L2 phonetic training research

Over the past 50 years, much attention has been given to the effects of training in L2 phonetic learning. A range of empirical studies has demonstrated that provision of explicit phonetic training promotes L2 phonetic perception on both segmental (e.g., Bradlow, Reference Bradlow, Hansen and Zampini2008 for Japanese speakers’ English [r] and [l] acquisition) and suprasegmental levels (e.g., Wang et al., Reference Wang, Sereno, Jongman and Hirsch2003 for Americans’ acquisition of Mandarin lexical tones). Such training effectiveness can be amplified through the use of multiple talkers (e.g., High Variability Phonetic Training; Uchihara et al., Reference Uchihara, Karas and Thomson2025), and key features are acoustically enhanced (e.g., F3 enhancement in English [r] and [l] contrast; Iverson et al., Reference Iverson, Hazan and Bannister2005).

Recent meta-analyses offer further support for these findings, revealing that phonetic training has a medium to large positive effect on L2 phonetic competence (e.g., Saito & Plonsky, Reference Saito and Plonsky2019; Sakai & Moorman, Reference Sakai and Moorman2018; Yao et al., Reference Yao, He, Chen and Zhu2025). In a recent comprehensive meta-analysis, Yao et al. (Reference Yao, He, Chen and Zhu2025) synthesized findings from 65 studies involving 2,793 learners and reported a large average effect size of d = 0.762 for phonetic training. More specifically, analyzing 25 years of research, Sakai and Moorman’s (Reference Sakai and Moorman2018) meta-analysis found that perceptual training led to medium-sized improvements in perception (d = 0.92) and small but significant improvements in production (d = 0.54). This suggests that perceptual training can indeed have a positive effect on production abilities in L2 phonetic learning (Flege & Bohn, Reference Flege, Bohn and Wayland2021). Finally, Saito and Plonsky (Reference Saito and Plonsky2019) conducted a meta-analysis of 77 studies and found that L2 pronunciation instruction is generally effective with a moderate overall effect size (d = 0.61). Greater improvements were observed when targeting specific features and using objective, controlled assessments rather than global, subjective evaluations.

In a typical phonetic training study, participants take a pre-test, complete a series of short training sessions (typically 10–30 minutes), and then undergo immediate and delayed post-tests. When it comes to the perception domain aspects of L2 phonetic training (i.e., the main focus of the study), recent review articles demonstrated that a few hours of training typically result in approximately 5–10% of gains (e.g., Saito et al., Reference Saito, Hanzawa, Petrova, Kachlicka, Suzukida and Tierney2022). One methodological issue that has become standard practice—without substantial scholarly scrutiny—is the distribution schedule of these training sessions. Most existing studies have employed massed or evenly spaced training input (e.g., using a one-day inter-session interval). More broadly, L2 phonetic learning research has long emphasized the importance of input. There is ample evidence that both the quantity of input (e.g., length of immersion; Trofimovich & Baker, Reference Trofimovich and Baker2006) and the quality of input (e.g., the frequency and intensity of L2 interaction; Derwing & Munro, Reference Derwing and Munro2013) shape the rate and ultimate success of speech development. However, the question of how best to schedule such input—particularly in short-term instructional contexts—remains largely underexplored.

Massed practice is a learning strategy in which study or skill rehearsal is concentrated into a single, continuous, and intensive period with little to no rest between repetitions. Often referred to as “cramming,” it contrasts with distributed practice, which spaces learning sessions over time (for a meta-analysis of verbal recall in cognitive psychology, see Cepeda et al., Reference Cepeda, Pashler, Vul, Wixted and Rohrer2006). In research in L2 learning outside phonetics, it has been shown that training outcomes can vary—and sometimes improve—depending on the distribution schedule (e.g., massed vs. equal vs. expanding spacing; Suzuki, Nakata, & DeKeyser, Reference Suzuki, Nakata and Dekeyser2019). Despite its potential importance, this issue has received little attention in L2 phonetics, with Alfotais et al. (Reference Alfotais, Mahdi and Alkhammash2025) being a rare exception (see below).

