Introducing Content-Based Structured Input to English-Medium Instruction: Evidence from Verb Pattern Acquisition in a Disciplinary Course for University Students in Poland

Abstract

The rise of English-medium instruction (EMI) in higher education presents challenges for supporting language development alongside disciplinary learning, as it is typically delivered by content specialists with limited time and little or no background in language pedagogy. Against this backdrop, this study examines the effectiveness of computer-assisted content-based structured input (CBSI), compared to input flood (IF) and unmodified disciplinary input (Control), in enhancing students’ acquisition of English verb patterns in an EMI university course in Poland. All participants received the same asynchronous, computer-assisted disciplinary instruction, aligned with the course syllabus. The groups differed only in the type of input enhancement, which was developed by a language specialist in coordination with the course instructor. A split-block design was used to measure learning gains through a pre-test, post-test, and delayed post-test. Students in the CBSI group showed significantly greater and more sustained improvement than those in the IF and Control groups. These findings support CBSI as an effective and pedagogically feasible way to promote language development in EMI, integrating linguistic and disciplinary concerns while respecting the roles and integrity of both domains.

1. English-Medium Instruction: Definitions, Potential, and Limitations

English-medium instruction (EMI) is commonly defined as “the use of the English language to teach academic subjects (other than English itself) in countries or jurisdictions where the first language of the majority of the population is not English” (Macaro, 2018, p. 19). Given the unquestionable status of English as a global language (Galloway & McKinley, 2022, p. 2), EMI has recently become a key feature of internationalization in higher education, with universities worldwide adopting English-taught programs to boost their global profile and attract international students (Doiz & Lasagabaster, 2021, p. 58; Galloway & Rose, 2021). This is reflected in the tenfold rise in the number of EMI programs in European higher education—from 725 in 2001 to 8089 in 2013—followed by a threefold surge to 24,043 in 2023/2024 (Wingrove et al., 2025, p. 1).

In Poland, EMI emerged in the early 1990s as part of post-communist higher education reforms further accelerated by Poland’s accession to the EU in 2004 (Mikołajewska & Mikołajewska, 2022). As a result, the number of international students surged more than 25-fold, from 4259 in 1990/91 to 107,100 by 2023/24 (Główny Urząd Statystyczny [Statistics Poland], 2011, 2024), as universities launched English-taught programs to attract international students, subsidize their budgets, offset demographic decline, and enhance competitiveness (Briggs et al., 2018).

While disciplinary knowledge remains the primary focus of EMI (Pecorari & Malmström, 2018, p. 499), many students also view it as a way to improve their English and gain a competitive edge in academic or professional contexts (Bowles & Murphy, 2020). This dual promise is often echoed in policy discourse, where language development is frequently assumed to occur as a natural side effect (Hillman et al., 2023, p. 1). The linguistic promise is further reflected in another widely cited definition of EMI, which describes it as the use of English to teach academic subjects “to improve students’ academic English proficiency” (Taguchi, 2014, p. 89). Thus, despite its content orientation, EMI is also expected to support language development (Galloway & Rose, 2021, p. 37).

In practice, however, this dual promise often goes unfulfilled. EMI courses are taught by subject specialists with little to no training in language pedagogy (Aguilar, 2017; Airey, 2020) and with varying levels of English proficiency (Mikołajewska & Mikołajewska, 2022). As a growing body of research confirms, language issues—except occasional lexis clarification (Macaro, 2020)—are seldom, if ever, addressed in EMI classrooms (Fürstenberg et al., 2022, p. 294; Hadingham, 2024, p. 584). Content lecturers typically view language support as beyond their remit and capacity, already busy with subject instruction (Airey, 2012; Dang & Vu, 2020)—let alone when delivered through a foreign language (Aguilar, 2017; Mikołajewska & Mikołajewska, 2022). As a result, while EMI has been shown to contribute to improvements in students’ general and discipline-specific vocabulary, more complex aspects of language development—such as grammatical accuracy and complexity, or discourse organization—tend to remain largely unaffected (Del Mar Sánchez-Pérez, 2023, p. 374). Similarly, although EMI students may develop receptive skills, they often struggle with productive ones, especially when required to use more complex academic language (Kamaşak et al., 2021, p. 11; Mikołajewska & Mikołajewska, 2022).

