Learning MWEs is considered essential for achieving L2 mastery (Boers et al., 2014; Nation, 2013; Siyanova- Chanturia & Pellicer-Sanchez, 2019; Tavakoli & Uchihara, 2020; Vu et al., 2023; Yu et al., 2025), yet it remains a challenge for L2 learners (Majuddin et al., 2021; Nguyen & Webb, 2017; Peters & Webb, 2018; Pu et al., 2024; Puimège et al., 2021; Puimège & Peters, 2020). In particular, insufficient exposure to the target language in the EFL settings restricts learners’ opportunities to learn MWEs. Therefore, many studies have examined the effectiveness of MWEs intake through various modes of input, and findings have proven positive learning outcomes (Dang et al., 2022a, 2022b; Majuddin et al., 2021; Pu et al., 2024; Puimège et al., 2021; Puimège & Peters, 2019, 2020; Teng, 2019; Webb & Chang, 2022). Studies comparing input modes such as reading and viewing have provided valuable insights into how different learning conditions influence MWE acquisition (Dang et al., 2022a, 2022b; Pu et al., 2024; Webb & Chang, 2022).

Dang et al. (2022a) compared the effects of various modes on MWE intake using unimodal input (e.g., reading, listening) and multimodal input (e.g., reading-while-listening, viewing with/without captions). It was proven that reading-only and viewing brought about higher MWE gains in form recognition compared to reading-while-listening and listening-only. Similarly, Pu et al. (2024) gained a similar result through three input modes in their research on incidental MWE gains. Learners across proficiency levels improved their MWE incidental knowledge, and multimodal input, such as viewing with captions, yielded greater results in receptive and productive MWEs. In addition, higher L2 proficiency was proven to predict more MWE learning gains.

In contrast, Webb and Chang’s study (2022) reported partially different results. Webb and Chang (2022) compared unimodal and bimodal input and demonstrated that the reading-while-listening mode benefited the learning of MWEs more than reading-only. Since their research did not include audio-visual input, further studies are needed to clarify how such input compares across modes. The finding might suggest that the prosodic feature of MWEs might have affected the acquisition of MWEs (Hallin & Sidtis, 2015; Lin, 2018), as MWEs, as a chunk, may be more easily processed while viewing and reading-while-listening. Thus, additional research is necessary to explore which modes are most conducive to MWE learning gains.

The efficacy of multimedia input led to more research on the incidental MWE acquisition through languages onscreen (Majuddin et al., 2021; Puimège & Peters, 2020; Teng, 2019). Puimège and Peters (2020) found that viewing without captions improved MWE knowledge, especially in form recall, achieving approximately 30% gains, which is greater than those observed for meaning recall without captions. However, other studies have reported that captions can facilitate incidental MWE acquisition (Dang et al., 2022a; Majuddin et al., 2021; Pu et al., 2024; Teng, 2019).

Teng (2019) investigated the impact of three captioning modes (i.e., full captioning, keyword captioning, and no captioning) on MWEs multidimensionally through video viewing. Receptive knowledge was measured through a multiple-choice test of form recognition, while productive knowledge was assessed through a form recall task requiring learners to write MWEs. Viewing with full captions produced the greatest learning gains, followed by keyword and no captions. However, form recall showed an insignificant difference among captioning types.

Majuddin et al. (2021) experimented with three different language modes—no captions, normal captions, and captions with MWE-enhanced—in combination with two viewing frequencies (single or twice watching). Their findings showed that participants watching twice in both caption modes performed better in form recall. Under the single-watching condition, the caption group outperformed the MWE-enhanced group. However, in watching twice, the MWE-enhanced captions resulted in better performance compared to full captions. These findings suggest that typographic enhancement may require more frequency and exposure to effectively draw learners’ attention and facilitate noticing.

In summary, incidental MWE acquisition has been investigated across unimodal and multimodal input. Viewing and viewing with captions have proven effective for improving form recognition (Dang et al., 2022a; Pu et al., 2024). However, findings regarding language modes on screen and their effects on form recall remain inconsistent. Few studies have simultaneously addressed incidental MWE acquisition in both form recall and recognition across different language modes. The purpose of this study is to investigate whether on-screen language modes while viewing impact incidental MWE learning in both form recall and recognition among Korean-speaking novice English learners.