In the case of L2 phonetic training, previous studies have employed a wide range of schedules, varying in duration, number of sessions, and overall intensity. As shown in Uchihara et al.’s (Reference Uchihara, Karas and Thomson2025) a methodological synthesis, programs have ranged from highly condensed formats delivered within a few days to more extended schedules lasting several weeks, with total training time varying from as little as 60 minutes (i.e., massed learning) to more than 20 hours (i.e., spaced learning; see Table 2 on p. 810). However, these spacing decisions were typically made on pragmatic rather than theoretical grounds, and no consistent pattern has been established linking timing/spacing parameters to learning outcomes. While perception gains tend to increase with overall training time up to a certain threshold (approximately six hours), neither the number of sessions nor their temporal distribution systematically predicted outcomes. Thus, although extant L2 phonetic training research provides abundant evidence for the efficacy of L2 phonetic training, the specific role of training distribution has not yet been directly addressed. The current study was designed to correspond to these concerns.

Distributed practice in L2 vocabulary research

Turning to L2 vocabulary literature, researchers have begun to investigate how the distribution of input—not just its amount or quality—affects learning outcomes (Suzuki et al., Reference Suzuki, Nakata and Dekeyser2019). This line of inquiry draws from the well-established spacing effect in cognitive psychology, which posits that learning is more robust when practice is spaced out over time rather than massed into a single session (Baddeley, Reference Baddeley1997; Cepeda et al., Reference Cepeda, Pashler, Vul, Wixted and Rohrer2006). One influential origin of this research is Ebbinghaus’s (1913) discovery that forgetting follows a predictable curve, but can be mitigated by repeated retrieval spaced over time.

In language learning, spaced practice has consistently been shown to yield better retention outcomes than massed practice (e.g., Bloom & Shuell, Reference Bloom and Shuell1981; Hamouda, Reference Hamouda2021; Namaziandost et al., Reference Namaziandost, Anwar and Neisi2020). Studies show that spaced learning (either equal or expanding) can lead to 20–40% higher retention compared to massed practice, and this effect is especially clearly observed in long-term retention (e.g., after a week or more).

Within the broader category of spaced practice, scholars have further distinguished between absolute spacing (i.e., total time between repetitions) and relative spacing (i.e., the pattern of intervals between repetitions), both of which have been shown to influence learning. Relative spacing has received particular attention in L2 vocabulary acquisition. It includes equal spacing—where intervals between study sessions are fixed—and expanding spacing—where intervals gradually increase across repetitions. Expanding spacing is often argued to be more effective because it increases retrieval difficulty over time, which can promote stronger memory consolidation (Karpicke & Bauernschmidt, Reference Karpicke and Bauernschmidt2011; Landauer & Bjork, Reference Landauer, Bjork, Gruneberg, Morris and Sykes1978; Pyc & Rawson, Reference Pyc and Rawson2009).

However, empirical evidence comparing equal and expanding spacing has yielded mixed results. Nakata (Reference Nakata2015), in a carefully controlled study of 128 Japanese learners, reported that expanding spacing resulted in significantly higher receptive vocabulary scores than equal spacing, across short, medium, and long absolute spacing schedules. Similarly, Schuetze (Reference Schuetze2015) found advantages for expanding spacing in short-term retention, though equal spacing outperformed in the long-term. Karpicke and Roediger (Reference Karpicke and Roediger2007) reported comparable findings, suggesting that the benefits of expanding spacing may be more pronounced in the early stages of learning.

Conversely, other studies have failed to find consistent advantages. Kang et al. (Reference Kang, Lindsey, Mozer and Pashler2014), for example, compared expanding and equal spacing schedules in a four-week vocabulary learning study and found no significant differences between the two groups on delayed post-tests. This echoes earlier work by Pyc and Rawson (Reference Pyc and Rawson2007), who also found no long-term advantage of expanding over equal spacing in their study of Swahili-English word pair retention. Overall, while the spacing effect is well established, the specific efficacy of expanding vs. equal spacing remains unsettled.

