The Production Effect in Second Language Verb–Noun Idiom Learning: Evidence From Pure-List Designs and Transfer-Appropriate Processing

Abstract

Learners of a second language often vocalize newly learned items, and previous studies have shown that this practice facilitates the learning of single words and multiword expressions. However, these effects have been demonstrated only in mixed-list designs, where a single list of items is learned using multiple learning methods, and have not been tested in pure-list designs, where each list is learned with one consistent method. Moreover, prior research has relied almost exclusively on written recognition tests, leaving spoken-modality tests and recall measures largely unexamined. This study employed a within-participant, pure-list design in which each list was associated with a single learning condition, and each participant studied a total of four lists. The participants were 68 Japanese learners of English. Participants learned 64 verb–noun idioms assigned across four active learning conditions (Listening, Repeating, Reading while Listening, and Reading while Repeating), with 16 items in a fifth condition serving as a no-exposure Control condition. Learning outcomes were assessed with three test types (form recall, form recognition, and meaning recognition) administered in two modalities (written vs. spoken). In the written test modality, form-recognition performance was higher for the Repeating and Reading-while-Repeating conditions, whereas in the spoken test modality, the strongest form-recall gains were produced for the Reading-while-Repeating condition. These results highlight modality-specific benefits of vocalization and are discussed in relation to the strong-trace account and the transfer-appropriate processing framework.

Keywords

multiword expressions production effect transfer appropriate processing verb–noun idioms vocalization

1. Introduction

Multiword expressions (MWEs), which consist of multiple words, such as collocations, idioms, and phrasal verbs, are central to second language (L2) learning because they are pervasive in native discourse (Conklin & Schmitt, 2012) and associated with higher L2 proficiency (Rafieyan, 2018; Wray & Perkins, 2000). However, empirical guidance on how to teach MWEs effectively remains limited, suggesting the need to explore pedagogical approaches that promote their learning. Although idioms, collocations, and related multiword expressions are theoretically distinct and differ in properties such as transparency and learning mechanisms (e.g. Cieślicka, 2006; Steinel et al., 2007), the present study focuses on verb–noun idioms. This is because such expressions are often semantically opaque, making them difficult to learn and therefore in need of instruction (Cieślicka, 2006). In addition, their semantic opacity allows learning outcomes to be assessed in terms of both form and meaning. Focusing on verb–noun expressions also makes it possible to control for phrase type and part-of-speech structure. In the present study, the term ‘verb–noun idioms’ is used operationally to refer to combinations of a verb and a noun that vary in transparency and may include items located at different points on Howarth’s (1998) collocational continuum, including restricted collocations, figurative idioms, and pure idioms. This terminological choice is made for clarity rather than to claim theoretical equivalence, and readers are encouraged to bear this in mind when interpreting the findings.

One candidate for facilitating the learning of verb–noun idioms is vocalization, or saying target items aloud during learning. A growing body of evidence has demonstrated positive effects of vocalization on both single-word learning (Icht & Mama, 2022) and verb–noun idiom learning (Hashizaki, 2024a).¹ However, these studies have examined vocalization only under mixed-list designs, in which items learned with vocalization and without vocalization appear within the same list. Thus, it remains unclear whether vocalization is effective in pure-list designs, where all items are studied under a single learning condition rather than a mixture of conditions.

Furthermore, previous studies on the production effect in learning L2 word and verb–noun idioms (Hashizaki, 2024a; Icht & Mama, 2022) have assessed learning solely through written, receptive tests, without examining productive knowledge or spoken performance. According to the Transfer Appropriate Processing (TAP) framework (Morris et al., 1977), learning is optimized when the processing operations during learning and testing align. Therefore, vocalization may yield stronger benefits when learning is evaluated through spoken productive tests, which require similar processing to the vocalization activity itself.

Additionally, previous studies (Hashizaki, 2024a; Icht & Mama, 2022) have examined the production effect in L2 word and verb–noun idiom learning under conditions in which written forms were presented. According to Paivio (2007), the simultaneous presentation of visual and auditory information facilitates memory because the two sources of information can be integrated during processing. This suggests that, independently of the production effect itself, the simultaneous presentation of written and auditory input may provide an additional benefit. Consequently, producing items aloud when written forms are available may lead to greater learning gains than producing them aloud without written support. However, it remains unclear whether the presence of written forms provides an additional benefit beyond the production effect itself.

To address the research gaps identified above – namely, the exclusive use of mixed-list designs, the reliance on written receptive tests, and the unclear role of written forms in the production effect – the current study investigated the effect of vocalization on L2 verb–noun idiom learning from the perspectives of both the production effect and TAP. Specifically, the current study used a pure-list design and assessed learning outcomes with three test types (form recall, form recognition, and meaning recognition) administered in two modalities (written and spoken). Furthermore, it manipulated the presence of written support during learning to assess its specific impact.

Clarifying these issues could provide concrete pedagogical guidance on the use of vocalization in verb–noun idiom instruction, thereby offering both theoretical and practical value.

2. Literature Review

Idioms are among the more semantically opaque types of MWEs and are therefore particularly difficult for L2 learners to acquire (Cieślicka, 2006), making it important to identify effective ways to support their learning. Recent studies (Hashizaki, 2024a; Icht & Mama, 2022) have suggested that vocalization may facilitate the learning of both single words and verb–noun idioms in an L2. However, the extent of this benefit remains unclear because previous findings may depend on methodological factors such as experimental design, the modality and format of the outcome measures, and the influence of simultaneous presentation of text and speech. To clarify these issues, the present literature review draws on two relevant theoretical frameworks: the production effect (MacLeod et al., 2010), which concerns the memorial advantage of overt production, and transfer appropriate processing (Morris et al., 1977), which emphasizes the match between learning and testing processes.

2.1. Definition of Verb–Noun Idioms and Rationale for Focusing on Them

As noted in the introduction, verb–noun idioms are defined in the present study as combinations of a verb and a noun that vary in transparency and may include items located at different points on Howarth’s (1998) collocational continuum, including restricted collocations, figurative idioms, and pure idioms. In Howarth’s phraseological model, this subcategorization is ‘not a simple two-way division but a continuum, derived from such criteria as restricted collocability, semantic specialization, and idiomaticity’ (p. 28). Within this continuum, restricted collocations are expressions in which one component is ‘used in a specialized, often figurative sense only found in the context of a limited number of collocates’ (p. 28). Figurative idioms, by contrast, ‘have metaphorical meanings in terms of the whole and have a current literal interpretation’ (p. 28), whereas pure idioms ‘have a unitary meaning that cannot be derived from the meanings of the components’ (p. 28) and are ‘the most opaque and fixed category’ (p. 28). This framework is useful for the present study because it allows the target items to be treated as a coherent set of verb–noun idioms while also acknowledging variation in transparency across items. Among these categories, the target items in the present study most closely align with figurative idioms overall, although individual items vary in transparency, and some are closer to restricted collocations or pure idioms. This is because many of the expressions have conventionalized phrase-level meanings that are partly metaphorical while still allowing varying degrees of literal interpretability.

The present study focuses on verb–noun idioms for both methodological and pedagogical reasons. Methodologically, idioms are often semantically opaque, which makes it possible to assess learning outcomes in terms of both form and meaning. In addition, restricting the target items to verb–noun expressions allows phrase type and part of speech to be controlled. Pedagogically, idioms are an important target of instruction. In first language (L1) use, they are prevalent in everyday language and are often understood with relative ease despite their semantic opacity (Cieślicka, 2006). Research on L1 processing further suggests that familiar idiomatic expressions can be processed more efficiently than novel strings (Siyanova-Chanturia et al., 2011). Moreover, because idiom learning may contribute to more native-like and fluent L2 performance (Rafatbakhsh & Ahmadi, 2019), developing knowledge of such expressions is also important for L2 learners. However, the semantic opacity of idioms often makes them difficult for L2 learners to learn (Cieślicka, 2006). Indeed, semantically opaque items have been shown to be more difficult to learn than more transparent idioms (Cucchiarini et al., 2022; Steinel et al., 2007). These difficulties underscore the need for effective instructional approaches that can support the efficient learning of verb–noun idioms. Yet, it remains unclear which learning methods are most effective for facilitating their acquisition.