None of this should come as a surprise. As Aguilar (2017, p. 3) explains, “unless a systematic approach [is] used to learn and develop disciplinary and academic discourse (…), the academic language proficiency of participants would hardly improve.” Yet the underlying assumption of linguistic benefits in EMI is precisely the opposite—that English will be acquired through exposure alone, without explicit instruction or defined learning goals (Hillman et al., 2023; Pecorari & Malmström, 2018, p. 502). Research shows this is not the case. Effective support requires intentional pedagogical measures, ideally involving collaboration between language and content specialists, given that the latter typically lack training in language pedagogy.

As Malmström and Zhou (2025) explain, such collaboration can take place at institutional, departmental, or individual levels through activities such as co-planning, team teaching, and assignment support (pp. 3–4). Evidence shows that this approach fosters EMI success: content teachers’ attitudes toward EMI become more positive (Lu, 2022, p. 10), and they develop greater awareness of the interdependence of content and language (Akıncıoğlu, 2024; Malmström & Zhou, 2025, p. 6). Yet such collaboration remains rare (Pecorari & Malmström, 2018, p. 498), as it is typically constrained by disciplinary divides, weak institutional support, time pressures, and unequal power relations (Malmström & Zhou, 2025, pp. 4–5).

This is all particularly concerning, as language proficiency consistently emerges as a key predictor of EMI success. Research indicates that students with stronger English skills—particularly those with specialized academic English preparation—tend to perform better in EMI courses (Hadingham, 2024; Rose et al., 2020). In contrast, insufficient language support may hinder content learning (Yuksel et al., 2023). To exemplify, Curle et al. (2024) found that English proficiency was the single strongest predictor of success in EMI programs, accounting for nearly 90% of the variance in achievement and showing a near-perfect correlation with content scores. Interview data confirmed this pattern, and even after controlling for variables such as gender and motivation, proficiency remained the most consistent predictor of EMI achievement, reinforcing findings from other studies that link L2 English skills directly to student academic outcomes (e.g., Soruç et al., 2021; Soruç et al., 2024; Trenkic & Warmington, 2019).

The content–language interdependence is hardly unexpected. As Richards and Pun (2022, pp. 73–75) explain, learning through EMI involves understanding disciplinary concepts and verbalizing this understanding to support the thinking and literacy processes that each subject requires. Since all these processes must be carried out in English, the underlying proviso is that “the conceptual foundations of a discipline cannot be understood or realized except through the use of language,” (p. 75). This echoes the long-established connection between language use and disciplinary knowledge development, as highlighted in other EMI-focused and related studies (e.g., Airey, 2020; Halliday, 2007; Lyster, 2007).

Despite this evidence, many EMI programs still operate under what Macaro (2018, p. 233) terms the “ostrich model,” ignoring language challenges. To remedy the situation, some scholars advocate for a “CLILised” EMI model (Moncada-Comas & Block, 2021), which would involve implementing content and language integrated learning (CLIL) methodology (Coyle et al., 2010; Coyle & Meyer, 2021). Yet CLIL has proven challenging even in its original primary and secondary school contexts (Lyster & de Zarobe, 2018), where language and content instructors are more equally involved—making its application in higher education even more aspirational (Hillman et al., 2023). The practical constraints of EMI—delivered by content experts with a clear disciplinary focus, varying levels of English proficiency, and limited time due to the demands of content delivery (Airey, 2012; Dang & Vu, 2020)—call for bespoke, low-intervention solutions that address linguistic challenges without compromising subject teaching or placing unrealistic expectations on content instructors.