It is important to note that much of the existing research has focused on vocabulary learning. Moreover, outcomes may depend on moderating factors such as the total number of exposures, the timing of retrieval, the nature of the linguistic targets, and whether feedback is provided during practice. As stressed earlier, surprisingly little is known about the optimal timing of input in the context of L2 phonetic learning and training.

The mechanisms underlying L2 vocabulary and phonetic learning may differ substantially. On the one hand, vocabulary learning (as tested in the existing literature) involves the conscious cognitive operation of higher-order information, such as facts, conceptual knowledge, vocabulary, and morphosyntactic rules. In this case, learning typically refers to the explicit acquisition of form-meaning mappings using declarative memory. In such contexts, introducing different types of input variability through distributed practice (massed vs. spaced; equal vs. expanding spacing) can strengthen memory traces and retrieval (Landauer & Bjork, Reference Landauer, Bjork, Gruneberg, Morris and Sykes1978). On the other hand, phonetic learning takes place on relatively lower-order and perceptual levels. Such learning processes entail the readjustment of abstract, acoustic, and articulatory details of new sounds relative to L1 phonetic categories. Here, learning relies on procedural rather than declarative memory, and the process tends to be implicit, subconscious, and gradual (Best & Tyler, Reference Best, Tyler, Bohn and Munro2007).

Given these fundamentally different learning mechanisms, researchers in L2 acquisition and teaching have increasingly shown interest in testing the generalizability (or specificity) of findings across different domains of L2 learning (e.g., extending task-based language teaching from vocabulary and grammar to phonology; Mora-Plaza et al., Reference Mora-Plaza, Mora, Ortega and Aliaga-Garcia2024; Xu et al., Reference Xu, Saito and Mora-Plaza2024). Following this line of thought, we argue that it is important to explore whether, to what degree, and how insights on distributed practice (derived mainly from L2 vocabulary and grammar research) can be generalized to L2 phonetic learning. At the very least, it is possible that the effects of distributed practice differ across vocabulary/grammar learning and phonetic training, and such discrepancies should be examined to further our understanding of the role of distributed practice in language learning more broadly.

Motivation for current study

To our knowledge, Alfotais et al.’s (Reference Alfotais, Mahdi and Alkhammash2025) recent empirical study represents the very first systematic attempt to compare the effects of massed and spaced repetition schedules. In their quasi-experimental investigation, 49 Saudi EFL learners were trained on the acquisition of the English /b/–/p/ contrast. Using a within-subjects design, the study assessed L2 phonetic development both immediately after training and following a four-week delay. The results revealed that spaced repetition consistently outperformed massed repetition for both short- and long-term retention, enhancing effectiveness by approximately 55–80% compared with massed repetition—a pattern strikingly similar to the figures reported in L2 vocabulary research (Nakata, Reference Nakata2015).

While Alfotais et al. (Reference Alfotais, Mahdi and Alkhammash2025) represent an important first step in applying distributed practice frameworks to L2 phonetic learning, their work raises two important questions that merit further investigation. First, can the benefits of spaced over massed repetition be replicated in a different L2 phonetic learning context? Second, might the effects of spaced repetition be further enhanced by considering types of spacing—specifically, by distinguishing between equal and expanding spacing formats, two theoretically significant subtypes of distributed practice?

The present study built on this foundation by not only replicating the massed vs. spaced comparison but also extending it to examine the relative efficacy of equal vs. expanding spacing schedules. Given that most existing phonetic training studies have employed massed practice, our ultimate aim was to provide principled methodological guidance on the extent to which the effectiveness of the same phonetic training can change or increase when researchers adopt different timing regimes—massed, equal, or expanding. These findings will ultimately inform key theoretical questions concerning the temporal distribution of phonetic input in L2 speech training and how differently distributed input may affect L2 phonetic learning.