2.2. Production Effect

Since verb–noun idioms are often challenging for L2 learners to acquire, it is crucial to explore instructional techniques that facilitate their learning. One promising approach is vocalization: the act of saying target items aloud during study. This technique is grounded in the production effect, a phenomenon where items spoken aloud are remembered more effectively than those read silently (Gathercole & Conway, 1988; Hopkins & Edwards, 1972; MacLeod et al., 2010). Three accounts have been proposed to explain the advantage of vocalization. The distinctiveness account argues that spoken items stand out against silently processed items within a learning context, thereby enhancing memory (Hunt, 2006; MacLeod et al., 2010). Icht and Mama (2022) further suggested that vocalization draws greater attentional resources because it engages additional encoding modalities. Specifically, vocal production involves ‘three distinct encoding processes’ – visual, auditory, and motor – whereas silent study typically relies on fewer encoding channels, which may make produced items more distinctive and memorable.

The strong-trace account, by contrast, posits that speaking draws greater attentional resources to the target items, resulting in stronger memory traces (Fawcett, 2013; Ozubko et al., 2012). However, this account does not provide a detailed explanation for why vocal production necessarily attracts more attention, particularly because the distinctiveness-based explanation may not fully apply in contexts where no silent baseline (e.g. silent reading) is included for comparison. One possible explanation is that vocalization forces learners to process each word in the input. Gao et al. (2017) argued that overt production requires learners to process all words included in a text or utterance. In contrast, silent reading often involves selective processing of the input. Eye-movement research indicates that readers frequently bypass a large proportion of words – particularly function words – and even when words are fixated, the duration of fixation tends to be brief (Rayner, 2009).

These two accounts are not necessarily incompatible. Instead, a third explanation – the dual-process account – proposes that the production effect reflects the combined contribution of both increased distinctiveness and stronger encoding (Fawcett & Ozubko, 2016).

Considering the three theoretical accounts is important for understanding variation across experimental designs. The magnitude of the production effect varies depending on the experimental design, with more robust effects observed in mixed-list designs than in pure-list designs. In mixed-list designs, vocalized and silent items are intermixed, so produced items may benefit both from stronger encoding and from increased distinctiveness relative to surrounding silent items (Fawcett & Ozubko, 2016). In contrast, pure-list designs involve conditions in which all items are vocalized or all items are learned silently. Because no contrasting items are present within the same learning context, the distinctiveness account is unlikely to operate, and any observed advantage is more likely to reflect the strong-trace account alone. As a result, the effect tends to be less robust, and individual studies may therefore fail to reach statistical significance (Fawcett, 2013; Fawcett & Ozubko, 2016; Saint-Aubin et al., 2021).

Vocalization facilitates both single-word and verb–noun idiom learning in L2. Among a variety of studies (Alali & Schmitt, 2012; Durrant & Schmitt, 2010; Ellis & Sinclair, 1996; Hashizaki, 2021, 2024b; Seibert, 1927), Icht and Mama (2022) and Hashizaki (2024a) are ones that investigated the effect of vocalization in the paradigm of production effect. Icht and Mama (2022) systematically examined the production effect in L2 vocabulary learning. Seventy-five Hebrew-speaking undergraduates learned Esperanto nouns in a mixed-list design, encountering vocalized and silently studied items intermixed. Learners viewed L1–L2 pairs with auditory input, and in the vocalization condition repeated the L2 word once. Memory was assessed using visual meaning-recognition tests at three intervals (immediate, 1 week, 2 weeks). Across all delays, vocalized items were recalled more accurately than silently studied ones, demonstrating both immediate and durable production benefits. Hashizaki (2024a) extended the production effect to multiword expressions. Thirty-two Japanese undergraduates learned 135 English verb–noun collocations in a mixed-list design, vocalizing or listening to each item during exposure. The study used visual recognition tests for both form and meaning, administered immediately and after one week. Vocalization yielded significantly higher accuracy in both test types at both intervals, indicating that the production effect also facilitated form–meaning learning of L2 verb–noun idioms over time.

Despite these positive findings, key methodological limitations remain. First, all studies used mixed-list designs in which aloud and silent items were intermixed; therefore, the effect has not been tested in pure-list designs. Second, although prior research on L2 MWE learning (e.g. collocations) has employed productive tests or assessments in the spoken modality (e.g. Boers et al., 2006; Estaji & Montazeri, 2026; Ha, 2017; Xu, 2018), the effects of vocalization in L2 verb–noun idiom learning have been examined primarily through written, receptive recognition tests. As a result, it remains unclear whether vocalization benefits extend to productive or spoken performance. In other words, the processing required during learning (overt articulation) and that required during testing (visual recognition) do not fully align, suggesting that current findings may underestimate the potential advantages of vocalization.

2.3. Transfer Appropriate Processing

Transfer appropriate processing (TAP) provides a useful theoretical lens for addressing the mismatch between encoding and retrieval processes that arises between learning and testing, as noted in the previous section. TAP is the principle that memory performance is optimized when the cognitive operations engaged during encoding match those required during retrieval (Morris et al., 1977). Applicability of TAP is emphasized in L2 as well (Lightbown, 2007). For instance, Uchihara (2023) reported a pattern consistent with TAP in L2 vocabulary learning. The study involved Japanese undergraduate learners of English as a foreign language who were divided into two groups. One group took weekly vocabulary quizzes in written form, seeing the target English words in print, whereas the other took quizzes in spoken form, hearing the target words. In both groups, learners retrieved the meanings of the target words and received feedback. Before and after the instructional period, participants completed vocabulary tests assessing their knowledge of the target items in both written and spoken modalities, in which they recalled the meanings of the words. Results showed that learners in the spoken-quiz condition demonstrated greater gains on a spoken vocabulary posttest than those in the written-quiz condition, whereas written vocabulary gains did not differ between groups. Uchihara suggested that the lack of a modality effect for written vocabulary may be because learners in both groups had comparable exposure to the written forms of the target words.

Taken together, these findings suggest that matching quiz modality to retrieval format can enhance L2 vocabulary retention. To fully determine how vocalization supports long-term retention of vocabulary, spoken productive assessments are needed. Such assessments would allow systematic examination of the relationship between learning modality and testing modality, clarifying when vocalization confers its greatest benefits.

2.4. Simultaneous Presentation of Text and Speech

According to Paivio (2007), the simultaneous delivery of visual and verbal stimuli facilitates superior memory retention by enabling the brain to integrate these distinct streams of information during processing. Accordingly, a growing body of research suggests that multimodal input can facilitate L2 verb–noun idiom learning (e.g. Huang et al., 2022; Pu et al., 2026; Wang et al., 2023; Webb & Chang, 2022).

Huang et al. (2022), for example, compared paper-based phrase-learning materials with multimodal materials that presented phrase form, meaning, and use through written text, audio recordings, pictures, and video clips. In their 4-week intervention with Chinese EFL learners, both groups improved, but the multimodal group outperformed the unimodal group on both the midtest and the posttest. More directly, Wang et al. (2023) investigated idiom learning under three CALL glossing conditions: text only, text plus audio, and text plus video. Their results showed that all three conditions promoted learning, but the text-plus-audio and text-plus-video conditions outperformed the text-only condition.

However, these studies did not test whether the presence or absence of written input affects learning under production effect conditions. Thus, it remains unclear whether written input provides an additional benefit in such conditions.

3. The Current Study

Because verb–noun idioms are often semantically opaque, they are difficult for L2 learners to acquire. At the same time, empirical guidance on how to teach them effectively remains limited. Previous studies have shown that vocalization supports L2 lexical learning for both single words and verb–noun idioms, as learners who vocalize exhibit higher accuracy on later recognition tests (Hashizaki, 2024a; Icht & Mama, 2022). However, three methodological limitations remain. First, prior studies (Hashizaki, 2024a; Icht & Mama, 2022) have used mixed-list designs, which differ from instructional settings where a learner typically studies items using one consistent learning method for one list. Thus, whether vocalization benefits persist under pure-list conditions remains unknown. Second, previous research on the effects of vocalization in L2 verb–noun idiom learning (Hashizaki, 2024a; Icht & Mama, 2022) has assessed outcomes primarily through written receptive tests, leaving it unclear whether its effects differ in productive tests or in tests administered in the spoken modality. Third, it remains unclear whether the effect of the simultaneous presentation of text and speech emerges regardless of whether learners engage in vocalization or whether it depends on the presence of vocal production.

This study addressed the following research questions:

Research question 1: Does vocalization enhance the learning of L2 verb–noun idioms under pure-list conditions?

Research question 2: Do the effects of vocalization (vocal vs. silent) and testing modality (written vs. spoken) interact in learning L2 verb–noun idioms?