MWEs are combinations of words that commonly appear as fixed ones (Siyanova-Chanturia & Pellicer-Sánchez, 2019; Webb et al., 2013). A considerable portion of MWEs in L2 poses challenges for L2 learners in effectively conveying communication. MWEs include various lexical types, such as collocations (e.g., “have a thing for,” “drink a toast”), verb phrases (e.g., “look up”), binomials (e.g., “ladies and gentlemen”), proverbs (e.g., “the early bird catches a worm”), and lexical bundles (e.g., “by the way”) (Siyanova-Chanturia & Sidtis, 2019).

L2 learners find it challenging to learn Multi-Word expressions due to the incongruent vocabulary between L1 and L2. Not only do their figurative features provide a challenge to L2 learners, but their low frequency increases the learning burden (Macis et al., 2021). Therefore, this study chose the target MWEs that occur frequently together and have high semantic transparency (Webb et al., 2013). Mutual information (i.e., a measure of collocational strength) was employed to ensure ecological validity. Moreover, non-target MWEs were included to encourage participant engagement and minimize potential floor effect.

The acquisition of MWEs has attracted attention from second language acquisition (SLA) scholars with regard to language processing and output (Dang et al., 2022a; Pu et al., 2024; Van Patten, 2007). Recent research has employed multimodal input to investigate the learning of Multi-Word expressions. Multimedia allows the presentation of nonverbal and verbal information simultaneously, potentially enhancing information processing (Pu et al., 2024). However, multimodal conditions, such as viewing with captions, can also cause additional cognitive load while processing multimodal input (Gass et al., 2019). This study investigated captioning types in multimedia that influence the acquisition of target MWEs.

A plethora of studies have demonstrated the advantages of viewing in vocabulary development (Dang et al., 2022a, 2022b; Majuddin et al., 2021; Montero Pérez & Rodgers, 2019; Montero Pérez et al., 2018; Peters & Webb, 2018; Teng, 2019, 2022). Studies have examined different captioning modes, such as full and keyword captions (Dang et al., 2022a, 2022b; Gass et al., 2019; Majuddin et al., 2021; Montero Pérez et al., 2018; Peters, 2018; Pujadas, 2019; Pujadas & Muñoz, 2020; Puimège & Peters, 2020, Teng, 2019, 2022; Yu et al., 2025). The findings on the effects of different captioning modes on vocabulary learning remain inconsistent. For example, Peters (2018) indicated that intermediate-level learners could benefit more from videos without captions, while learners with limited L2 exposure are suggested to view with captions as an effective supplementary resource (Lindgren & Muñoz, 2012; Suarez & Gesa, 2019). Captioned videos are argued to support comprehension and facilitate vocabulary acquisition (Gass et al., 2019; Montero Pérez et al., 2013; Puimège & Peters, 2020).

Such benefits may also extend to the learning of MWEs. Majuddin et al. (2021) experimented with the three language modes and different viewing times. Two caption groups with double viewing outperformed the no caption group, with MWE-enhanced captions slightly higher but not significantly compared to full captions. However, a single viewing with full captions resulted in better MWE gains than enhanced MWEs. The researchers suggested that due to the limitation of fast-moving scenes, learners do not have enough time to process enhanced MWEs compared to reading. Similarly, Teng (2019) experimented with the learning gains of verb-noun collocations using whole and keyword captions and found that the whole caption group outperformed keyword (MWEs) captions. However, there are some issues in the two studies. Majuddin et al. (2021) investigated the effect of captions on MWE acquisition in form recall (productive knowledge), but Teng (2019) focused on form recognition (receptive knowledge). Limited research has investigated which language modes most effectively facilitate MWE acquisition across both receptive and productive dimensions. To deal with the inconsistency, the present study aims to examine the effects of three distinct on-screen languages on the acquisition of MWEs, targeting both form recall and form recognition.