In the context of Mandarin learners’ acquisition of the English vowels /ɛ/ and /æ/, the current study was designed to address the following three research questions, each accompanied by a corresponding prediction:

1. 1. Does a spaced practice schedule facilitate L2 phonetic learning more effectively than a massed practice schedule? If so, how much?

2. 2. How do different types of spaced practice schedules (equal vs. expanding) influence L2 phonetic learning? If so, how much?

3. 3. Do the relative benefits of input distribution (massed, equal, expanding) vary as a function of learners’ perceptual–cognitive abilities?

Prediction 1: Massed vs. spaced practice schedules

Following previous research in L2 vocabulary learning (e.g., Nakata, Reference Nakata2015), we first aimed to test the generalizability of the spacing effect—specifically, whether the relative advantage of spaced over massed practice would extend to L2 phonetic learning. Drawing on prior phonetic training studies (e.g., Alfotais et al., Reference Alfotais, Mahdi and Alkhammash2025), we predicted that spaced practice would lead to greater learning gains than massed practice. This effect was expected to be substantial, as Alfotais et al. reported significantly larger effect sizes for spaced conditions (d > 2) than for massed conditions (d = 1–2) at both immediate and delayed post-tests. Such findings suggest that spaced practice can result in nearly twice the retention gains of massed practice (Alfotais et al., Reference Alfotais, Mahdi and Alkhammash2025; Nakata, Reference Nakata2015).

Prediction 2: Equal vs. expanding spacing

Although spaced repetition is a well-established method for enhancing memory, debate remains as to whether expanding spacing—gradually increasing the intervals between reviews—yields greater learning gains than equal spacing, where intervals remain constant. Proponents of expanding spacing argue that increasing retrieval difficulty strengthens memory by engaging effortful retrieval and reconsolidation processes (Karpicke & Bauernschmidt, Reference Karpicke and Bauernschmidt2011; Landauer & Bjork, Reference Landauer, Bjork, Gruneberg, Morris and Sykes1978; Pyc & Rawson, Reference Pyc and Rawson2009). In contrast, equal spacing provides a simpler and more consistent schedule, which may be more effective when the initial learning task is particularly demanding (e.g., Kang et al., Reference Kang, Lindsey, Mozer and Pashler2014). In the present study, given the moderate difficulty Mandarin learners experience in acquiring the English /ɛ/–/æ/ contrast, we predicted that expanding spacing would lead to greater improvement than equal spacing. Previous research suggests that the gains from expanding spacing may yield moderately higher retention—approximately 10–20% more—than equal spacing (Nakata, Reference Nakata2015).

Prediction 3: Aptitude–treatment interaction

Recent research has shown that individual differences in perceptual-cognitive abilities can strongly influence the outcomes of L2 speech instruction. While well-structured, explicit training tends to benefit learners broadly, those with lower aptitude profiles may struggle more under less scaffolded (e.g., Suzukida & Saito, Reference Suzukida and Saito2023 for the null aptitude effects in classroom L2 speech learning), more naturalistic conditions—such as study abroad settings (Saito & Tierney, Reference Saito and Tierney2025), incidental rather than intentional phonetic training (Correia et al., Reference Correia, Rato, Ge, Fernandes, Kachlicka, Saito and Rebuschat2025; Saito et al., Reference Saito, Hanzawa, Petrova, Kachlicka, Suzukida and Tierney2022), or meaning-oriented over form-focused instruction (Ruan & Saito, Reference Ruan and Saito2023; Xu et al., Reference Xu, Saito and Mora-Plaza2024). In the current study, we hypothesized that learners with lower aptitude would make limited gains in the more demanding massed condition, where cognitive load is concentrated and consolidation opportunities are minimal. In contrast, spaced practice—particularly expanding spacing—was expected to mitigate these individual differences, leading to robust gains across aptitude levels.

Methodological implications

In response to R1-R3, the paper aims to provide critical methodological implications regarding whether, to what degree, and how introducing different practice schedules (equal and expanding spacing) can influence and enhance the effectiveness of the same phonetic training compared to traditional massed practice. Given that many existing studies traditionally rely on massed practice schedules, the findings may allow us to reinterpret their potential effectiveness if the distribution of training sessions is adjusted.

