and Horie 2013). Some studies have also noted that when learners whose L1 is satellite-framed learn a verb-framed L2, they also tend to attempt use satellite-framed expressions in the L2. For example, Cadierno and Lund (2004), Cadierno and Ruiz (2006), and Navarro and Nicholadis (2005) found that Danish and English speakers learning of Spanish overly used manner of motion constructions, generally mentioning the manner of motion more than native speakers and often attempted to connect manner of motion verbs to pseudo-satellites.

It should be noted that not all studies are in perfect agreement regarding the aforementioned tendencies. Specifically, Brown (2015) conducted a video description task with L1 Japanese and L1 Chinese learners, specifically focusing on how often they would explain the manner of motion, finding that the two groups of learners seemed to focus on it about the same amount. Filipović and Vidaković (2010) found that L1 Serbian learners (typically classified as satellite-framed) were more likely to use satellite-framed expressions in L2 English than in L1 Serbian, concluding that L2 framing patterns may be influenced by more than just L1 type. Previous studies offer several suggestions as to why this could happen, including economy of the L2 form (Filipović and Vidaković 2010), lexical and processing demands (Brown 2015), or the confusion that arises when some L1 patterns are available for transfer into the L2 but not others (e.g., Hendriks and Hickmann 2015). However, it could also be due to variability within native speaker groups (e.g., Lewandowski and Özçalışkan 2023) and the samples that were taken in these studies. Finally, some studies have also noted that learners’ ability to acquire L2 framing can be influenced by the directionality of the motion events obesrved (e.g., Brown and Gullberg 2008; Cadierno 2004, 2010]; Hendriks and Hickmann 2015; Lewandowski and Özçalışkan 2023). Regardless, it seems that while many studies have found that L1 framing patterns persist in L2 usage, there are some cases where this is not true, and more study is still required especially in different registers and under varying circumstances.

2.2 Expressing change-of-state events in a second language

Though there are far fewer studies on the L2 acquisition of change-of-state events as compared to motion events, there is some theoretical and experimental evidence to suggest that the trend of difficulty in L2 acquisition depending on the L1-L2 pairing is congruent between motion and change-of-state events. From a theoretical perspective Washio (1997) denotes a difference between “weak” resultatives and “strong” resultatives, which are congruent with satellite-framed change-of-state events. Specifically, weak resultatives are those in which both the verb and satellite imply change of state, such as in (2a) where both the verb, paint, and the satellite, green, imply a change of color. Conversely, strong resultatives are those in which the verb does not inherently imply the change of state, such as in (2b) where the verb, hammer, does not necessarily result in something becoming flat. As evidenced by (3a–d), strong resultatives are allowed in English, but not in Japanese, which Washio (1997) suggests prevents L1 Japanese learners of English from easily interpreting them. Furthermore, according to Mateu (2012), this distinction between strong and weak resultatives is also valid for prepositional-like resultatives, as illustrated in the ungrammatical Italian sentence in (3e). However, it should be noted that although Korean is generally considered to be verb-framed in terms of motion (e.g., Choi 2009; Park 2019; Sun et al. 2022), it does allow strong-resultatives as illustrated in (3f) and suggested by Lee (2004).

Furthermore, Spring (2014) used both an image-sentence matching task similar to that of Inagaki (2001) and a video-description task similar to Brown 2015; Spring and Horie 2013 to investigate the L2 English acquisition of both strong and weak resultatives by L1 Japanese and L1 Mandarin learners. The findings of the former task indicated that L1 Japanese learners could correctly interpret weak resultatives, but not strong resultatives, whereas L1 Mandarin learners had little difficulty interpreting both, as predicted by Washio’s (1997) findings. Furthermore, the results of the video description task indicated that learners generally used L2 framings similar to other native speakers who conducted the task in their L1. Specifically, L1 Japanese learners in L2 English and Japanese native speakers in Japanese tended to use verb-framed expressions more prevalently, whereas L1 Mandarin learners in L2 English and Mandarin native speakers in Mandarin tended to use more satellite-framed (or equipollently-framed) expressions.