Research question 3: Does orthographic support during learning enhance the learning of L2 verb–noun idioms independently of vocalization?

4. Methods

4.1. Participants

The participants were 68 Japanese EFL learners (36 females, 32 males; M age = 24.310 years, SD = 4.250). Participants were recruited via an online participant recruitment website (https://www.jikken-baito.com/). Recruitment was restricted to native speakers of Japanese aged between 18 and 35 years who owned a Windows-based personal computer. Participants who were not current students were required to have completed at least a bachelor’s degree. Each participant received 2,400 yen as compensation. Participants were randomly assigned to either the written-modality group (n = 40) or the spoken-modality group (n = 28) at recruitment. The resulting imbalance in group sizes, with more participants in the written-modality group, was due to unequal attrition prior to participation rather than the assignment procedure itself. Vocabulary knowledge was measured with the V_YesNo v1.01 (Meara & Miralpeix, 2016), revealing that participants knew an average of 4,397.338 English words (SD = 1,398.116), consistent with an intermediate proficiency level. When broken down by test modality, the written-modality group had a mean vocabulary size of 4,124.825 words (SD = 1,398.559), whereas the spoken-modality group showed a slightly higher mean of 4,786.643 words (SD = 1,325.952). To control for individual differences in lexical proficiency, vocabulary size was included as a covariate in all statistical models. Ethical approval for this study was obtained from Matsuyama University (Ethics approval number: 松大倫15号).

4.2. Programming Language

Hot Soup Processor (https://hsp.tv/) was employed to create the questionnaires, control the learning procedures, and administer the immediate posttests, as well as to record responses. Scoring and data logging for the form and meaning recognition tests and data logging for the form recall test were automated, and these results were automatically saved as text files on each participant’s computer.

4.3. Materials

4.3.1. Target Items

Eighty verb–noun idioms from Hashizaki and Sugiura (2026) were chosen. To reduce the influence of vocabulary knowledge, idioms were restricted to those composed of words within the first 3,000 entries of the JACET 8,000 list (JACET Committee of Basic Words Revision, 2003), except for rainbows and pants. Because the JACET list is based on both the British National Corpus and materials frequently encountered by Japanese learners (Uemura & Ishikawa, 2004), it is appropriate for this population. The 3,000-word cutoff was also justified by the estimated average vocabulary size of Japanese university students (3,715 word families; McLean et al., 2014). Idioms sharing the same verb were avoided to prevent lexical overlap across items.

Expressions consisting of multiple words, including idioms, are characterized by a range of properties, including semantic transparency, familiarity, and corpus frequency. These properties have been shown to influence L2 learners’ knowledge of such expressions (Hubers et al., 2020), their processing (Siyanova-Chanturia & Sonbul, 2025), and their learning (Cucchiarini et al., 2022; Steinel et al., 2007). However, it is extremely difficult to perfectly control all of these numerous properties across learning conditions. Therefore, in the present study, the program was designed so that the 80 items were randomly assigned for each participant, in order to prevent items with particular properties from clustering in specific learning conditions. This procedure ensured that items with particularly high (or low) values on certain properties did not systematically concentrate in any given learning condition. For the target items and their Japanese translations, see Table S1 in supplemental material.

4.3.2. Tests for Learning Outcomes

4.3.2.1. Theoretical Rationale for the Tests

The test battery followed the framework of Webb et al. (2013), based on Nation and Webb’s (2011) distinction between form knowledge and form–meaning connections, each measurable receptively or productively.

Form Recall (FRc): Learners produced the missing noun of each idiom. This test targeted productive form knowledge and the most demanding level of form-related knowledge.

Form Recognition (FRg): Learners selected the studied idiom from three options. This test measured receptive form knowledge and familiarity with verb–noun idiom forms without requiring production.

Meaning Recognition (MRg): Learners chose the correct Japanese meaning. This test assessed receptive knowledge of form–meaning connections and understanding of the form–meaning link (L2 form → L1 meaning).

Because the meanings of the target items were not vocalized in the present experiment, a meaning-recall test was not administered. Together, these tests captured complementary dimensions of verb–noun idiom learning and aligned with established models of lexical knowledge.

4.3.2.2. Form Recall Test (FRc)

To measure the production knowledge of verb–noun idiom forms, a form recall test was conducted. Each item provided the first word of the verb–noun idioms – a verb, sometimes followed by an article – as a cue. In the written modality, participants saw this cue on-screen and typed the second word (the noun) that they had studied in each learning condition. In the spoken modality, participants clicked the button on the screen, listened to the cue, and said aloud the second word that they had studied in each learning condition. For the control items, participants produced the second word that they thought would complete the verb–noun idiom, either by typing in the written modality or by saying it aloud in the spoken modality. When unable to produce the correct noun, those in the written modality typed ‘4’, while those in the spoken modality said ‘I don’t know.’ For the spoken modality, all recorded responses were transcribed by the author. After transcription, all answers – except those that participants themselves judged as ‘I don’t know’ or those that were clearly incorrect – were double-checked (24.107% of all responses in the spoken recall test) by a highly proficient Japanese L2 English user (C1 level). Two raters showed high agreement, with a percent agreement of 96.111% and Cohen’s κ = .880 (95% CI [.830, .930]). In cases where discrepancies arose between the transcription and the checker’s judgment (3.148% of double-checked responses), the author re-examined the audio and made the final decision. Regarding scoring, in both the written and spoken modalities, errors related to plural forms were treated as correct responses, whereas all other errors were marked as incorrect. In the written test, spelling mistakes were also counted as errors. An example of the test format is provided in Figure 1. For the details of the form recall test, see Table S1 in supplemental material.

Figure 1.

Examples of the form recall test.

4.3.2.3. Form Recognition Test (FRg) and Meaning Recognition Test (MRg)

Two recognition tests based on Hashizaki (2024a) measured learning gains, one targeting form and the other meaning. In the written modality, participants saw three idioms and indicated which one they had studied in each learning condition. For the spoken modality, three options were presented with audio when participants clicked a button on the screen, and they typed the number of the choice. One option was a real idiom, and the remaining two were distractors. These distractors were created by recombining verbs and nouns from the target items and were confirmed to lack idiomatic meanings through two checks: an online search and verification by two native English speakers. For control items, participants chose the one they thought was an idiom from three options.

The meaning test followed the same three-alternative structure, except participants selected the Japanese translation they had learned; for control idioms, they picked the translation they considered most appropriate. For the written modality, a verb–noun idiom from the learning session and three Japanese translation options were presented on the screen. Then, participants chose the correct answer by typing the number for the option. For the spoken modality, three Japanese translation options were presented on the screen, and the participants clicked the button to hear a verb–noun idiom from the learning session. One option was the correct meaning, while the other two were distractors. These distractors were literal interpretations generated by three Japanese learners, including the author. All translations were checked using two online Japanese–English dictionaries (Eijiro on Web and Weblio), and when no suitable translation existed, the author provided one in consultation with a native English speaker.

Both form and meaning tests offered a fourth option, ‘I don’t know’, to capture uncertainty. Furthermore, whenever a participant answered correctly, they were immediately asked whether the answer had been known prior to the learning session, allowing pre-existing knowledge to be taken into account. For the examples of the recognition tests, see Figures 2 and 3; for the details of the form and meaning recognition tests, see Table S2 and Table S3 in supplemental material.

Figure 2.

Examples of the form recognition test.

Figure 3.

Examples of the meaning recognition test.

4.3.3. Vocabulary Size Test

Participants’ English vocabulary knowledge was assessed using the V_YesNo v1.01 test (Meara & Miralpeix, 2016). This instrument was chosen because it provides a rapid estimate of vocabulary size, requiring only about ten minutes to complete. The test was delivered online, and learners indicated whether they recognized each item by selecting ‘Yes’ or by choosing ‘Next’ when they did not know or were unsure about a word. After finishing the task, each participant received a vocabulary size estimate on a 0–10,000 scale, categorized into three bands: beginner (2,000–3,500 words), intermediate (3,500–6,000 words), and proficient (6,000–10,000 words) (Meara & Miralpeix, 2016).

4.4. Procedure

Data collection was conducted from June to October 2025. Participation was fully online, and all participants completed the study remotely via Zoom on their own computers (i.e. no face-to-face sessions were conducted). The study used a single-session design: each participant completed all learning and testing cycles in one day. The total time required per participant – including instructions and setup checks, all learning and test phases, the vocabulary size measure, and the post-task questionnaire/file upload – was approximately 90–120 minutes.

