Knowledge of academic collocations is important for academic communication. Examining its relationship with knowledge of academic words and general high-frequency words should provide insights into the relative value of prior knowledge of different kinds of single words for the development of collocational knowledge in academic contexts, thus expanding our understanding of vocabulary knowledge beyond the current conception of individual word knowledge. Yet no studies have addressed this issue, mainly due to the lack of validated tests for academic words and collocations. This study employed three pairs of newly developed tests to measure Vietnamese EFL learners’ knowledge of academic collocations, academic words, and general high-frequency words. Results showed that at the form-recall level, both academic word and general high-frequency word knowledge were positively associated with academic collocation knowledge, with academic words having a stronger relationship. At the form-recognition level, only academic words and academic collocations were positively correlated.

Listening ability in second language (L2) assessment is elicited through test methods that often require other abilities, which may introduce construct-irrelevant variance and threaten the validity of the test. While previous research has examined the effects of item presentation and item format on test performance, studies on test-takers’ cognitive processing remain exploratory. This study investigates the effects of two test method variables, item presentation (operationalized as while-listening performance [WLP] vs. post-listening performance [PLP]) and item format (operationalized as multiple-choice questions [MCQs] vs. open-ended questions [OEQs]), on test-takers’ cognitive processing, using gaze behaviors as real-time indicators. A Graeco-Latin square design was employed to administer four psychometrically validated short-talk listening testlets across the test conditions. Eye-tracking data were collected while controlling for word count and typing speed as covariates. Linear mixed-effects modeling revealed higher total fixation duration, fixation counts, total visit duration, and visit counts in the WLP condition than in the PLP condition. Interaction effects for all four metrics further indicated that these differences were more pronounced for MCQs than for OEQs. Additionally, typing speed and word count contributed to the variance in eye-tracking measures. The findings contribute to our growing understanding of how test methods shape listening processes and offer implications for the design and interpretation of L2 listening assessments.

Over the past decade, there has been growing use of mobile-assisted language learning (MALL) applications such as Duolingo. The effectiveness of MALL applications is thus of great interest among language acquisition researchers and practitioners. This study compared French language development among beginners in three learning conditions: Classroom-Only (n = 58), Duolingo-Only (n = 65), and Classroom + Duolingo (n = 60). The Classroom-Only group completed a standard first-semester curriculum, the Classroom + Duolingo group used Duolingo as supplemental material, and the Duolingo-Only group learned exclusively through the app. All participants completed pretests and posttests measuring overall proficiency, grammar, vocabulary, pragmatics (tu vs. vous), and communicative competence over a 16-week period. Linear mixed-effects models revealed that all three groups showed significant improvement across nearly all measures between pretest and posttest, with similar magnitudes of improvement. The only exception was in learning tu vs. vous (pragmatic competence), where the Classroom + Duolingo group showed larger gains than Classroom-Only and Duolingo-Only groups. Results suggest that both traditional classroom instruction and Duolingo are comparably effective for beginning French language learners. Results are discussed in light of mobile app-based language learning with the potential role of learner characteristics.

Measurement invariance (MI) ensures that a given measure holds the same conceptual meaning for individuals from different groups and across multiple measurement occasions. Structural invariance (SI) is a logical extension of MI that examines whether relationships between latent constructs (e.g., structural paths within the model) hold equally across groups. To examine the status quo of MI and SI in second-language (L2) research, we systematically investigated the extent to which primary studies adhered to best practices in invariance testing and reporting. A total of 4,272 full-text records were screened, and 113 articles (116 independent samples; 147,856 participants) were included. The sample was fully double-coded to ensure accuracy and reliability. The results indicated alarming inconsistencies in how key invariance steps were implemented and reported. We offer empirically grounded recommendations for (a) improving methodological rigor of invariance assessments in the field and (b) contributing to more equitable and interpretable comparisons in multilingual settings.