2.3 Unresolved issues

Based on the results of several of the previous studies introduced in Section 2.1 and their own experimentation, Yoshinari et al. (2021, 2023] created the Asymmetry Hypothesis, which proposes that learners of certain L1s have more difficult acquiring the predominant framing patterns of an L2 than others. Specifically, they suggest that L1 verb-framed language (e.g., Japanese) learners have more difficulty acquiring satellite-framing in an L2 than learners whose L1 is satellite-framed or ‘mixed-framed’ (referring to how frequently native speakers use non-verb-framing). However, their data is limited to three languages (English, Hungarian, and Japanese) and they also note that there are considerable fine-grained differences within languages and task types. Furthermore, their datasets were rather limited in size and mostly dealt with the differences in the number of expressions of learners in different bands rather than across a scaled range. Therefore, it is still important to test this theory both generally (i.e., across language “types”) and specifically (i.e., across languages and a range of proficiency levels) to explore how generalizable it is.

Finally, though producing L1-framed expressions in an L2 can create unnatural turns of phrase and might be found to be less native-like, it is still unclear how much of a problem this truly is in terms of communication. Naturally, L1-L2 mismatches that cause incorrect interpretation (Inagaki 2001, 2002]) or disallowed framing patterns (e.g., Hendriks and Hickmann 2015; Spring 2014) will generally cause communication breakdown in that the intended message is not correctly interpreted either by the learner or the listener. However, as noted by a number of the aforementioned studies, many times the phrasings that L2 learners use are not necessarily grammatically incorrect; they are simply not what is colloquial (e.g., Brown 2015; Cadierno 2008; Navarro and Nicholadis 2005; Spring and Horie 2013; Yoshinari et al. 2021, 2023]). Nevertheless, if L2 phrasings stray too far from what is standard in an L1, they can be misconstrued as having a deeper meaning or connotation that was not intended by the L1 learner. For example, consider the statements in (3a–d). Though the expressions (3a) and (3c) technically have the same meanings as those in (3b) and (3d), respectively, not only are they verb-framed, which is less common in English, but they have the potential to give a false impression to a native speaker. Specifically, by putting the information contained in come in (3b) (that she was moving towards the speaker) in a relative clause headed by which, it implies that this information is not necessarily relevant to the main clause. Similarly, not only is (3c) less common amongst native speakers of English, it places excess information on the causal relationship between the act of pushing and the act of entering, diverting much more attention to the push and implying there was something particular about it, whereas there is no such implication in (3d).

However, it is still unknown how much the use of less common framing patterns in the L2 affects communication and how impactful the L1-L2 framing mismatch problem is for L2 ability in general. Furthermore, it is still unclear how likely learners of particular L1s are to utilize their L1 framing patterns in an L2 when they are not prompted by a specific motion or change of state event. Finally, though some languages are argued to fit into certain typological groups (or be on a cline), there are significant differences in the linguistic resources available, especially between motion-events and change-of-state events (e.g., Korean is well-argued to be verb-framed for motion events, but seems to allow strong resultatives which means that it might be more satellite-framed in terms of change-of-state expressions, whereas Japanese is verb-framed for both event types).

Therefore, in order to attempt to fill in these gaps in the previous literature, this study looks at the framing patterns of conflated events in high stakes tests in an independent production task (i.e., a task with an open-ended prompt). If the Asymmetry Hypothesis is true, one would expect there to be more of an association between L2 ability and L2 framing choices for learners with L1s that are of a different type than for learners whose L1s are of the same type as the L2. However, this has not yet been made clear via standardized test scores that include both a range of proficiency levels and native speakers of various L1s. Furthermore, it is unclear if the Asymmetry Hypothesis is appropriate for both motion and change-of-state events, and it is still unknown if should be applied to specific languages or if it can be applied to groups of languages. Thus, this study is innovative in its use of a range of L2 proficiencies, a wide range of L2 learners, data from the verified scores of a learner database in a communicative task, a focus on correlation between proficiency and L1-like usage, and a focus on wider range of event types than previous studies. Specifically, this study seeks to answer the following research questions:

1. Do L2 learners exhibit a tendency to use L1 framing patterns in high-stakes testing?

2. Do L2 learners who use L1 framing patterns generally receive higher scores, and does the strength of this association vary by L1 type?