The experimental sequence is summarized in Figure 4. The present study adopted a pure-list design. In a pure-list design, each list of items is learned using a single, consistent learning method, and learning conditions are not intermixed within the same list. This design contrasts with mixed-list designs, in which multiple learning methods are combined within a single list of items. At the same time, the present study implemented this pure-list structure within participants, such that each participant studied multiple lists, with each list assigned to a different learning condition. This approach differs from typical between-participants pure-list designs, in which each participant experiences only one learning condition. By implementing a pure-list design within participants, the present study preserved the methodological clarity of pure-list learning while controlling for individual differences across learning conditions.

Figure 4.

Procedure of the experiment.

The same procedure was applied to participants in both written and spoken modalities, except for the modality of the three tests. Prior to the main session, participants reviewed the written instructions, submitted informed consent through Google Forms, and confirmed that the software ran properly on their computers. During the real-time session using Zoom, the author explained the overall procedure of the experiment. Participants then completed a short background survey and began smartphone audio recording to verify that vocal responses were produced when required.

During the learning phase, randomization was implemented at two levels: the order of the four learning conditions and the order of the items within each condition. Each participant studied 64 target verb–noun idioms, with 16 verb–noun idioms randomly allocated to each condition. Every participant experienced all learning conditions rather than being assigned to a subset. The four learning conditions – listening, repetition, reading while listening (RWL), and reading while repeating (RWR) – are described in detail in Table 1. Figures 5 and 6 show examples of learning sessions.

Table 1.

Explanation of the Four Learning Conditions and the Control Condition.

Learning condition	Visual presentation	Verb–noun idiomaudio presentation	Vocalization
Listening	Japanese translation	Twice	–
Repeating	Japanese translation	Once	Once
Reading while listening (RWL)	Verb–noun idiom + Japanese translation	Twice	–
Reading while repeating (RWR)	Verb–noun idiom + Japanese translation	Once	Once
Control	–	–	–

Notes. Each trial began with a fixation cross (+) presented for 0.5 s. Visual stimuli (Japanese translations alone or verb–noun idioms with their Japanese translations) were displayed for 5 s in all learning conditions except the control condition.

Figure 5.

Learning sessions for listening and repeating.

Figure 6.

Learning sessions for reading while listening (RWL) and reading while repeating (RWR).

After completing the learning of 16 verb–noun idioms under a given condition, participants immediately proceeded to the corresponding tests. To prevent subsequent tasks from influencing performance on earlier ones, the assessments were administered in a fixed sequence: (1) form recall, (2) form recognition, and (3) meaning recognition. For instance, the form recognition test followed the recall task to ensure that the response options provided did not inadvertently cue recall. Likewise, since the meaning recognition test displayed the correct forms alongside meaning options, it was positioned last to avoid biasing the preceding form-based assessments. When testing for one condition had finished, participants advanced to the next set of 16 verb–noun idioms and repeated the same procedure. For the 16 verb–noun idioms assigned to the control condition, participants completed only the test phase.

Following the completion of all learning and assessment cycles, participants took a vocabulary size measure (V_YesNo v1.01). Lastly, they reported demographic information using Google Forms and uploaded all resulting files.

4.5. Analysis

R 4.5.2 (R Core Team, 2025) software and the lme4 package (Bates et al., 2015) were used for all analyses. To avoid excessive model complexity, separate models were constructed for the written modality and the spoken modality, following the same analytical procedure. Generalized linear mixed-effects modeling (GLMM) was used for all accuracy analyses (binomial distribution and logit link function). The five learning conditions, test type, and their interactions were the fixed effects. For the learning conditions, the control condition was included in the analysis because it served as a theoretically important baseline representing no exposure. At the same time, it should be noted that including this condition may have accentuated the overall main effect of learning condition relative to analyses limited to the four learning conditions alone. Vocabulary size was also added as a covariate. The response variables were accuracy in each test.

Categorical variables were coded using sum-to-zero (effects) contrasts, with coefficients interpreted as deviations from the grand mean. The effects package (Fox & Weisberg, 2019) was applied to plot all models, and the Bobyqa optimizer was used to avoid convergence errors. Only random slopes justified by the design (Barr et al., 2013) and contributing to improved model fit were included, to avoid unnecessary loss of statistical power (Matuschek et al., 2017).

Modeling began by fitting the best model that included only random effects. Each fixed effect was then tested as a candidate random slope, and models were fitted for every possible combination and sequence of these slopes. Models were compared using AIC values, and the model with the lowest AIC was selected. When the difference in AIC between models was less than 2, the more parsimonious model was preferred. After the random-effects structure had been finalized, learning condition, test type, and their interaction were entered as fixed effects, with vocabulary size included as a covariate, to construct the final model. All final models were evaluated for assumptions – including overdispersion, normality of the random effects, and multicollinearity – using the DHARMa (Hartig, 2024) and performance (Lüdecke et al., 2021) packages, and no violations were detected.

5. Results

For the written modality, Cronbach’s alpha was .905 for the form recognition test, .894 for the form recall test, and .888 for the meaning recognition test. For the spoken modality, Cronbach’s alpha was .743 for the form recognition test, .870 for the form recall test, and .699 for the meaning recognition test. The alpha value for the spoken meaning recognition test was marginally below the conventional .700 cutoff and should therefore be interpreted with caution. Cronbach’s alpha is not ideal for binary accuracy data, but it is reported here for descriptive purposes. Importantly, item-level variability was explicitly modeled using random effects in all analyses. Tables 2 and 3 present the descriptive statistics for the recall and recognition tests.

Table 2.

Descriptive Statistics for Written Tests.

Condition	Total	Correct	Incorrect	Correct (%)	Incorrect (%)
Form recall (FRc):
Control	640	2	638	0.312	99.690
Listen	640	144	496	22.500	77.500
Repeat	640	135	505	21.090	78.910
RWL	640	193	447	30.160	69.840
RWR	640	158	482	24.690	75.310
Form recognition (FRg):
Control	640	205	435	32.030	67.970
Listen	628	415	213	66.080	33.920
Repeat	629	459	170	72.970	27.030
RWL	623	446	177	71.590	28.410
RWR	630	478	152	75.870	24.130
Meaning recognition (MRg):
Control	640	166	474	25.940	74.060
Listen	628	446	182	71.020	28.980
Repeat	627	436	191	69.540	30.460
RWL	621	418	203	67.310	32.690
RWR	630	413	217	65.560	34.440

Notes. The differences in the total number of items are due to the exclusion of items that participants indicated they already knew (1.563% for FRg and 1.688% for MRg). The form-recall test was not adjusted for prior knowledge because many learners in the control condition scored zero.

Table 3.

Descriptive Statistics for Spoken Tests.

Condition	Total	Correct	Incorrect	Correct (%)	Incorrect (%)
Form recall (FRc):
Control	448	1	447	0.223	99.780
Listen	448	97	351	21.650	78.350
Repeat	448	120	328	26.790	73.210
RWL	448	133	315	29.690	70.310
RWR	448	142	306	31.700	68.300
Form recognition (FRg):
Control	448	148	300	33.040	66.960
Listen	434	302	132	69.590	30.410
Repeat	445	327	118	73.480	26.520
RWL	435	337	98	77.470	22.530
RWR	440	326	114	74.090	25.910
Meaning recognition (MRg):
Control	448	170	278	37.950	62.050
Listen	434	292	142	67.280	32.720
Repeat	439	320	119	72.890	27.110
RWL	435	310	125	71.260	28.740
RWR	438	306	132	69.860	30.140

Notes. The differences in the total number of items are due to the exclusion of items that participants indicated they already knew (1.696% for FRg and 2.054% for MRg). The form-recall test was not adjusted for prior knowledge because many learners in the control condition scored zero.

5.1. Effect of Learning Conditions on Accuracy: Written Tests

To reveal the effect of conditions on each test type in written tests, a generalized linear mixed model (GLMM) was fitted. The fixed-effects structure included condition, test type, vocabulary size, and the condition × test type interaction. The random-effects structure consisted of random intercepts for participants and items, as well as by-participant random slopes for condition and test type.

For the written tests, likelihood ratio tests using the drop1 function showed that vocabulary size significantly improved model fit, χ²(1) = 13.338, p < .001. As vocabulary size increased, participants learned more successfully (estimate = 0.504; OR = 1.656, 95% CI [1.286, 2.132], z = 3.915). The interaction between condition and test type was also significant, χ²(8) = 60.887, p < .001 (for the summary of the model, see Table 4).