2. Processing Instruction and Structured Input

The central idea behind the current research is that content-based structured input (CBSI) (Walenta, 2018, 2019) can be effectively implemented in EMI settings to target morphosyntactic aspects of student language development—areas that, as discussed above, typically do not improve through exposure alone, yet are vital for successful academic performance. To lay the groundwork for this approach, it is important to first outline the key principles of processing instruction (PI) and its core component—structured input (SI) (Benati & Lee, 2010; VanPatten, 2015a)—on which CBSI is based.

PI is a pedagogical intervention grounded in input processing theory, which posits that learners tend to prioritize meaning over grammatical form due to limited processing capacity (Lee & Benati, 2009, pp. 3–4). The core assumptions of input processing theory are captured in two overarching principles: (1) the Primacy of Meaning Principle, which holds that “[l]earners process input for meaning before they process it for form” (VanPatten, 2004, p. 11), and (2) the First Noun Principle, which states that “[l]earners tend to process the first noun or pronoun they encounter in a sentence as the subject or agent” (VanPatten, 2004, p. 22). PI aims to counter both tendencies by promoting more accurate form–meaning mapping, thereby leading to grammatically richer intake (VanPatten, 1996, p. 8).

The key instructional technique in PI is SI, which manipulates input so that morphosyntactic cues become essential for message interpretation. For example, in the sentence Jane laughed yesterday, learners may rely on the adverb yesterday and overlook the past tense marker -ed. SI tasks remove such redundant lexical cues, ensuring that learners must attend to grammatical form to extract meaning (Benati & Lee, 2008, pp. 26–27; VanPatten, 2015a). Crucially, SI never sacrifices meaning—it works precisely because it maintains a focus on meaning while increasing the salience of form (Benati, 2019; VanPatten, 1996, p. 67).

SI activities fall into two main types: referential and affective. Referential activities require learners to attend simultaneously to form and meaning, and include interpretation tasks with clear right–wrong answers—for example, matching sentences to images or choosing between interpretations (VanPatten, 1996, p. 170; Zhong & Benati, 2024, p. 4). Affective activities invite personal responses and are often used to reinforce form–meaning connections established earlier (Benati, 2019, p. 355).

Numerous studies (Benati & Lee, 2010; Henry et al., 2024; Lee & Benati, 2009; VanPatten, 2015b) have demonstrated that learners receiving SI perform better than those receiving rule-focused or output-based instruction. Importantly, SI has a positive effect on both interpretation and production tasks (Benati, 2023), including at the discourse level (Benati & Lee, 2010). This is because structured input practice helps learners build form–meaning connections during input processing, which in turn supports the development of their underlying linguistic competence and enables access to a given linguistic feature in production (Benati, 2023, p. 6).

Moreover, learners receiving structured input alone often achieve gains comparable to those who receive both SI and explicit information on the target form (VanPatten & Oikkenon, 1996), underscoring SI as the principal mechanism behind observed learning effects (Benati & Lee, 2010, pp. 67–76). Recent research also confirms that referential activities—without the addition of affective ones—are the key mechanism in helping learners process morphosyntactic cues more effectively (Zhong & Benati, 2024).

Studies have also found no significant difference in outcomes between classroom-based and computer-assisted SI delivery (Lee & Benati, 2007), and recent research confirms its effectiveness in digital formats (Benati, 2023).

3. Enter Content-Based Structured Input

CBSI draws on all of the above and extends this logic to content-based contexts (Walenta, 2018, 2019), embedding SI tasks within subject-specific material so that grammatical form must be processed in order to access disciplinary knowledge. Thus, rather than relying on incidental language gains—a strategy common in many CLIL contexts yet largely ineffective for morphosyntax (see Section 1)—CBSI actively supports grammar acquisition through disciplinary content comprehension. To exemplify, a referential CBSI task for CLIL students learning English alongside the history of architecture might involve the paired questions: What supports the flying buttresses? and What do the flying buttresses support? Both items contain the same vocabulary, but learners must process grammatical structure to extract the disciplinary information—the mechanics of Gothic cathedrals. In an affective CBSI task, the same students may be asked to respond to statements about their learning experience, e.g., I learned today what supports the flying buttresses (Yes/No). I learned today what the flying buttresses support (Yes/No). In this way, CBSI channels attention to form through learners’ engagement with content, fostering form–meaning mapping within content-based, goal-oriented tasks.