3. Are there differences in the above trends when L1 types are observed versus when specific L1s are observed?

3.1 Dataset

The data for this study was taken from the independent writing task of the TOEFL iBT ^® Public Use Dataset. This dataset was selected because it represents a gold-standard standardized test with reliable rating and learners with a wide variety of L1s who are all learning a satellite-framed L2 (English). The independent writing task was observed as opposed to the integrated writing task because the former provides only a short prompt, whereas the latter requires test-takers to write based on source material which would likely influence learners’ framing choices. I selected learners whose L1s could most clearly be categorized into groups according to previous studies. Specifically, Germanic and Slavic languages were used to represent satellite-framed L1s as they are often considered to be very clearly satellite-framed in terms of tendency to include co-events, tendency to encode path in satellites, and availability of linguistic resources (Filipović and Vidaković 2010; Lewandowski 2021; Mano et al. 2019; Slobin 2004; Talmy 1991, 2000]; Zlatev and David 2003; etc.). Romance languages, Korean, and Japanese were used to represent verb-framed L1s because they are generally considered to be less likely to encode co-events, tend to encode path in main verbs, and often have limited satellite-framed options (Brown and Gullberg 2008; Beavers et al. 2010; Choi 2009; Hendriks and Hickmann 2015; Inagaki 2001; Ono 2009; Park 2019; Sun et al. 2022; etc.). Finally, three languages (Chinese, Vietnamese, and Thai) that have been argued in the literature to employ an intermittent tendency to encode manner, encode path in main verbs, and routinely employ serial verbs or have been argued to be equipollently-framed in the literature were selected as mixed-framed languages (Brown 2015; Chen and Guo 2009; Ly 2020; Pham 2013; Takahashi 2020; Tran and Le 2022; Zlatev and David 2003; etc.). The numbers and specific languages are represented in Table 2.

For further analysis, I also did the same comparisons with groups of learners from specific languages that had at least 20 participants (i.e., German, Chinese, French, Spanish, Japanese, and Korean) to check for more language specific patterns.

3.2 Procedure

I used the Event Conflation Finder (ECF) “full data” mode to extract all potential instances of path encoding from the learners’ writing samples and organize them for proofreading and further analysis (Spring and Ono 2024). The ECF is an automated tool that automatically extracts possible instances of “path” (i.e., main events) from a folder of text files and creates a list of marked sentences that it is sure contain path, and a list of marked sentences that it suspects contain path. It utilizes a large language model (i.e., SpaCy) for parsing and then original commands to detect main events. According to Spring and Ono (2024), using the ECF “full data” mode to create lists of all sentences that potentially include path and then making decisions on those sentences is far more accurate than a solo human-rater, as satellite, particularly, can be difficult to recognize as existing in motion of change-of-state events (a single human rater missed as many as 57 % of satellite-framed expressions of change-of-state events, a pair of raters missed 13 %, and the ECF did not miss any). Furthermore, using this tool in this manner helps to reproduce replicability, as other researchers can also obtain the dataset and tool, and obtain the same list.

I then checked the sentences extracted by the ECF individually for accuracy, as suggested by Spring and Ono (2024). I felt qualified to do so as I have a PhD in comparing motion and change-of-state event conflation expressions and have published extensively on the topic. When making decisions, I counted a clause or sentence as a valid motion expression if it included a path of motion and counted it as a change-of-state expression if it included a change of state (i.e., the main-events of these two event types). If a verb could take either a main-event of co-event meaning (e.g., climb can be either a manner of motion or a path upwards and grow can either be a manner/cause of change or imply a change to a larger size), I followed the manner/result complementarity generalization (Rappaport-Hovav and Levin 2010) and determined the verb to be taking one of the two meanings. In the event that participants combined a true main-event verb with a main-event satellite (e.g., in the expression he entered through the window, both enter and through imply a path of motion), I counted this as one instance of verb-framing and one instance of satellite-framing because one main-event is encoded in each, following Spring and Ono (2024).

As the participants were justifying personal opinions, many of the expressions were either personal examples, which often had actual physical motion (i.e., a participant recounting a life episode in which many people walked through a busy shopping center, see Table 3), or reasons, which often had figurative motion (e.g., “the driving forces which [are] driven from the market”, “the population has reached a worrying level”). I included both actual physical motion expressions and figurative motion expressions because the basic expressions were the same and could have been expressed either with a satellite- or verb-framed expression. Many participants used change-of-state expressions to describe changes they expected, worried about, or as reasons for their opinions and were more prevalent in the data (there were 308 motion expressions and 809 change-of-state expressions).

According to Spring and Ono (2024), the ECF checks for specific phrasal verbs taking aspect, correlation of activities, or realization of goals and includes these as “other” conflated events. I included these in the study, but only sparingly as these final three forms of conflated events are understudied and Spring and Ono (2024) doubt the ability of the ECF to properly find all instances of these final three event types.