Table 4.

Generalized Linear Mixed-Effects Modeling (GLMM) Results for the Written Modality.

Fixed effect:
Predictor	Estimate	OR	95% CI	SE	z	p
(Intercept)	–0.252	0.777	[0.594, 1.017]	0.137	–1.834	.067
Condition [Listen]	–2.421	0.089	[0.059, 0.135]	0.213	–11.371	< .001
Condition [Repeat]	0.478	1.613	[1.357, 1.918]	0.088	5.417	< .001
Condition [RWL]	0.605	1.832	[1.539, 2.181]	0.089	6.811	< .001
Condition [RWR]	0.721	2.057	[1.684, 2.514]	0.102	7.062	< .001
Test type [FRc]	1.121	3.068	[2.675, 3.520]	0.07	15.994	< .001
Test type [MRg]	–2.008	0.134	[0.108, 0.167]	0.111	–18.074	< .001
Vocabulary size	0.504	1.656	[1.286, 2.132]	0.129	3.915	< .001
Condition [Listen]: Test type [FRc]	0.730	2.075	[1.403, 3.070]	0.200	3.654	< .001
Condition [Repeat]: Test type [FRc]	–0.420	0.657	[0.556, 0.777]	0.086	–4.903	< .001
Condition [RWL]: Test type [FRc]	–0.100	0.905	[0.762, 1.074]	0.088	–1.143	.253
Condition [RWR]: Test type [FRc]	–0.245	0.783	[0.659, 0.930]	0.088	–2.778	.005
Condition [Listen]: Test type [MRg]	–1.351	0.259	[0.123, 0.545]	0.380	–3.557	< .001
Condition [Repeat]: Test type [MRg]	0.321	1.378	[1.086, 1.748]	0.121	2.642	.008
Condition[RWL]: Test type [MRg]	0.191	1.211	[0.954, 1.536]	0.121	1.576	.115
Condition [RWR]: Test type [MRg]	0.520	1.682	[1.326, 2.133]	0.121	4.286	< .001
Random effect:
Group	Effect	SD		Variance
Participant	(Intercept)	0.673		0.453
	Condition [Listen]	0.537		0.289
	Condition [Repeat]	0.338		0.114
	Condition [RWL]	0.34		0.116
	Condition [RWR]	0.466		0.217
	Test type [FRc]	0.226		0.051
	Test type [MRg]	0.276		0.076
Item	(Intercept)	0.569		0.324

Notes. n (participants) = 40; n (items) = 80; observations = 9,496. Model formula: Accuracy ~ Condition × Test type + Vocabulary size + (1 + Condition + Test type | Participant) + (1 | Item). AIC = 9,348.300. Condition labels: Listen = Listening, Repeat = Repeating, RWL = Reading while listening, RWR = Reading while repeating. Coding: sum coding. R² (marginal) = .549.

Because the interaction was statistically significant, post-hoc multiple comparisons were carried out using Tukey-adjusted pairwise contrasts (via the emmeans and contrast functions from the emmeans package [Lenth, 2025]) to examine condition-specific differences within each test type. Across all test types, scores in the four learning conditions (Listening, Repeating, RWL, RWR) were higher than those in the control condition (ps < .001). Beyond these overall advantages, additional differences emerged only in form recognition: both the Repeat and RWR conditions outperformed the Listening condition. In particular, the Listening condition showed lower accuracy than Repeating (estimate = –0.447, OR = 0.640, 95% CI [0.415, 0.986], SE = 0.102, z = –2.814, p = .039) and RWR (estimate = –0.592, OR = 0.553, 95% CI [0.335, 0.913], SE = 0.102, z = –3.220, p = .011) (see Table 6 and Figure 7). For the other comparisons, see Table S4 in the supplementary material.

Figure 7.

Interaction between condition and test type in the written modality.

5.2. Effect of Learning Conditions on Accuracy: Spoken Tests

To reveal the effect of conditions on test types in spoken tests, a generalized linear mixed model (GLMM) was fitted. The model assumed a binomial distribution with a logit link function. The fixed-effects structure included condition, test type, vocabulary size, and the condition × test type interaction. The random-effects structure consisted of random intercepts for participants and items, as well as by-participant random slopes for condition and by-item random slopes for vocabulary size, specified with an independent (‘‖’) variance structure.

For the spoken tests, likelihood ratio tests showed that vocabulary size did not reach statistical significance, χ²(1) = 2.598, p = .107. Consistently, the model estimate for vocabulary size suggested only a minimal influence on learning performance (estimate = 0.16; OR = 1.17, 95% CI [0.97, 1.42]). The interaction between condition and test type was again significant, χ²(8) = 62.045, p < .001 (for the summary of the model, see Table 5).

Table 5.

Generalized Linear Mixed-Effects Modeling (GLMM) Results for the Spoken Modality.

Fixed effect:
Predictor	Estimate	OR	95% CI	SE	z	p
(Intercept)	–0.275	0.760	[0.592, 0.975]	0.127	–2.158	.031
Condition [Listen]	–2.345	0.096	[0.053, 0.175]	0.307	–7.638	< .001
Condition [Repeat]	0.320	1.377	[1.106, 1.714]	0.112	2.865	.004
Condition [RWL]	0.622	1.862	[1.489, 2.329]	0.114	5.452	< .001
Condition [RWR]	0.762	2.142	[1.684, 2.723]	0.123	6.213	< .001
Test type [FRc]	1.037	2.821	[2.418, 3.29]	0.079	13.208	< .001
Test type [MRg]	–1.971	0.139	[0.106, 0.183]	0.139	–14.134	< .001
Vocabulary size	0.159	1.173	[0.97, 1.418]	0.097	1.646	.100
Condition [Listen]:Test type [FRc]	0.754	2.126	[1.237, 3.652]	0.276	2.731	.006
Condition [Repeat]:Test type [FRc]	–0.173	0.841	[0.682, 1.037]	0.107	–1.618	.106
Condition [RWL]:Test type [FRc]	–0.218	0.804	[0.651, 0.994]	0.108	–2.019	.044
Condition [RWR]:Test type [FRc]	–0.092	0.912	[0.736, 1.131]	0.109	–0.837	.403
Condition [Listen]:Test type [MRg]	–1.855	0.156	[0.055, 0.447]	0.535	–3.466	< .001
Condition [Repeat]:Test type [MRg]	0.360	1.434	[1.049, 1.96]	0.160	2.257	.024
Condition [RWL]:Test type [MRg]	0.363	1.438	[1.052, 1.965]	0.159	2.279	.023
Condition [RWR]:Test type [MRg]	0.456	1.578	[1.158, 2.15]	0.158	2.890	.004
Random effect:
Group	Effect		SD		Variance
Participant	(Intercept)		0.442		0.195
	Condition [Listen]		0.760		0.577
	Condition [Repeat]		0.350		0.123
	Condition [RWL]		0.369		0.136
	Condition [RWR]		0.437		0.191
Item	(Intercept)		0.509		0.259
Item	Vocabulary size		0.242		0.058

Notes. n (participants) = 28; n (items) = 80; observations = 6,636. Model formula: Accuracy ~ Condition × Test type + Vocabulary size + (1 + Condition | Participant) + (1 + Vocabulary size || Item). AIC = 6,986.100. Condition labels: Listen = Listening, Repeat = Repeating, RWL = Reading while listening, RWR = Reading while repeating. Coding: sum coding. R² (marginal) = .538.

For the spoken tests, scores in all four learning conditions (Listen, Repeat, RWL, RWR) were higher than those in the control condition. Additional contrasts revealed that, within form recall, the Listening condition yielded lower accuracy than the RWR condition. Specifically, Listening was significantly outperformed by RWR (estimate = –0.637, OR = 0.529, 95% CI [0.329, 0.851], SE = 0.092, z = –3.655, p = .002) (see Table 6 and Figure 8). For the other comparisons, see Table S5 in the supplementary material.

Table 6.

Results of Multiple Comparisons for the Written and Spoken Modalities.

Test type	Contrast	Estimate	OR	95% CI	SE	z	p
Written test:
FRg	Listen / Repeat	–0.447	0.640	[0.415, 0.986]	0.102	–2.814	.039
FRg	Listen / RWR	–0.592	0.553	[0.335, 0.913]	0.102	–3.220	.011
Spoken test:
FRc	Listen / RWR	–0.637	0.529	[0.329, 0.851]	0.092	–3.655	.002

Notes. For ease of interpretation, only significant contrasts are reported here. For the full set of results, including all comparisons, see Tables S4 and S5 in supplemental material.