The current study explores beginners’ incidental acquisition of MWEs through watching a TV program. A few studies have examined the incidental learning of MWEs with intermediate-level learners in various language modes (Majuddin et al., 2021), but it’s uncertain whether audio-visual input would influence beginners’ learning gains of MWEs. Furthermore, this study aims to delve into the difference in form recognition and recall acquisition of MWEs. The research questions are as follows:

1) Does viewing a video in different captioning modes affect the incidental recognition of MWEs’ form?

2) Does viewing a video in different captioning modes affect the incidental recall of MWEs’ form?

This research included 55 Korean-speaking adult EFL learners (33 males and 22 females). Five participants who did not complete all the sessions were excluded, resulting in 50 participants (No Captions = 16, Whole Captions = 18, MWE-Enhanced Captions = 16). The participants were undergraduates majoring in various disciplines, including Marine Biotechnology, Information & Communications Engineering, and Sports Science, and their ages ranged from 19 to 26. Three intact classes were assigned to one of the language modes: no caption, whole caption, and MWE-Enhanced Captions. Before the experiment, all the participants completed the Cambridge English Language Assessment test to determine their proficiency level. Table 1 presents the mean scores and standard deviations of their proficiency test results: No Caption (M = 8.38, SD = 2.65), Whole Caption (M = 8.50, SD = 2.12), and MWE-Enhanced Caption (M = 9.88, SD = 2.63).

All participants scored between 5 to 14 out of 25 on the Cambridge English Language Assessment, corresponding to the A2 (beginner) level of CEFR, with one participant at B1. A one-way ANOVA confirmed that the three groups were statistically almost homogeneous in their proficiency level, F(2, 47) = 1.85, p = .16. Prior to the experiment, participants completed pretests on the MWEs form recall and recognition. As shown in Table 3, all groups scored low on the productive form recall pretest, indicating limited prior MWEs knowledge. From the pilot study of participants with similar proficiency, the participants’ proficiency level was unsuitable for measuring productive MWE learning through incidental exposure. To address this, an MWE form recognition test was included to assess receptive knowledge. The recognition results were higher than recall scores, suggesting moderate receptive knowledge. The Enhanced Caption group showed the highest mean score (M = 3.06, SD = 2.46) in the form recognition pretest, followed by the Whole Caption group (M = 2.61, SD = 2.03) and the No Caption (M = 2.38, SD = 2.36) (see Table 2). The form recall pretest, however, showed different results: No Caption (M = .47, SD = 1.20); Whole Caption (M = .33, SD = .59); MWE-Enhanced Caption (M = .31, SD = .60) (see Table 3). The findings indicated that while participants’ productive skills were limited, their receptive knowledge provided a suitable basis for examining incidental MWE acquisition through captioned video input.

The American sitcom series How I Met Your Mother (Fryman, 2005) was selected as the input because it provides authentic discourse and rich exposure to MWEs (Montero Pérez & Rodgers, 2019). A 22-minute episode from the first series was chosen to match participants’ proficiency, as half of them knew between 2,000 and 3,000 of the most frequent words, making the episode largely comprehensible and engaging. Lexical analysis indicated 3168 tokens and 736 types in AntConc (Anthony, 2024), while Lexical Tutor Profile showed 3151 tokens and 719 types, with minor differences. Lexical profiling showed that the 3,000 most frequent words covered 87.5% (AntConc) and 94.8% (Lexical Tutor) of the episode, while even a 1,000-word base covered 91.5%. These results suggest that the episode’s vocabulary was appropriate for participants and suitable for supporting incidental MWE learning.