I further categorized uses of satellite-framed expressions into those with “bare” verbs (Navarro and Nicholadis 2005: those that contain no information regarding the main- or co-event, but are grammatically necessary for the satellite to express one of Talmy’s five event types, such as get, go, and put) and those with “manner” verbs (those that provide information about how or why the main-event occurred such as run, fly, and pull). I also counted the instances of unique event conflation expressions per learner text. Exact examples of each type of framing from the data and how I coded them are provided in Table 3.

For each learner text, I checked the number of satellite- and verb-framed expressions for both motion and change of state events and calculated the percentage of each event type that was satellite-framed. Furthermore, I also used the counts of other types of event conflation (i.e., aspect, correlation of activities, and realization of goals) to create a measure of overall satellite-framed expression usage. Finally, I also counted the number of satellite-framed expressions using manner verbs and the number of unique satellite-framed expressions. These measures are summarized in Table 4.

In order to answer the first research question, I compared the rate at which the measures in Table 2 were used by each L1 type, examining the descriptive data, and making comparisons via Kruskal-Wallis tests with Dunn’s Test for post-hoc analysis and partial eta squared for effect size. This statistical methodology was used because the data tended to skew towards 0 as many participants did not use every single type of framing in their responses. Shapiro Wilk’s tests confirmed that most measures (all except for total satellite-framing usages and satellite-framing variety) failed the test of normalcy, and so non-parametric tests were used throughout. While in most cases a combined model would be used to observe overall trends, because the measures themselves are different and there is incongruency between the data by L1 type and by L1 (i.e., the latter removes some of the data points), a combined model for testing all measures across all data sets was not possible. To answer the second research question, I compared the magnitude of correlation between rater scores and the measures in Table 2 by conducting Spearman’s correlation tests. Spearman’s correlation was used instead of Pearson’s due to the previously discussed non-normal nature of the data and the ordinal, yet normalized, nature of the rater scores. To answer the third research question, I repeated these tests not only for language types, but also by language. I then compared the magnitude of these correlations by language type and looked for general trends. I also extracted example expressions to provide qualitative data and made comparisons of the variety of expressions amongst L1 types and L1s.

4.1 Descriptive analysis and comparison of learner scores

The descriptive data for the measures per typological group is presented in Table 5 and visualized in Figure 1. ANOVA tests showed that there were differences in the groups’ use of satellite-framed motion expressions, overall satellite-framed expressions, the percentage of motion expressions that were satellite framed, the number of manner verbs used in satellite framed expression, and the amount of satellite-framed expression variation, as shown in Table 6. The results suggest that while there were no differences in how frequently the different groups used verb-framing or satellite-framing to describe change of state events, there was a general trend for satellite-framed L1 learners to use more typical satellite-framed patterns (i.e., path satellites, greater frequencies of path satellites to path verbs, more manner verbs and greater variety of satellite-framed expressions) for both motion events and in general than mixed-framed L1 learners, who in turn used these patterns more than verb-framed L1 learners. However, looking at effect sizes, this trend was especially strong for the number of total satellite-framed expressions used and the variety of satellite-framed expressions used.

Figure 1:

Visualization of descriptive data by L1 type.

The same descriptives and comparisons by language rather than L1 type are provided in Tables 7 and 8 and visualized in Figure 2. The results of these comparisons are less conclusive, with the only statistically significant difference being that L1 German speakers clearly used more satellite-framing in total than L1 Japanese speakers.

Figure 2:

Visualization of descriptive data by L1.

4.2 Comparison of framing tendencies across languages and types

In looking at the qualitative data, such as the specific satellite-framed expressions used by various L1 learners in Table 9, it becomes clear that when learners with verb-framed L1s, and to a lesser extent those with equipotently-framed L1s, used path satellites, they tended to use more bare verbs and path verbs. In comparison, L1 German speakers tended to use a wide range of manner verbs and various particles despite there being far fewer L1 German speakers than L1 Chinese, Korean, Spanish, or Japanese speakers. Furthermore, there was a tendency for satellite-framed L1 learners to use path verbs with path satellites in motion events, such as reach to and move to, and a similar pattern in for change of state events of weak resultatives such as join together and translate into. With regards to other satellite-framed expressions (i.e., correlation of activities, aspect, and realization of goals), most of these appeared in all learners’ language as quite common set examples of phrasal verbs taking one of these meanings (see Spring and Ono 2024).