In a quasi-experimental study, Walenta (2018) found that, in a CLIL context, a CBSI group made significant and lasting gains in their use of complex relative clauses, outperforming a comparison group that received only comprehension-based instruction. Follow-up research (Walenta, 2019) confirmed CBSI’s effectiveness in promoting students’ morphosyntactic development, with the CBSI group outperforming both comprehension-only and CBSI-plus-output-practice groups in both short- and long-term outcomes. These findings underscore the decisive role of CBSI in supporting successful input processing and position it as a promising approach to integrating linguistic and content development in CLIL.

4. Motivation and Research Questions

Building on the above, this paper aims to extend the scope of CBSI to EMI settings by arguing that CBSI offers a low-intervention, pedagogically grounded approach to supporting grammar development in EMI.

While the effectiveness of SI is well documented (Benati & Lee, 2010; Henry et al., 2024; Lee & Benati, 2009; VanPatten, 2015b), and CBSI has shown promise in CLIL classrooms (Walenta, 2018, 2019), to the best of my knowledge, no published research has yet examined its applicability to EMI. This study addresses that gap by testing whether CBSI—delivered in an asynchronous, computer-assisted format, and thus acknowledging the time constraints of content instructors—can enhance the acquisition of English verb patterns, which are often challenging for learners due to L1 transfer, collocational misuse, and semantic ambiguity (Kaleta, 2020; Kang, 2009; Liang & Dong, 2022).

To evaluate its effectiveness, CBSI—consisting of referential-only tasks—is compared to input flood (IF), a widely used, unobtrusive input enhancement technique that increases the frequency of target forms in the input without requiring learners to attend to them explicitly (Loewen & Inceoglu, 2016; Trahey & White, 1993), and to unmodified disciplinary input (Control). Referential (rather than affective or combined) activities were selected for CBSI because, as research shows (Zhong & Benati, 2024), they are sufficient to support successful processing of target form–meaning connections. They also lend themselves more readily to asynchronous, computer-assisted delivery, as—unlike affective tasks—they do not require personalized interaction or individualized feedback.

The study addresses the following research questions:

Q1: Do EMI students in the CBSI group improve their use of English verb patterns in timed, controlled production tasks in both the short and long term?

Q2: Do EMI students in the CBSI group perform comparably to those in the input flood (IF) and Control groups?

5. Materials and Methods

5.1. Overall Procedure

An experimental study was conducted to compare the effects of three different types of asynchronous pedagogical intervention—content-based structured input (CBSI), input flood (IF), and no additional enhancement (Control)—on the acquisition of English verb patterns from the same disciplinary input. The independent variables were instructional condition (CBSI, IF, Control) and time (pre-test, post-test 1, post-test 2). The dependent variable was performance on timed controlled production (TCP) tasks.

A split-block design was used, with testing and instruction spread across three phases: a pre-test (week 1), instructional treatment—one day (week 2), and two post-tests—immediate (week 3) and delayed (week 9). TCP tasks were administered at each test point to assess participants’ ability to produce the target forms in semi-spontaneous discourse.

Before the study, all participants signed a consent form. Students scoring above 60% on the pre-test (week 1) were excluded, resulting in a final sample of 109 participants. These students were then randomly assigned to one of the three groups.

Table 1. Overview of the study.

Week	Procedure
week 1	consent forms signed by all participants
	pre-test
	randomization following exclusion criteria (>60% test performance)
week 2	instruction (CBSI, IF, and Control)—1 day (1 × 60 min + 1 × 45 min)
week 3	post-test 1
week 9	post-test 2

provides an overview of the study.

5.2. Participants

The study was carried out within an EMI Professional Practice for Architects course at a Polish university. All first-year students enrolled in the course were invited to participate. Initially, 120 students were randomly assigned to three equal groups of 40 (CBSI, IF, and Control). Due to subsequent exclusion criteria (see Section 5.1), the final sample consisted of 109 participants: CBSI (n = 40), IF (n = 37), and Control (n = 32). Participants were between 20 and 24 years old; 7 were Ukrainian, 5 were Belarusian, and the remaining 97 were Polish. All were L2 English users with B2-level proficiency, as defined by the Common European Framework of Reference for Languages (Council of Europe, 2020; CEFR)—the threshold required for course admission.