Figure 8.

Interaction between condition and test type in the spoken modality.

6. Discussion

The present study examined how vocalization influences L2 verb–noun idiom learning under a pure-list design and across different testing modalities (written, spoken) and formats (form recall, form recognition, meaning recognition). The results showed a selective but robust pattern. In the written tests, the Repeat and RWR conditions outperformed listening in form-recognition, whereas no significant differences emerged for form-recall or meaning-recognition. In the spoken tests, RWR led to higher scores than listening in spoken form-recall, while no between-condition differences were observed for spoken form-recognition or spoken meaning-recognition. In other words, production benefits were confined to written form-recognition and spoken form-recall, with no detectable advantages for meaning-recognition in either modality.

Vocabulary size was a significant predictor only in the written modality, not in the spoken modality, although this was not a primary focus of the present study. This asymmetry may partly reflect modality congruence, as vocabulary size was measured using a written test and may therefore have been more closely aligned with the written posttests than with the spoken posttests. As Uchihara and Clenton (2023) noted, written and spoken vocabulary knowledge are related but partially independent, and discrepancies between the two may arise at the individual level. In addition, the smaller sample size in the spoken-modality group may have reduced the statistical power to detect a comparable effect. The apparent modality asymmetry should therefore be interpreted with caution.

These findings provide a ‘yes’ to research question 1 (Does vocalization enhance the learning of L2 verb–noun idioms under pure-list conditions?). Even under a within-participant pure-list design – where each list is learned using a single, consistent method – production benefits were observed for L2 verb–noun idioms. This is noteworthy given prior claims that the production effect is less robust in pure-list designs compared to mixed-list designs (Fawcett, 2013; Fawcett & Ozubko, 2016; Saint-Aubin et al., 2021). At least for some test formats (written form-recognition and spoken form-recall), vocalization-based conditions (Repeat, RWR) yielded better performance than listening alone, suggesting that the production effect can generalize to pure-list situations in L2 verb–noun idiom learning.

From the perspective of research on the production effect, the advantage observed in a pure-list context is likely to reflect stronger memory traces rather than item-level distinctiveness (Fawcett, 2013; Fawcett & Ozubko, 2016). This is because distinctiveness is assumed to arise only in mixed-list designs, in which produced and non-produced items are presented together within the same list (Hunt, 2006; MacLeod et al., 2010). In such a pure-list context, vocalization may enhance learning by promoting more complete lexical processing during encoding (Gao et al., 2017) and by reducing the skipping of input often associated with silent reading (Rayner, 2009), thereby strengthening memory representations.

The fact that effects were clearest on measures targeting formal knowledge (form-recognition, spoken form-recall) is consistent with the view that production primarily strengthens surface-form representations (Ozubko et al., 2012). In contrast to the findings of Icht and Mama (2022) and Hashizaki (2024a), no effect emerged for meaning recognition. This may be because only the verb–noun idiom forms were vocalized in the present experiment; the meanings themselves were not articulated, which may have prevented the formation of strong memory traces for L2 translations.

The results also answer research question 2 (Do the effects of vocalization (vocal vs. silent) and testing modality (written vs. spoken) interact in learning L2 verb–noun idioms?) and research question 3 (Does orthographic support during learning enhance the learning of L2 verb–noun idioms independently of vocalization?) affirmatively. Regarding research question 2, production benefits did not appear ‘across the board’ but were contingent on the combination of test modality and test format. Vocalization was facilitative in written form-recognition and spoken form-recall, but not in written form-recall, written meaning-recognition, spoken form-recognition, or spoken meaning-recognition. Thus, the effectiveness of vocalization depended on how learners were tested, not only on how they studied. TAP offers a complementary explanation for why production benefits emerged selectively across tests. According to TAP, memory performance is enhanced when the cognitive operations at encoding and retrieval align (Morris et al., 1977). In the RWR condition, learners heard the verb–noun idiom, read it, and produced it aloud. The spoken form-recall test required retrieving and articulating the verb–noun idiom in the spoken modality. This match between learning and testing demands likely boosted performance in RWR relative to Listening on spoken form-recall.

However, one might ask why the Repeat condition, in which the learning modality also matched the test modality, did not produce a significant advantage on the spoken form-recall test. This issue is closely related to research question 3. Unlike Repeat, the RWR condition involved the simultaneous presentation of text and speech. Therefore, in addition to the production effect and TAP, the retention of verb–noun idioms may also have been enhanced by the simultaneous presentation of visual and verbal information, as proposed by Paivio (2007).

Taken together, the findings indicate that the production effect and TAP frameworks play complementary roles in interpreting L2 verb–noun idiom learning. Accounts of the production effect explain whether and to what extent vocalization can strengthen memory traces for verb–noun idioms, particularly at the level of form (Fawcett, 2013; Fawcett & Ozubko, 2016; MacLeod et al., 2010). TAP clarifies when these strengthened traces are most likely to manifest as performance gains; that is, under test conditions that partially mirror the multimodal, articulatory processing engaged during learning (Lightbown, 2007; Morris et al., 1977; Uchihara, 2023). In this sense, vocalization can be viewed as a form-strengthening technique whose observable benefits depend on the degree of processing overlap between study and test. Furthermore, the findings extend previous studies on multimodal input in L2 verb–noun idiom learning (e.g. Huang et al., 2022; Wang et al., 2023) by suggesting that written input may provide an additional benefit.

However, these findings should be interpreted with two important qualifications. First, the study adopted an operational focus on verb–noun idioms. This choice limits the generalizability of the results to other types of multiword expressions. Second, all findings are based on immediate posttest measures; therefore, the claims should be appropriately constrained to short-term learning rather than extended to long-term retention.

7. Pedagogical Implications

From an L2 pedagogical perspective, the findings have several implications for the instruction of verb–noun idioms. First, production benefits were observed even under a within-participant pure-list design, in which each list was learned using a single, consistent method and no non-vocalized comparison items were included within the same list. This suggests that the benefits of vocalization are not limited to situations in which only some items in a list are vocalized in contrast to others. Rather, vocalization may facilitate learning even when an entire list is practiced through vocalization.

Second, Repeat and RWR outperformed Listening in written form recognition, and RWR also outperformed Listening in spoken form recall. These results suggest that combining auditory input with overt articulation can help consolidate form-related knowledge of verb–noun idioms. Classroom practices such as reading aloud, shadowing, and choral repetition may therefore be particularly beneficial when the instructional goal is to stabilize orthographic and phonological forms of verb–noun idioms. Third, because no production advantage was observed for meaning-recognition, it would be premature to treat vocalization as a stand-alone method for enhancing semantic learning. For meaning, vocalization may need to be supplemented with tasks that explicitly promote semantic elaboration, such as contextualized examples, meaning-judgment tasks, or activities requiring learners to integrate verb–noun idioms into meaningful discourse.

Finally, the advantage of RWR in spoken form-recall has clear relevance for speaking-oriented instruction. When teachers aim to prepare learners to retrieve and use verb–noun idioms fluently in oral communication, incorporating vocalization at the learning stage appears particularly promising. Activities that combine listening, reading, and overt production (e.g. shadowing verb–noun idioms, repeated oral reading of example sentences) may help build the kind of robust, articulatory-based representations that support rapid retrieval in speech. Overall, the present findings suggest that vocalization is best conceptualized as a targeted tool for strengthening the formal, especially phonological and orthographic, aspects of L2 verb–noun idioms, whose benefits are maximized when assessment and classroom practice are aligned with the processing demands of production.

It should be noted, however, that the present study focused exclusively on verb–noun idioms and examined learning only through immediate posttests. Readers should therefore bear in mind that it remains unclear whether similar effects of vocalization would be observed for other types of multiword expressions (e.g. collocations, phrasal verbs, or longer lexical bundles) or over longer retention intervals. Accordingly, any pedagogical implications in the present study should be interpreted with caution rather than taken as directly generalizable.

8. Conclusions

The present study examined the effects of vocalization on verb–noun idiom learning, yet several issues remain open for further investigation.