Regardless of whether one is analyzing quantitative data from research involving generative artificial intelligence (GenAI) or more classical methods, testing for normality remains a necessary step in statistical analysis. Although over 60 methods have been proposed for assessing univariate normality, previous systematic reviews show that normality testing remains underreported in L2 research. This paper addresses this gap by first reviewing the concept of normality and its role in parametric statistical inference. We then examine 12 normality assessment methods including five graphical and seven analytical methods selected based on their prominence in statistical literature and availability in commonly used software. Each method is explained in terms of its underlying mechanism and sensitivity to specific forms of nonnormality, such as skewness, tail heaviness, and multimodality. In the second part of the study, we review 237 empirical articles published between 2020 and 2025 in ten selected L2-focused Q1 journals, using AI-assisted annotation. Our findings reveal inconsistencies in how graphical tests are reported, a tendency to rely on tests such as the Kolmogorov-Smirnov without explicit attention to sample size constraints, and limited justification provided for critical values of skewness and kurtosis. These results indicate some divergence between recommended statistical practices and the procedures for normality testing reported in the L2 publications examined. The paper concludes with actionable recommendations for selecting and interpreting normality tests in L2 research contexts.

This meta-analysis investigates the contributions of viewing audiovisual input on second language (L2) learning. We calculated 75 effect sizes from 56 experiments (n = 1954) and assessed the effects of audiovisual input on language learning using a within-group (pre-post) meta-analytic approach. Fifteen moderator variables were included in the analysis. Results showed that a) there was a small effect of audiovisual input on L2 learning (g = .89); b) no differences were found between the effects of viewing audiovisual input on different areas of L2 learning (vocabulary, grammar, pronunciation, speaking, listening proficiency); and c) video category had a significant impact on L2 learning with entertainment-focused videos (e.g., TV series, movies, and mixed videos) yielding lower effects than educational videos (e.g., TED Talks, documentaries, and language-focused).

The present study investigated how initial proficiency impacted development in adolescent learners’ L2 speaking. The study reports on a longitudinal study with dense measurements with twelve adolescent L2 English learners. The participants were tested every week throughout their first year of secondary school (30 datapoints). The participants’ learning trajectory was modeled using generalized additive mixed models (GAMMs). Results showed large differences in development between various lexical and syntactic measures and between individual learners. The learners’ initial proficiency had a significant impact on their development. Overall, more periods of growth were observed in lexical measures than in syntactic measures. In line with previous results, some stabilization was observed once learners reached a certain proficiency level, but this stabilization was dependent on task type. Closed tasks lead to a ceiling effect in some measures, whereas more open tasks give learners from various proficiency levels opportunities to demonstrate their L2 English speaking skills.

Structural equation modeling (SEM) is a powerful and flexible modeling framework for testing complex relationships among observed and latent variables. However, its methodological complexity and analytical flexibility also increase the risk of questionable research practices (QRPs), especially in fields like applied linguistics, where training in advanced statistics may be limited. This article synthesizes the literature on QRPs and applies it to SEM by identifying seven categories of problematic practices: not checking assumptions, not validating a measurement model, not testing competing models, not sufficiently justifying modeling decisions, relying on post hoc model modification, overemphasizing global fit indices, and incomplete or nontransparent reporting. Each practice is described with examples and linked to broader issues in research ethics and transparency. The paper concludes with concrete recommendations for improving the credibility and reproducibility of SEM research, emphasizing the integration of best practices with the principles of open science.

Second language (L2) pronunciation research has measured speech comprehensibility by asking listeners to assess L2 learners’ speaking performance with rating scales. While some studies have provided validity evidence for these rating scales, few studies have examined the extent to which those scales effectively distinguish among L2 speakers. To fill this gap, the present study examines the 9-point scale used in Saito et al. (2020: Annual Review of Applied Linguistics, 40, 9–25.) and the 100-point scale in Huensch and Nagle (2023: Studies in Second Language Acquisition, 45(2), 571–585.) from a Rasch measurement perspective and showcases post hoc score category collapsing as a potential countermeasure against suboptimal rating scale functioning. Findings suggested that different score categories represented the same ability level and were therefore interchangeable. Collapsing these score categories yielded shorter but more functional scales without compromising the psychometric qualities of the original scales. These findings suggest that researchers need to empirically refine their scale lengths rather than uncritically following their conventional measurement practices.