5.3. Target Feature

The linguistic target of the intervention was English verb patterns with try, remember, and forget, each followed by either a to-infinitive or -ing form with a change in meaning. These structures were chosen because they are often misunderstood due to L1 transfer, unclear collocational patterns, and the fact that small formal changes result in meaning shifts (Kaleta, 2020; Kang, 2009; Liang & Dong, 2022). Learners tend to overlook these grammatical cues and rely on lexical content alone—consistent with the Primacy of Meaning Principle (VanPatten, 2004). For instance, the contrast between Don’t forget to reinforce the floor slab and Don’t forget reinforcing the floor slab is often missed, leading to interpretation and production errors. In disciplinary settings, such misinterpretations may result in erroneous—and potentially serious—misunderstandings, especially when grammatical form signals key disciplinary distinctions.

5.4. Instructional Material

All materials were based on the same disciplinary text provided by the course instructor. Using this text, the linguist (current author) developed three versions of a 105 min asynchronous Moodle-based training, delivered in two sessions (60 + 45 min). The content instructor reviewed the materials to ensure that the disciplinary content was accurate and appropriately conveyed but was not otherwise involved in their delivery. This arrangement can be seen as an instance of individual-level collaboration between the content and language specialists (see Section 1), with the language specialist providing assignment support based on input from the content expert. The disciplinary text used in the instruction was scheduled for follow-up discussion in class (outside the scope of the study).

5.4.1. Content-Based Structured Input (CBSI) Group

The CBSI treatment consisted of 42 target tokens in total—21 -ing and 21 to-infinitive forms of the three verbs (try, remember, forget)—evenly spread across seven referential activities. For each activity, participants read a short disciplinary text and completed CBSI exercises embedded within it by selecting the correct option (a or b). If they wished, they could also listen to both the texts and the a/b options (in line with SI design guidelines; Benati, 2019). Figure 1 shows a CBSI activity fragment. To ease comparison for the reader, the same text fragment is later used to exemplify the IF and Control conditions. The full version of this CBSI activity is provided in Appendix A.

All tasks were referential, requiring learners to select the correct interpretation in English. This task type was chosen because it effectively engages the target processing mechanisms (see Section 2) and is more feasible for asynchronous delivery than affective tasks, which require personalized interaction. Immediate feedback was provided after each choice, indicating whether it was correct or incorrect. The goal was to draw learners’ attention to grammatical form in order to successfully process disciplinary meaning, without overt metalinguistic explanation, in line with the principles of PI. At no point were learners asked to produce the target structures.

CBSI tasks were delivered asynchronously via Moodle. Neither the content instructor nor the researcher (current author) was involved in their delivery. Student engagement was monitored by the researcher using Moodle log data. Students were also explicitly instructed not to use any external resources throughout the duration of the study.

5.4.2. Input Flood (IF) Group

Participants in the IF group worked with the same disciplinary text, divided into seven activities, each modified to increase the frequency of the target verb patterns, using the same number and distribution as in the CBSI condition. No SI tasks were included. For each activity, after reading the full text, learners answered comprehension questions. They were then informed whether their answers were correct. The same delivery method and instructions—not to use external resources during the study—were maintained. Figure 2 shows the same text fragment presented in Figure 1 but now paired with the IF intervention. The full version of this IF activity is provided in Appendix B.

5.4.3. Control Group

The Control group received the original disciplinary text, divided into seven activities, with no grammatical enhancement, and answered the same comprehension questions as the IF group. This condition mirrored a standard EMI scenario in which learners engage with content input without any form-focused enhancement. The same delivery method and instructions were preserved. Figure 3 shows the same text fragment as in Figure 1 and Figure 2, this time paired with no intervention (Control). The full version of this Control activity is provided in Appendix C.