First, the generalizability of the findings remains limited. It is unclear whether the observed effects would extend to other types of multiword expressions, such as collocations, phrasal verbs, and lexical bundles. Replicating the present experimental design with a broader range of linguistic items would allow for stronger claims regarding the generalizability of vocalization-based learning advantages. In addition, the instructional approach adopted in this study was decontextualized, and the effects of contextualized input were therefore not examined. Likewise, learning conditions involving visual input, such as images or videos, were not considered. Future research should investigate whether the facilitative effects of vocalization extend to contextualized settings, for example, when verb–noun idioms are embedded in discourse, as well as to multimedia learning environments. In particular, it would be valuable to compare the effects of vocalization across different input modalities, such as reading, listening, reading while listening, and viewing with or without captions, under both contextualized and decontextualized conditions. Such research would provide a clearer understanding of how vocalization interacts with input modality and would further strengthen the ecological validity of the findings. Furthermore, the present study did not include a delayed posttest. Previous studies (Hashizaki, 2024a; Icht & Mama, 2022) have shown that the production effect can persist in delayed tests when a mixed-list design and written modality are used. However, it remains unclear whether the advantages of vocalization are also maintained over time under a pure-list design or in a spoken modality. Examining the durability of vocalization effects under these conditions would allow for broader generalization of the present findings.

Second, the study did not examine potential moderating variables. Item-level and participant-level characteristics were neither systematically controlled for nor included in the analyses. At the item level, factors such as semantic transparency, familiarity, and length may influence the effectiveness of vocalization. At the participant level, learner variables – including proficiency level, working memory capacity, prior knowledge of verb–noun idioms, and L1–L2 typological distance – may also moderate their effects. Future research should therefore measure and statistically model these factors to investigate their interaction with vocalization. Incorporating such variables would further strengthen the ecological validity of the findings.

Third, this study was also limited in terms of outcome measures and the investigation of underlying mechanisms. Although the findings indicate improvements in test performance, no measures of real-time processing or fluency were collected. Incorporating psycholinguistic measures, such as reaction time, eye-movement patterns obtained through eye-tracking, or self-paced reading, would allow researchers to examine whether vocalization facilitates not only greater accuracy but also more efficient processing. In addition, the present study did not examine whether the acquired expressions were incorporated into actual language production. Follow-up writing or speaking tasks in which participants use verb–noun idioms communicatively would help determine whether vocalization promotes the uptake and productive use of target items. Furthermore, participants in the present study were required to vocalize only the orthographic forms of the expressions and not their semantic content. Future research should investigate whether vocalizing form together with meaning enhances learning, particularly for multiword expressions whose interpretation depends on compositional or idiomatic semantics.

Finally, there were design-related constraints. Although the imbalance in group sizes across modalities resulted from attrition rather than the assignment procedure, it may have created unequal statistical power between the written and spoken models, particularly for detecting interaction effects. Accordingly, cross-modality differences should be interpreted with caution. Replication with a larger sample would allow for more fine-grained comparisons of learning effects across modalities.

Taken together, these directions highlight the need for a more fine-grained investigation of how and why vocalization supports verb–noun idiom learning, and under what linguistic, cognitive, and instructional conditions the advantages emerge. By identifying these mechanisms, future research can refine theoretically grounded approaches to L2 instruction and harness the potential of vocalization to promote durable and transferable language learning.

Supplemental Material

sj-docx-1-ltr-10.1177_13621688261454077 – Supplemental material for The Production Effect in Second Language Verb–Noun Idiom Learning: Evidence From Pure-List Designs and Transfer-Appropriate Processing

Supplemental material, sj-docx-1-ltr-10.1177_13621688261454077 for The Production Effect in Second Language Verb–Noun Idiom Learning: Evidence From Pure-List Designs and Transfer-Appropriate Processing by Ryotaro Hashizaki in Language Teaching Research

Footnotes

Acknowledgements

This article is based on a paper presented at the 2025 JACET Kansai Chapter Conference on 7 March 2026.

ORCID iD

Ryotaro Hashizaki

Ethical Considerations

This study received approval from the Ethical Review Committee for Research Involving Human Participants at Matsuyama University.

Consent to Participate

Written informed consent was obtained from all participants prior to their participation in the study.

Funding

The author disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by a JSPS KAKENHI Grant-in-Aid for Research Activity Start-up (grant number 24K22480).

Declaration of Conflicting Interests

The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

The data that support the findings of this study are available from the author upon reasonable request.

Statement on AI-Based Language Editing

Generative artificial intelligence (ChatGPT, OpenAI) was used solely for grammatical and stylistic editing of the manuscript. No content generation or data analysis was performed by the AI. The author has reviewed and approved all revisions and takes full responsibility for the final version of the manuscript.

Supplemental Material

Supplemental material for this article is available online.

Notes

References

Alali

F. A.

Schmitt

(2012). Teaching formulaic sequences: The same as or different from teaching single words? TESOL Journal, 3(2), 153–180. https://doi.org/10.1002/tesj.13

Barr

D. J.

Levy

Scheepers

Tily

H. J.

(2013). Random effects structure for confirmatory hypothesis testing: Keep it maximal. Journal of Memory and Language, 68(3), 255–278. https://doi.org/10.1016/j.jml.2012.11.001

Bates

Mächler

Bolker

Walker

(2015). Fitting linear mixed-effects models using lme4. Journal of Statistical Software, 67(1), 1–48. https://doi.org/10.18637/jss.v067.i01

Boers

Eyckmans

Kappel

Stengers

Demecheleer

(2006). Formulaic sequences and perceived oral proficiency: Putting a Lexical Approach to the test. Language Teaching Research, 10(3), 245–261. https://doi.org/10.1191/1362168806lr195oa

Cieślicka

A. B.

(2006). Literal salience in on-line processing of idiomatic expressions by second language learners. Second Language Research, 22(2), 115–144. https://doi.org/10.1191/0267658306sr263oa

Conklin

Schmitt

(2012). The processing of formulaic language. Annual Review of Applied Linguistics, 32, 45–61. https://doi.org/10.1017/S0267190512000074

Cucchiarini

Hubers

Strik

(2022). Learning L2 idioms in a CALL environment: The role of practice intensity, modality, and idiom properties. Computer Assisted Language Learning, 35(4), 863–891. https://doi.org/10.1080/09588221.2020.1752734

Durrant

Schmitt

(2010). Adult learners’ retention of collocations from exposure. Second Language Research, 26(2), 163–188. https://doi.org/10.1177/0267658309349431

Ellis

N. C.

Sinclair

(1996). Working memory in the acquisition of vocabulary and syntax: Putting language in good order. The Quarterly Journal of Experimental Psychology, 49(1), 234–250. https://doi.org/10.1080/713755604

10.

Estaji

Montazeri

M. R.

(2026). Mapping oral collocational use by measuring collocational accuracy, complexity, and fluency across different proficiency levels. International Review of Applied Linguistics in Language Teaching. Advance online publication. https://doi.org/10.1515/iral-2025-0199

11.

Fawcett

J. M.

(2013). The production effect benefits performance in between-subject designs: A meta-analysis. Acta Psychologica, 142(1), 1–5. https://doi.org/10.1016/j.actpsy.2012.10.001

12.

Fawcett

J. M.

Ozubko

J. D.

(2016). Familiarity, but not recollection, supports the between-subject production effect in recognition memory. Canadian Journal of Experimental Psychology/Revue Canadienne de Psychologie Experimentale, 70(2), 99–115. https://doi.org/10.1037/cep0000089

13.

Fox

Weisberg

(2019). An R companion to applied regression (3rd ed.). Sage.

14.

Gao

Xiang

Xue

Radach

(2017). Parafoveal preview benefits during silent and oral reading. Acta Psychologica Sinica, 49(11), 1357–1369. https://doi.org/10.3724/SP.J.1041.2017.01357

15.

Gathercole

S. E.

Conway

M. A.

(1988). Exploring long-term modality effects: Vocalization leads to best retention. Memory & Cognition, 16(2), 110–119. https://doi.org/10.3758/bf03213478

16.

D. T.

(2017). The impact of different instructions on Vietnamese EFL students’ acquisition of formulaic sequences. English Language Teaching, 10(8), 18–30. https://doi.org/10.5539/elt.v10n8p18

17.

Hartig

(2024). DHARMa: Residual diagnostics for hierarchical (multi-level/mixed) regression models (R package version 0.4.7). https://doi.org/10.32614/CRAN.package.DHARMa

18.

Hashizaki

(2021). The effect of shadowing on memorization of English chunks and its theoretical basis: Influence of repetition and material difficulty. LET Journal of Central Japan, 31, 39–52. https://doi.org/10.20656/letcj.31.0_39

19.