This study examined the impact of video-based synchronous computer-mediated communication (SvCMC) on engagement in collaborative pre-task planning and L2 content used in task performance. One hundred twenty-eight Hong Kong learners of English were assigned to either a face-to-face (FTF) group (n = 64) or an SvCMC group (n = 64), where they completed planning for a monologic task. Based on Philp and Duchesne’s multidimensional model, planning was analyzed for cognitive (negotiation of meaning, semantically engaged talk), social (affiliative responses) and emotional engagement (enjoyment, anxiety), and task performance was coded for L2 content. Results showed that FTF mode led to significantly lower anxiety, more semantically engaged talk, and more affiliative responses, as well as conceptually richer task performances. Use of planned content was predicted by L2 proficiency and semantically engaged talk and negatively predicted by anxiety. Our findings contribute to an understanding of engagement in SvCMC and FTF modes and their impact on L2 learning.

Recent research has shown that adult learners can rapidly acquire novel words of a foreign language by tracking cross-situational statistics, but learning is substantially reduced when the target words are phonologically similar and contain non-native contrasts. We expand on this research by investigating whether perceptual discrimination training on non-native target contrasts facilitates cross-situational learning of new words (CSWL). Our design combines perceptual training and CSWL to test the transfer of perceptual gains to lexical learning—an approach that integrates methods from L2 speech and statistical learning. In two studies, we tested English-native and Portuguese-native speakers’ learning of 24 Portuguese pseudowords via a CSWL task. In Study 1, we examined baseline learning in both language groups without prior training. In Study 2, English-native speakers were assigned to one of three conditions: phonetic training with an AX discrimination task, phonetic training with an oddity discrimination task, or no phonetic training prior to the CSWL task. Results confirmed that adults can learn non-native words from cross-situational statistics, and that phonological overlap between words decreases learning. Perceptual training improved the discrimination of target contrasts, but this did not transfer to statistical learning of words that contain these contrasts. These findings suggest that phonetic training alone may not be sufficient for vocabulary acquisition, suggesting the need for instructional approaches that integrate phonetic training with more explicit teaching methods or meaning-based practice.

Many adults learn languages with written forms that differ from their first language(s). Empirical research has demonstrated the influential role of written input on developing L2 phonology. However, existing studies are limited by (1) focusing on learning languages that share the same orthographic script, predominantly the Latin alphabet, (2) small sample sizes, and (3) limited consideration of L2 proficiency. This study investigated the influence of Arabic and English written input when lexically encoding the difficult /f-v/ phonological contrast for L1 Arabic-speaking learners of L2 English. A word learning study was completed by 114 L1 Arabic speakers, with varying English proficiency, and 117 L1 English-speaking controls. Mixed-effects modeling of L1 Arabic accuracy revealed an inhibitory effect of any written input when learning words differing by the difficult contrast. Performance improved with increasing L2 proficiency; however, the inhibitory effect of written input for words differing by /f-v/ persisted into high levels of L2 proficiency.

Research on how cognitive and affective factors shape bilinguals’ moral judgments in their first (L1) and second (L2) languages remains limited. This study advances our understanding of both language-related variables (L2 proficiency, age of onset of L2 acquisition, length of L2 immersion) and nonlinguistic variables (emotional intelligence, executive functions) in relation to the moral judgments and emotional intensity experienced by 90 Chinese–English bilingual speakers. Participants read five moral dilemmas and completed a set of questionnaires and cognitive ability tests. Mixed-effects models revealed no evidence of a moral foreign language effect, raising questions about the robustness of this phenomenon. Specifically, neither language-related variables nor emotional intelligence had a significant effect. However, participants with stronger updating ability made more deontological judgments in both their L1 and L2, while those with better inhibitory control also made more deontological decisions, but only in their L1. These findings offer new insights into the relationship between language and cognition.