6. Assessment

Timed Controlled Production (TCP) tasks were used to assess participants’ ability to produce the target verb patterns under time constraints, with a view to minimizing conscious rule monitoring and thus encouraging more automatized responses (Ellis, 2005, 2009). This design aimed to tap into whether learners had successfully internalized the target form–meaning mappings and could access them for quasi-spontaneous production (Benati, 2023). Beyond methodological considerations, discourse-level production tasks were also selected for their relevance to the EMI context, where productive skills often lag behind receptive ones (see Section 1).

Three versions (A, B, and C) of the TCP task were developed for use at the pre-test, post-test 1, and post-test 2, following a split-block design. For each version, learners first read a short text that included 12 tested items (3 distractors and 9 target items, with try, remember, and forget each appearing three times in both the to-infinitive and -ing forms). Then, without referring back to the text, learners had seven minutes to reconstruct it in writing using a series of visual and verbal prompts. The prompts consisted of italicized phrases presented in a fixed order and accompanied by bespoke, AI-generated images (created using ChatGPT-4o) illustrating their meaning. Students were required to use each phrase as given, combining it with a verb of their choice and making any necessary adjustments to recreate the intended meaning of the story. An example was provided at the start of the task.

The tasks were administered in pen-and-paper format. For practical reasons, the initial reading text was collected after two minutes to eliminate reliance on it during reconstruction.

Only the target items were scored. Each correct instance earned two points. In a few cases, one point was awarded if the correct verb pattern was used but other errors occurred in the target form, as long as the intended meaning remained clear (e.g., He tryed enquiring to his brother).

See Appendix D for a fragment of Version A of the TCP task.

7. Results

7.1. Data Analysis

A 3 × 3 mixed-design ANOVA was conducted to assess the presence of any significant effects. The between-subjects factor was the type of instruction (CBSI, IF, or Control). The within-subjects factor was Time, representing performance on TCP tasks at pre-test, post-test 1, and post-test 2. As the mixed-design ANOVA failed the assumption of sphericity (Mauchly’s Test, p = 0.014), a Greenhouse–Geisser correction was applied.

Statistically significant effects were found for Instruction (F(2, 97) = 7.946, p < 0.001; η²_p = 0.141), for Time (F(1.84, 178.81) = 100.59, p < 0.001; η²_p = 0.509), and for the Time*Instruction interaction (F(3.69, 178.81) = 9.50, p < 0.001, η²_p = 0.164), all representing large effect sizes.

Given the significant interaction, follow-up analyses were conducted to address the research questions directly (see Section 7.2). All relevant assumptions were tested. When assumptions of normality or homogeneity of variance were violated, data transformation was attempted. If this proved ineffective, post hoc comparisons were conducted using 5000-sample BCa bootstrapping with a 95% confidence interval (Field, 2024). This method is robust to violations of normality and homogeneity of variance and less sensitive to outliers. In all cases, bootstrap results were consistent with the original analyses, confirming the reliability of the findings.

Effect sizes for partial eta squared (η²_p) were interpreted according to Cohen’s (1988) guidelines (≥0.01 = small, ≥0.06 = medium, ≥0.14 = large). Eta squared (η²) values were interpreted using Plonsky and Oswald’s (2014) benchmarks for applied linguistics (≥0.06 = small, ≥0.16 = medium, ≥0.36 = large). In cases of missing data, pairwise case deletion was applied, as recommended for quantitative analysis (Field, 2024). All analyses were performed using SPSS (version 29).

7.2. Timed-Controlled Production Data

Descriptive statistics for the TCP task for each group across time are presented in

Table 2. Group means and standard deviations for the TCP task across time.

	Pretest		Posttest 1		Posttest 2
Groups	M	SD	M	SD	M	SD
CBSI (n = 40)	9.20	3.41	16.35	1.88	14.45	1.84
IF (n = 37)	9.27	3.50	12.38	3.26	12.81	2.38
C (n = 23)	9.52	3.76	12.52	3.69	12.26	3.15

and are plotted in Figure 4. These data show no group differences at baseline. All groups appear to have improved over time, but the CBSI group demonstrated the most substantial learning, particularly at post-test 1, where it recorded the highest scores.