Participants

The participants were 60 adult Chinese ESL learners recruited through social media platforms in a specific region of China. Participation was incentivized by advertising the project as a free English phonetic training program, thereby attracting only individuals genuinely interested in the project. This approach was intended to avoid recruiting participants who might otherwise participate solely for monetary incentives and thus complete the training without sufficient concentration. All were over 18 years old, self-reported no hearing impairments, and had received at least 7 years of formal English education as part of China’s compulsory schooling system. According to national curriculum guidelines, this educational background corresponds to CEFR A2–B1 proficiency levels. None of the participants reported extensive study abroad experience. Interested individuals contacted the researcher to enroll. Prior to participation, they received an information sheet and signed a consent form. One participant withdrew before completing the study due to personal reasons, leaving a final sample of 59 participants. The average age was 29.6 (Range = 20–54). On average, they began learning English at age 7.6 (Range = 1–13 years). Participants were randomly assigned to one of three groups: Expanding spacing (n = 19), Equal spacing (n = 20), or Control (Massed practice) (n = 20).

Power analysis of sample size

Importantly, for robust statistics, participants with extremely high scores (>90% accuracy) were excluded at the outset of the project to avoid ceiling effects. This resulted in a final sample of 53 participants for analysis. To estimate the required sample size, a power analysis was conducted using G*Power (Faul, Erdfelder, Lang, & Buchner, Reference Faul, Erdfelder, Lang and Buchner2007). Although Alfotais et al. (Reference Alfotais, Mahdi and Alkhammash2025) reported relatively large effects (d = 2) in their precursor research, we adopted a more conservative approach and set the predicted effect size to “medium,” in line with Yao et al.’s (Reference Yao, He, Chen and Zhu2025) meta-analysis, which suggested that L2 phonetic training typically yields medium effects (d = 0.762). For three groups (massed, equal spacing, and expanding spacing) assessed across three testing points (pre, immediate post, and delayed post), achieving medium effects (f = 0.25) with sufficient power (0.95) required a sample size of 54. Thus, our final sample size (n = 53) was considered adequate. Notably, this was comparable to existing research (n = 49; Alfotais et al., Reference Alfotais, Mahdi and Alkhammash2025).

Setup

The experiment was conducted online via the Gorilla platform (Anwyl-Irvine et al., Reference Anwyl-Irvine, Massonnié, Flitton, Kirkham and Evershed2020), with participants completing all tests and training individually on their own computers and headsets under the guidance of a researcher (a native Mandarin speaker). Several precautionary measures were implemented to ensure that participants fully understood and engaged with the tasks, remained focused, and completed the procedures without distraction. Any participants who showed evidence of deviation were excluded to maintain data quality. First, we explicitly advertised the project as an opportunity to learn English pronunciation for free, ensuring that participants were genuinely motivated by the project’s objective (i.e., improving their L2 English speech proficiency). Second, the investigator arranged an individual videoconferencing meeting with each participant, explained the project’s purpose and procedures in their L1 (Mandarin), and confirmed comprehension. During this meeting, participants also completed an online survey about their EFL experience on Gorilla as a test run and checked the sound level of their headsets. Third, the investigator monitored participants’ progress in each training session via Gorilla. Pilot data collection confirmed that each training session lasted about 10 minutes. A threshold of 15 minutes per session was set for potential elimination; however, no participants exceeded this limit. In sum, we are confident that all participants included in the final analyses (n = 53) completed the tasks with adequate levels of understanding, engagement, and concentration.