Watching English programs exposes learners to a large number of MWEs. Previous studies on MWEs have handled specific collocations, such as ‘adjective + noun,’ ‘verb phrases,’ and ‘prepositional phrases,’ often by modifying texts to include sufficient target items. However, such materials lack ecological and authentic value (Majuddin et al., 2021; Peters et al., 2019; Puimège & Peters, 2019) The present study aimed to investigate incidental MWE acquisition from authentic input, providing both language and socio-cultural exposure. Following Majuddin et al. (2021), the study targeted 16 MWEs primarily consisting of phrasal verbs (e.g., “plan out,” “go out”) and idioms (e.g., “a long shot”). The transcript of the input was thoroughly analyzed to identify target MWEs likely unfamiliar to participants. The selected items were validated using online dictionaries, the Corpus of Contemporary American English, and Lexical Tutor. All MWEs had over 100 hits in COCA and mutual information (MI) values above 3, except ‘bear with (MI = 0.35),’ which was retained for ecological validity. A total of 27 MWEs were presented to participants to maintain attention and avoid floor effects (i.e., a condition where numerous scores cluster at the possible lowest point of measurement). After a pilot test with students of similar proficiency, 16 target MWEs were finalized for form recall and recognition tests. Item frequency in the input varied, with one MWE occurring three times, five occurring twice, and the remainder occurring once (see Table 4).

For the Whole Caption group, the captions appeared at the bottom of the running time. 16 target MWEs, including 11 nontarget MWEs, were enhanced with bold, underlined, and a different color from other expressions.

MWE recognition was assessed at pretest, immediate posttest (3-week interval), and delayed posttest (one month later) to examine both learning gains and retention (see Table 5). Pretest scores were low and comparable across groups: No Caption (M = 2.37, SD = 2.36), Whole Caption (M = 2.61, SD = 2.03), and MWE-Enhanced Caption (M = 3.06, SD = 2.46) (Table 5). After viewing, all groups improved, with the MWE-Enhanced Caption group showing the highest immediate posttest scores (M = 5.50, SD = 3.20), followed by Whole caption (M = 5.39, SD = 2.28), and No Caption (M = 4.19, SD = 3.31). Delayed posttest scores declined slightly but remained above pretest scores, with MWE-Enhanced Caption still retaining the highest mean (M = 4.44. SD = 3.33).

Figure 4 presents the estimated marginal means of the MWE form recognition between groups in pretest, immediate post-test, and delayed post-test. All the groups performed better than pretests. The Enhanced Caption outperformed the other groups after the intervention, and in the delayed posttest, followed by the Whole Caption.

A repeated-measures ANOVA showed a significant main effect of Time, F(2,94) = 17.23, p < .001, indicating improvement over time. The Time x Group interaction was not significant, F(4,94) = .35, p = .85, and between-group differences were non-significant F(2,47) = .60, p = .55. These findings suggested that while all participants benefited from audio-visual input, the differences between caption modes were not statistically significant.

MWE form recall was measured using pretests, immediate posttests, and delayed posttests through viewing under different on-screen language modes. The descriptive analysis presented in Table 6 shows that all groups began with similar levels of MWE form recall knowledge at pretest. Mean scores were comparably low across groups: No Caption (M = .47, SD = 2.36), Whole Caption (M = .33, SD = 2.03), and the MWE-Enhanced Caption (M = .31, SD = 2.46). Following the viewing, all groups showed slight improvement on the immediate posttests. The MWE-Enhanced Caption group showed the highest improvement (M = 1.00, SD = 1.52), followed by No Caption (M = .81, SD = 1.55). and Whole Caption (M = .50, SD = .84). Delayed posttest scores indicated slight retention across different captioning types, with No Caption group (M = .81, SD = 1.83) outperforming the caption groups (Whole Caption: M = .42, SD = .62; Enhanced Caption: M = .69, SD = .99). Overall, incidental learning of MWE form recall was limited in Form Recall gains, suggesting that although Enhanced Captions may facilitate immediate learning of MWEs, the retention over time appears inconsistent.

Figure 5 presents the estimated marginal means of the MWE form recall between groups in pretest, immediate post-test, and delayed post-test. All the groups scored low on the pretest. The Enhanced Caption outperformed the other groups after the intervention, but No Caption retained the most MWEs knowledge on form recall in the delayed posttest, which is unexpected. The difference between captions and no captions was not statistically significant.

A repeated-measures ANOVA revealed a significant effect of Time on MWE recall, F(2,94) = 5.56, p = .005, indicating that participants’ recall scores improved over time. Consistent with the MWE recognition findings, participants’ MWE recall improved over time due to audiovisual input, but differences between caption conditions were not statistically significant. These results indicate that varying on-screen language modes did not influence MWE learning for beginner-level learners.