Hashizaki

(2024a). Does vocalization enhance L2 formulaic sequence learning? Language Teaching Research. Epub ahead of print 31 January 2024. https://doi.org/10.1177/13621688231224040

20.

Hashizaki

(2024b). Theoretical validation of the effect of shadowing on the memorization of multiword units. JACET Kansai Journal, 26, 1–19. https://doi.org/10.60375/jacetkansaijournal.26.0_1

21.

Hashizaki

Sugiura

(2026). Vocalization, text presentation, and learner-related factors in learning L2 multiword expressions in sentences. International Journal of Applied Linguistics. Advance online publication. https://doi.org/10.1111/ijal.70143

22.

Hopkins

R. H.

Edwards

R. E.

(1972). Pronunciation effects in recognition memory. Journal of Verbal Learning and Verbal Behavior, 11(4), 534–537. https://doi.org/10.1016/S0022-5371(72)80036-7

23.

Howarth

(1998). Phraseology and second language proficiency. Applied Linguistics, 19(1), 24–44. https://doi.org/10.1093/applin/19.1.24

24.

Huang

Zhang

Liu

Huang

(2022). English phrase learning with multimodal input. Frontiers in Psychology, 13, 828022. https://doi.org/10.3389/fpsyg.2022.828022

25.

Hubers

Cucchiarini

Strik

(2020). Second language learner intuitions of idiom properties: What do they tell us about L2 idiom knowledge and acquisition? Lingua, 246, 102940. https://doi.org/10.1016/j.lingua.2020.102940

26.

Hunt

R. R.

(2006). The concept of distinctiveness in memory research. In Hunt

R. R.

Worthen

J. B.

(Eds.), Distinctiveness and memory (pp. 3–25). Oxford University Press. https://doi.org/10.1093/acprof:oso/9780195169669.003.0001

27.

Icht

Mama

(2022). The effect of vocal production on vocabulary learning in a second language. Language Teaching Research, 26(1), 79–98. https://doi.org/10.1177/1362168819883894

28.

JACET Committee of Basic Words Revision. (2003). JACET list of 8000 basic words: JACET 8000. Japan Association of College English Teachers. https://ci.nii.ac.jp/ncid/BB21043002

29.

Lenth

R. V.

(2025). emmeans: Estimated marginal means, aka least-squares means (R package version 1.11.2-8). https://doi.org/10.32614/CRAN.package.emmeans

30.

Lightbown

P. M.

(2007). 3. Transfer appropriate processing as a model for classroom second language acquisition. In Han

(Ed.), Understanding second language process (pp. 27–44). Multilingual Matters. https://doi.org/10.21832/9781847690159-005

31.

Lüdecke

Ben-Shachar

M. S.

Patil

Waggoner

Makowski

(2021). performance: An R package for assessment, comparison and testing of statistical models. Journal of Open Source Software, 6(60), 3139. https://doi.org/10.21105/joss.03139

32.

MacLeod

C. M.

Gopie

Hourihan

K. L.

Neary

K. R.

Ozubko

J. D.

(2010). The production effect: Delineation of a phenomenon. Journal of Experimental Psychology: Learning, Memory, and Cognition, 36(3), 671–685. https://doi.org/10.1037/a0018785

33.

Matuschek

Kliegl

Vasishtha

Baayen

Bates

(2017). Balancing type I error and power in linear mixed models. Journal of Memory and Language, 94, 305–315. https://doi.org/10.1016/j.jml.2017.01.001

34.

McLean

Hogg

Kramer

(2014). Estimations of Japanese university learners’ English vocabulary sizes using the vocabulary size test. Vocabulary Learning and Instruction, 3(2), 47–55. https://doi.org/10.7820/vli.v03.2.mclean.et.al

35.

Meara

Miralpeix

(2016). Tools for researching vocabulary. Multilingual Matters. https://doi.org/10.21832/9781783096473

36.

Morris

D. C.

Bransford

J. D.

Franks

J. J.

(1977). Levels of processing versus transfer appropriate processing. Journal of Verbal Learning and Verbal Behavior, 16(5), 519–533. https://doi.org/10.1016/S0022-5371(77)80016-9

37.

Nation

Webb

(2011). Researching and analyzing vocabulary. Cengage Learning.

38.

Ozubko

J. D.

Gopie

MacLeod

C. M.

(2012). Production benefits both recollection and familiarity. Memory and Cognition, 40(3), 326–338. https://doi.org/10.3758/s13421-011-0165-1

39.

Paivio

(2007). Mind and its evolution: A dual coding theoretical approach (1st ed.). Psychology Press. https://doi.org/10.4324/9781315785233

40.

Chang

D. Y.-S.

Wang

(2026). Young EFL learners’ incidental collocation learning through different multimodal inputs: A mixed-methods study. International Journal of Applied Linguistics, 36(1), 916–926. https://doi.org/10.1111/ijal.12828

41.

R Core Team. (2025). R: A language and environment for statistical computing. R Foundation for Statistical Computing. Available online at: https://www.R-project.org

42.

Rafatbakhsh

Ahmadi

(2019). A thematic corpus-based study of idioms in the corpus of contemporary American English. Asian-Pacific Journal of Second and Foreign Language Education, 4(1), 1–21. https://doi.org/10.1186/s40862-019-0076-4

43.

Rafieyan

(2018). Role of knowledge of formulaic sequences in language proficiency: Significance and ideal method of instruction. Asian-Pacific Journal of Second and Foreign Language Education, 3(1), 1–23. https://doi.org/10.1186/s40862-018-0050-6

44.

Rayner

(2009). Eye movements and attention in reading, scene perception, and visual search. The Quarterly Journal of Experimental Psychology, 62(8), 1457–1506. https://doi.org/10.1080/17470210902816461

45.

Saint-Aubin

Yearsley

J. M.

Poirier

Cyr

Guitard

(2021). A model of the production effect over the short-term: The cost of relative distinctiveness. Journal of Memory and Language, 118, 1–26. https://doi.org/10.1016/j.jml.2021.104219

46.

Seibert

L. C.

(1927). An experiment in learning French. Journal of Educational Psychology, 18(5), 294–309. https://doi.org/10.1037/h0074206

47.

Siyanova-Chanturia

Conklin

Schmitt

(2011). Adding more fuel to the fire: An eye-tracking study of idiom processing by native and non-native speakers. Second Language Research, 27(2), 251–272. https://doi.org/10.1177/0267658310382068

48.

Siyanova-Chanturia

Sonbul

(2025). The processing of multi-word expressions: A research agenda for the next 10 years. Language Teaching. Epub ahead of print 27 August 2025. https://doi.org/10.1017/s0261444825000606

49.

Steinel

M. P.

Hulstijn

J. H.

Steinel

(2007). Second language idiom learning in a paired-associate paradigm: Effects of direction of testing, idiom imageability, and idiom transparency. Studies in Second Language Acquisition, 29(3), 449–484. https://doi.org/10.1017/S0272263107070271

50.

Uchihara

(2023). How does the test modality of weekly quizzes influence learning the spoken forms of second language vocabulary? TESOL Quarterly, 57(2), 595–617. https://doi.org/10.1002/tesq.3176

51.

Uchihara

Clenton

(2023). The role of spoken vocabulary knowledge in second language speaking proficiency. The Language Learning Journal, 51(3), 376–393. https://doi.org/10.1080/09571736.2022.2080856

52.

Uemura

Ishikawa

(2004). JACET 8000 and Asia TEFL vocabulary initiative. Journal of Asia TEFL, 1(1), 333–347.

53.

Wang

Liu

Lee

C. I.

(2023). The efficacy of CALL glossing on EFL learners’ idiom learning. Acta Psychologica, 241, 104060. https://doi.org/10.1016/j.actpsy.2023.104060

54.

Webb

Chang

(2022). How does mode of input affect the incidental learning of collocations? Studies in Second Language Acquisition, 44(1), 35–56. https://doi.org/10.1017/S0272263120000297

55.

Webb

Newton

Chang

(2013). Incidental learning of collocation. Language Learning, 63(1), 91–120. https://doi.org/10.1111/j.1467-9922.2012.00729.x

56.

Wray

Perkins

(2000). The functions of formulaic language: An integrated model. Language & Communication, 20(1), 1–28. https://doi.org/10.1016/S0271-5309(99)00015-4

57.

(2018). Measuring ‘spoken collocational competence’ in communicative speaking assessment. Language Assessment Quarterly, 15(3), 255–272. https://doi.org/10.1080/15434303.2018.1482900

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.06 MB