This study explored the effects of repeating a video-lecture-based task on second language (L2) learners’ input processing and the relationship between online processing patterns and lecture comprehension. Participants were randomly assigned to one of three groups. The comparison group (n = 30) performed the task once, whereas the repetition group (n = 30) repeated the same task three times, with each group completing a free-recall test after their last performance. The stimulated-recall participants (n = 15) completed the task once, twice, or three times and described their thought processes during their last task performance. The lecture featured an instructor introducing fundamental concepts of neurobiology with labeled diagrams. Participants’ visual attention to the instructor and diagrams was captured using an eye-tracker. Results revealed an increase in learners’ visual attention to the instructor and a decline in their visual attention to the diagrams across repetitions. Additionally, more visual attention to the instructor was related to lower comprehension.

This methodological study investigated how the distribution of training sessions—massed, equal spacing, and expanding spacing—affects L2 phonetic learning, focusing on Mandarin-speaking learners’ perception of the English /ɛ/–/æ/ contrast. While most previous phonetic training studies have used massed schedules, the current quasi-experimental design revealed that both types of spaced practice significantly outperformed massed practice in terms of immediate gains and long-term retention. Effect sizes in the spaced groups were approximately double those of the massed group. No significant differences emerged between equal and expanding spacing. These findings suggest that distributed practice—regardless of spacing type—can enhance both the magnitude and durability of L2 phonetic learning. Crucially, this study makes it possible to revisit past findings based on massed training paradigms and to consider whether adopting alternative timing schedules could unlock greater learning potential—for instance, by doubling the size and durability of training effects through the use of spaced conditions.

This study aims to illuminate the underlying mechanisms of sentence processing in L2 speakers. The phenomenon of interest in the study is the passive structure, which prior research has shown can be challenging for both L1 speakers and L2 speakers to process compared to active structures. Using a visual-world eye-tracking paradigm, this study investigates whether L1-English speakers and L1-Cantonese L2-English speakers employ a morphological cue within the verb to process English actives and passives, and if so, specifically when these cues are integrated into their processing. The results from a growth curve analysis and a divergence point analysis show that the L2-English speakers were slower than the L1-English speakers, but did use the morphological cue to process both actives and passives, even though this cue is absent in their L1 Cantonese. These results suggest that, despite differences in processing speed, the mechanisms underlying L1 and L2 processing are similar.

This study investigates whether the second language (L2) development of d/Deaf and hard of hearing (d/DHH) learners in comparison to the L2 development of hearing learners, based on the processability theory (PT) developed by Pienemann (1998, 2005) in the field of second language acquisition (SLA). Thirty-eight d/DHH and 32 hearing secondary school learners of English as a foreign language (EFL) completed a series of speaking tasks designed to elicit specific morphosyntactic structures that, according to PT, align with L2 developmental stages. Implicational scaling revealed that although d/DHH learners followed a similar developmental sequence to their peers, they displayed a noticeable delay. They also appeared to require additional time and practice to fully produce the entire target structures after reaching certain developmental stages. The results provide stronger empirical evidence for d/DHH learners’ L2 development, supporting the qualitative similarity hypothesis in the field of special education, which posits that despite quantitative delays, d/DHH learners exhibit qualitatively similar L2 development to their hearing peers.

The role of morphology in complex word acquisition was examined in Chinese (L1)–English (L2) bilinguals. Participants learned words consisting of two novel constituents, by pairing them with pictures. Items either belonged to large (torbnel, torbilm, torbla, torbiph) or small morphological families (torbilm, torbla). After training, participants completed recognition and spelling tasks with novel words that either included or excluded a trained morpheme. Results revealed robust stem-training effects, showing that items including a trained constituent were harder to reject and easier to spell than items including two untrained constituents. There was also a significant effect of morphological family size, with greater training effects for items belonging to large than small families. Effect sizes were overall smaller in L2 than in L1. These findings point to the important role of morphological structure in L2 word acquisition and suggest that large morphological family-clusters lead to better learning outcomes.