To address Q1—whether EMI students in the CBSI group improved their use of English verb patterns in timed, controlled production tasks in both the short and long term—a repeated-measures ANOVA was conducted on the CBSI group’s scores across all three time points. As the assumption of sphericity was violated (Mauchly’s Test, p < 0.001), a Greenhouse–Geisser correction was applied.

The ANOVA revealed a significant effect with a very large effect size (F(1.29, 50.22) = 161.24, p < 0.001, η²_p = 0.805). Pairwise comparisons with Bonferroni correction showed that post-test 1 scores were significantly higher than pre-test scores (p < 0.001). Although some forgetting occurred between post-test 1 and post-test 2 (p < 0.001), performance at post-test 2 remained significantly higher than at pre-test (p < 0.001). These findings suggest that CBSI effectively supported the learning of English verb patterns in EMI, with gains largely retained over time.

To address Q2—whether CBSI students outperformed their peers in the IF and Control groups—three separate one-way ANOVAs were conducted for each time point.

At pre-test, the ANOVA was not significant (F(2, 104) = 0.160, p = 0.852, η² = 0.003), with a very small effect size, indicating no initial differences among groups. At post-test 1, the ANOVA was significant (F(2, 104) = 22.226, p < 0.001, η² = 0.304), reflecting a medium-to-large effect size. Games–Howell post hoc tests (validated through bootstrapping; see Section 7.1) revealed that CBSI outperformed both the IF and Control groups (p < 0.001), while no significant difference was found between IF and Control (p = 0.929).

At post-test 2, the ANOVA was also significant (F(2, 103) = 8.140, p < 0.001, η² = 0.136), with a small-to-medium effect size. Again, Games–Howell post hoc tests showed that CBSI outperformed both IF (p = 0.001) and Control (p = 0.007), with no significant difference between IF and Control (p = 0.898). In sum, CBSI led to significantly greater gains than both comparison groups at both post-test stages.

7.3. Summary of Results

The present study addressed two research questions:

Q1: Do EMI students in the CBSI group improve their use of English verb patterns in timed, controlled production tasks in both the short and long term?

Q2: Do EMI students in the CBSI group perform comparably to those in the Input Flood (IF) and Control groups?

With regard to Q1, results from the repeated-measures ANOVA showed that the CBSI group made statistically significant gains from pre-test to post-test 1. While some decline was observed between post-test 1 and post-test 2, performance at post-test 2 remained significantly higher than at pre-test. These findings suggest that CBSI led to meaningful and largely sustained improvement in the semi-spontaneous productive use of English verb patterns, supporting a positive answer to Q1.

In relation to Q2, between-group comparisons revealed that CBSI participants significantly outperformed both the IF and Control groups at both post-test stages. No statistically significant differences were found between the IF and Control groups at any time point. These results support a positive conclusion for Q2.

Taken together, the results confirm that referential CBSI activities were the most effective instructional method for learning English verb patterns in this EMI setting. They also show that input flooding is no more effective than the Control condition (no intervention) for facilitating grammar learning in this context—despite the IF group being exposed to the same number of target tokens as in the CBSI condition. This confirms that frequency alone is insufficient for grammar development without structured opportunities to process form–meaning connections.

8. Discussion

This study set out to investigate whether CBSI, delivered asynchronously and without synchronous teacher intervention, could support EMI students’ acquisition of English verb patterns, as reflected in quasi-spontaneous timed, controlled production. The findings support its effectiveness. Participants in the CBSI group not only outperformed their peers in both the IF and Control groups but also largely retained their gains over a six-week period. This confirms that, within an EMI context, CBSI creates favorable processing conditions for mapping form onto disciplinary meaning and, as a result, leads to sustained grammar gains accessible in production.