In comparing the path-verbs of the learners shown in Table 8, it is also apparent that while L1 German speakers did use many path verbs, especially for change of state events, they seem to show less variation (in comparison to other L1s) than they did for satellite-framed expressions. Though there were fewer L1 German learners in total, there was still a long list of expressions, implying a lot of variety, in Table 9, whereas Table 10 shows much less variety in change of state path verbs, in particular. Interestingly, L1 French and L1 Spanish learners did not use as many different path verbs in motion events as L1 Chinese, Korean, Japanese, or even L1 German learners. This is likely due in part to the fact that there were simply many more L1 Chinese, Korean, and Japanese learners, but it does not explain why L1 German learners had a wider variety of path of motion verbs, albeit still less variety in terms of path of change of state verbs. These results suggest that most L1 verb-framed language learners showed greater variation in L1 framed expressions, as would be expected according to the theory of event conflation typology, but this was not true for L1 French and L1 Spanish learners in terms of motion events.

4.3 Analysis of association between L2 framing and rater scores

Finally, the results of the correlation analyses between various measures of framing usage and rater scores are shown in Table 11 and visualized in Figures 3 and 4. The results show that in general, using more various types of satellite-framed expressions seems to be associated with higher scores only for L1 verb-framed language learners. Conversely, higher scores were positively associated with more verb framed patterns by both satellite-framed and mixed-framed L1 learners. With regards to specific languages, most L1 groups exhibited significant correlations to rater scores for verb-framed change of state expressions and greater variety of satellite-framed expression types. French and Korean L1 learners also exhibited some correlation between rater scores and satellite-framing usage overall and for greater use of manner verbs. However, the same trend was not observed for Japanese and Spanish L1 learners. Finally, it should be noted that the one common trend amongst all L1 verb-framed language learners seems to be that the greater the variety of satellite-framed expressions used by the learners, the greater their scores tended to be. Although L1 German speakers show some tendency for this to be true as well, this does not appear to be true for all of the L1 satellite-framed speakers in general.

Figure 3:

Magnitudes of correlation (r _s ) between framing usage and rater score by L1 type.

Figure 4:

Magnitudes of correlation (r _s ) between framing usage and rater score by L1.

Taken together, the results suggest that there is an overall trend for more satellite-framed expressions to be associated with higher scores for L1 verb-framed learners than for learners with other types of L1s. Specifically, for only the verb-framed learners group there is some low level correlation between rater scores and the number of satellite-framed motion events, overall number of satellite framed expressions, percentage of manner verbs used, and amount of satellite-framed variation. While there is some correlation between verb-framed change of state expressions and raters scores for verb-framed L1 learners, and a couple of correlated measures for the other language groups, verb-framed language learners show the only clear pattern of mostly consistent correlation between raters scores and these measures. Because correlation is not the same as causation, it is unclear if the use of these expressions spurred higher scores or if having higher proficiency resulted in the learners having more ability to use satellite-framed expressions. However, the fact that this pattern of correlation persist across only one type of L1s suggests that it is more likely that learners with satellite- and mixed-framed L1s simply were able to use more of these expressions from lower proficiency levels, which is also predicted by the Asymmetry Hypothesis (Yoshinari et al. 2021, 2023]).

It should be noted that although at the L1 type general level, consistent correlations were found that support the Asymmetry Hypothesis, the same trends were not found when looking at specific L1s individually. For example, both L1 German and L1 French learners showed quite a bit of correlation across many of the measures, whether verb- or satellite-framed. Furthermore, though the group of all L1 verb-framed learners showed much more correlation between satellite-framed expressions and rater scores than verb-framed expressions and rater scores, this trend did not appear for any specific L1 group. This could be due to the fact that the individual L1 groups simply represented less data and thus are less indicative results, but it could also be due to individual differences in the languages themselves.

Another finding of this study seems to be that while L1 German learners of English seemed to exhibit a high level of variety in their satellite-framed expressions, in general, and in their verb-framed expressions of motion, they did not exhibit as much variety in their verb-framed expressions of change of state. Furthermore, though L1 French, Spanish, Japanese, and Korean learners all exhibited great variety in their verb-framed expressions of change of state, L1 Japanese and Korean learners actually exhibited more variation in their verb-framed expressions of motion. Conversely, whereas L1 French, Spanish, and Japanese learners exhibited little variety in their satellite-framed expressions, both L1 Korean and L1 Chinese learners exhibited quite a lot of variation. Though event conflation typology would suggest that L1 Chinese learners would likely have more than L1 verb-framed learners, it does not explain the amount of variety present in L1 Korean learners’ satellite-framed expressions.