Speech stimuli

Lexical items

L2 phonetic training in the current study employed words that contrasted /ɛ/ and /æ/ in a consonant-vowel-consonant structure. A total of 20 lexical items were selected—10 trained items, which appeared in the training and both tests, and 10 untrained items, which appeared only in the tests. This allowed for an assessment of both trained learning and generalization to novel items. The majority of items (19 out of 20) belonged to the first 2,000 word families in the British National Corpus (BNC), based on frequency profiles available via Tom Cobb’s Vocabulary Profiler. Thirteen items were classified as K-1, and six as K-2. The remaining word (lag) was from the K-5 level but followed regular phonological patterns and was therefore deemed appropriate. According to Milton (Reference Milton2009), a vocabulary size of 1,500–2,500 words corresponds to CEFR A2 level, suggesting that the selected items were appropriate for the study’s target participants, all of whom had a minimum estimated proficiency of CEFR A2. A list of the target items was displayed in Table 1.

Table 1. Target Word pairs used during training and testing (trained items) vs. testing only (untrained items)

Pre/post-test measures

To assess the effects of training on learners’ perception of the English vowels /ɛ/ and /æ/, a forced-choice identification test was administered at pre-test, immediate post-test, and delayed post-test. To accommodate participants’ availability and geographical distribution, the tests were delivered online using the Gorilla Experiment Builder platform. During the test, participants completed a two-alternative forced-choice identification task. Using their own devices in a quiet environment, they listened to pairs of minimally contrasting words (e.g., pen vs. pan) that differed only in the target vowel sound. The stimuli included both trained and untrained items (see Table 1), and each item was recorded by 3 native speakers, resulting in 60 audio tokens in total. For each trial, participants heard a single word and were asked to identify what they heard by selecting one of two written options displayed on the screen (see Figure 2).

Figure 2. Sample screenshot of the two-alternative forced-choice identification task interface used in the perception test. On each trial, participants listened to a spoken word and selected the corresponding written form from two options (e.g., said vs. sad). Instructions were presented bilingually in English and Chinese (“Click the word you heard.” / 点击您所听到的单词) to ensure clarity for all participants. Responses were submitted by clicking one of the two options, after which the task automatically advanced to the next item.

Phonetic training

Training setup

Following the pre-test, participants completed six sessions of perceptual training using a high-variability paradigm with trial-by-trial feedback. Each session lasted approximately 10 minutes, amounting to about 1 hour of training in total. In each session, participants identified 20 audio tokens of the 10 trained words (see Table 1), spoken by both a male and a female talker. After each response, participants received immediate visual feedback, followed by a replay of the correct word accompanied by a bilingual message (see Figure 3). This feedback format was informed by Nakata (Reference Nakata2015), who emphasize the importance of corrective feedback and repeated exposure for consolidating difficult phonetic contrasts. To minimize reliance on rote memorization, the order of word presentation was randomized in each session. The six-session structure was based on research suggesting that 2–20 exposures are typically required to establish lexical knowledge (Saragi et al., Reference Saragi1978; Uchihara et al., Reference Uchihara, Webb, Saito and Trofimovich2022; Webb, Reference Webb2007). In the present design, each word was presented at least 12 times across sessions (6 sessions × 2 talkers), providing sufficient repetition to support perceptual learning.

Figure 3. Example screenshots from the phonetic training session. After selecting one of the two word options, participants received immediate visual feedback indicating whether their response was correct (green tick; 3A) or incorrect (red cross; 3B). This was followed by a feedback screen presenting the correct word along with its pronunciation and a bilingual prompt (“The correct word is” / 正确的单词是; 3C). The two-step feedback procedure was designed to enhance phonetic category learning by encouraging accurate form-meaning mapping and reinforcing correct auditory representations.

Auditory processing measures

To assess participants’ individual differences in auditory processing abilities, two types of tests were administered: (a) auditory acuity tasks targeting spectral and temporal discrimination, and (b) auditory–motor integration tasks assessing rhythm and melody reproduction. All tasks were delivered via the Gorilla platform. Written instructions were provided in both English and Mandarin, and participants completed a brief practice trial before each task to ensure comprehension. All test materials are available through SLA Speech Tools (Mora-Plaza, Saito, Suzukida, Dewaele, & Tierney, Reference Mora-Plaza, Saito, Suzukida, Dewaele and Tierney2022), with full methodological details and validation reported in Saito and Tierney (Reference Saito and Tierney2025).