As such, these results extend previous findings on the efficacy of SI (Benati & Lee, 2010; Henry et al., 2024; Lee & Benati, 2009; VanPatten, 2015b) and CBSI (Walenta, 2018, 2019) by introducing—to the best of my knowledge—the first evidence of CBSI’s effectiveness in an EMI setting.

While input flooding has at times been shown to raise awareness of form (Trahey & White, 1993), this study also confirms that increased exposure alone is insufficient for grammar learning in EMI. As detailed in Section 5.4.1, Section 5.4.2 and Section 5.4.3, the IF group—despite receiving input saturated with the same frequency and distribution of target verb patterns as the CBSI condition—performed comparably to the Control group, which received unmodified disciplinary input. This aligns with previous research (Del Mar Sánchez-Pérez, 2023, p. 374), which shows that mere exposure in EMI settings does not suffice to support language development beyond vocabulary gains.

8.1. Pedagogical Implications

While CBSI does not claim to address all language-related challenges in EMI, the findings suggest that it offers an unobtrusive and discipline-sensitive way to focus on grammar in contexts where explicit instruction is rarely feasible and disciplinary aims are primary. Moreover, by embedding SI tasks within disciplinary materials, CBSI supports meaningful grammar processing not only without detracting from content goals but also by reinforcing them—since grammatical form is mapped onto disciplinary meaning. In doing so, it aligns with calls for the principled integration of language and content in higher education (Lasagabaster, 2022; Morton, 2020).

While input flooding also aligns with the content focus of EMI, the present findings suggest that it does not promote the depth of processing required for acquisition and is therefore insufficient as a standalone approach.

Moreover, CBSI appears to offer a platform for productive collaboration between disciplinary instructors and language specialists, which—as discussed above—is indispensable for EMI effectiveness. In this study, materials were designed by a linguist based on input from a disciplinary expert, ensuring both linguistic precision and content relevance. Such cooperation may help bridge the persistent divide between language and content expertise in EMI, as highlighted by researchers such as Airey (2012) and (Fürstenberg et al., 2022). As EMI continues to expand, pedagogical models like CBSI may represent a principled way forward— offering a platform for constructive dialogue while respecting both the roles and the integrity of each domain.

In sum, this study confirms that asynchronous, referential CBSI tasks can effectively promote grammar learning in EMI. The way they integrate linguistic and disciplinary concerns may also signal a promising step toward more interdisciplinary and pedagogy-informed EMI instruction.

8.2. Limitations of the Study

Several limitations should be acknowledged in this study. First, it focused exclusively on one linguistic structure—verb patterns—and one specific disciplinary course. While the results are promising, generalization to other grammatical targets or content areas should be made with caution.

Second, the instructional treatment was limited to two short asynchronous sessions totaling 105 min. Although learning gains were observed, further research is needed to determine whether longer or repeated exposure to CBSI yields stronger or more durable effects.

Additionally, the use of an asynchronous, computer-assisted format—while realistic given the time and expertise constraints of EMI—necessarily limited control over how participants engaged with the tasks.

Moreover, although teacher involvement was intentionally minimized, future studies might explore how varying levels of scaffolding or interaction influence learning outcomes.

8.3. Further Research

Future research should explore the applicability of CBSI to a broader range of morphosyntactic targets. It would also be valuable to test CBSI in other EMI disciplinary contexts to assess its flexibility and transferability. Further studies might compare different types of CBSI tasks—e.g., referential vs. affective, or those incorporating output practice—to determine which are most effective and feasible under the time and resource constraints typical of EMI. Longitudinal designs tracking how CBSI affects productive academic skills (e.g., writing or speaking) could also shed more light on its pedagogical value. At the same time, monitoring content gains alongside language development would help ensure that disciplinary learning is not compromised.

In addition, research is needed to explore how CBSI can be scaled and sustained in EMI practice—for example, through the development of ready-to-use templates or training modules that enable disciplinary instructors to adapt CBSI tasks more independently.

Finally, it is important to examine how EMI students and teachers perceive CBSI. Despite its demonstrated effectiveness, understanding their perspectives will be key to implementing the approach successfully in real-world educational settings.