Shungo Suzuki

This study proposes and demonstrates a computer-assisted qualitative approach for examining L2 collocational development through within-learner comparisons in longitudinal learner corpus data. Previous corpus-based research has documented developmental tendencies through aggregate quantitative measures such as frequency and association strength, but these measures are less suited to examining how change is realized in individual learners’ expression choices. Using longitudinal spoken learner English corpus data from Japanese secondary school learners at two time points separated by three years, the study focuses on verb–object combinations as a manageable analytic window. Verb–object pairs were automatically extracted from repeated-task performances, and candidate replacement cases were identified through same-learner, same-task, and shared-object comparison across time. Script-assisted processing and spreadsheet filtering reduced 7,268 extracted dependency pairs to 451 candidate replacement cases, of which 121 were retained after human qualitative review for contextual comparability. The validated cases showed multiple forms of within-learner qualitative change, including changes in targetlikeness, verb specificity, compactness, subject-related reformulation, and expanded- and reduced-content reformulation. A recurrent pattern was also identified in cases derived from [not] eat breakfast, where the negative meaning expressed in [not] eat was re-expressed through skip, alongside variation in the object phrase and a gerund-complement realization. The main contribution is methodological: the study demonstrates the value of linking automatic candidate identification with human contextual validation and qualitative annotation to produce contextually comparable replacement cases for analyzing within-learner collocational change. It also clarifies challenges related to extraction scope, replacement-based comparison, qualitative coding, and corpus comparability.

This study examined how the location of silent pausing (mid- vs. end-clause) may relate to neural processes during spontaneous L1 and L2 speech production. Twenty intermediate-level Japanese speakers of L2 English carried out eight monologic, oral decision-making tasks, four in English and four in Japanese. While completing the tasks, their brain activity was recorded through fMRI scanning. Participants’ speech was transcribed, and the resulting transcripts were annotated for mid- and end-clause silent pauses. Then, for the pauses identified, we conducted whole-brain analyses to identify relevant activation patterns, followed by region-of-interest analyses in language-related areas. We found that mid-clause pauses were linked to increased activation in language-related brain regions, with stronger effects for L2 speech. We also observed heightened activity in conceptualisation-related brain areas at end-clause positions in both L1 and L2 speech. In L2 English, participants also displayed greater activation in a concept-retrieval-related brain region in end-clause than in mid-clause position.

This study investigates the nature of co‐construction in roleplays conducted with human versus AI interlocutors for assessing interactional competence (IC) in L2 English. Seventy‐five university students in Japan completed roleplay tasks with both human tutors and an AI agent. The AI agent is a multimodal dialog system integrated with a large language model (LLM), designed to allow synchronous interaction with the participant through autonomous turn‐taking. Using conversation analysis, 24 interactions were analyzed to investigate how participants managed preference organization, sequence expansion, and turn‐taking. The analysis revealed that the AI‐delivered roleplays elicited some IC‐relevant practices and that participants treated the roleplay as a co‐constructed interaction, responding contingently to the AI's contributions. While the data suggested both human and AI interlocutors maintained mutual understanding, striking differences in turn‐taking practices were observed, including more frequent overlaps and inter‐turn gaps in the AI‐delivered condition. The study concludes that LLM‐integrated multimodal dialog systems, by producing recognizable verbal actions and multimodal signals, have the potential to effectively elicit co‐constructed interactional performances relevant to IC assessment.

The current study proposes a new approach to weakness identification in diagnostic language assessment (DLA) for speaking skills. We also propose to design actionable and contextualised diagnostic feedback through the systematic integration of feedback and remedial learning activities. Focusing on lexical use in second language speaking, the current study developed and validated our DLA programme in terms of actual learning gains, using an experimental design. A total of 59 beginner-to-intermediate-level Japanese learners of English were randomly assigned to control or experimental groups. While both groups engaged in task repetition with a conversational artificial intelligence (AI) agent on six occasions, only the experimental group received the diagnostic feedback on lexical use including the paraphrased utterances of their original utterance. The results showed that the control group (task repetition only) demonstrated significant improvement during the task repetition sessions but failed to transfer and retain the learning gains. In contrast, despite the lack of practice effects, the experimental group (task repetition with diagnostic feedback) outperformed the control group at the posttest with a near-medium effect size. A qualitative investigation into learners’ perceptions further confirmed that the proposed contextualised diagnostic feedback succeeded in heightening their awareness of weaknesses.

This study explored how second language (L2) speakers’ use of multiword sequences in speech predicted perceived fluency ratings while controlling for their utterance fluency. A total of 102 Japanese speakers of English delivered an argumentative speech, which was analyzed for bigram and trigram measures (frequency, proportion, and mutual information) and utterance fluency measures capturing three subdimensions: speed, breakdown, and repair fluency (Tavakoli & Skehan, 2005). Perceived fluency was assessed by 10 experienced L2 raters. Mixed-effects regression analyses revealed that after establishing the parsimonious model solely by UF predictors (marginal R² = .61), a frequency-based n-gram predictor––bigram proportion––slightly but significantly explained the remaining variance of fluency rating scores (0.8%). The results indicated that multiword sequences in speech had a small but systematic impact on perceived fluency, even controlling for the effects of utterance fluency. This finding contributes to the discussion concerning the role of multiword sequences in fluent speech production.

Second language researchers have recently recognized the importance of flow experience as an optimal state for learning, but still little is known about the necessary conditions for flow and its longitudinal change and impact on language learning outcomes. This study tracked the changes of 329 Japanese university students’ flow experience and oral proficiency in a semester-long English as a foreign language (EFL) conversation course. A flow questionnaire was administered four times during the semester, while their oral proficiency was assessed by an AI-based interactional speaking test at the beginning and end of the semester. A series of longitudinal analyses using multilevel modeling showed a U-shaped trajectory of flow experience. Furthermore, it was found that flow experience was predicted from the learner-perceived balance between task challenge and skills, concurring with Csikszentmihalyi’s (1975/2000, 1990) flow theory. In addition, flow experience contributed to improved oral proficiency scores, providing further positive evidence for the theory.

Second language (L2) speaking research has highlighted intra-speaker variability of fluency performance across tasks. To better understand such task effects on fluency, the framework of speech processing demands has been proposed as a systematic approach to relating task characteristics to L2 speech production mechanisms and the limited capacity of attentional resources (Skehan 2009. Modelling second language performance: Integrating complexity, accuracy, fluency, and lexis. Applied Linguistics 30(4). 510–532, 2014. Limited attentional capacity, second language performance, and task-based pedagogy. In Peter Skehan (ed.), Processing perspectives on task performance, 211–260. Amsterdam: John Benjamins). However, the framework has been tested on a limited range of task types, using carefully designed experimental tasks. For the sake of the ecological validity of findings, the current study thus further explores how L2 learners’ fluency varies across four spontaneous speaking tasks differing in their processing demands. A total of 128 Japanese learners of English completed four speaking tasks: Argumentative task, Picture narrative task, Reading-to-Speaking task, and Reading-while-listening-to-speaking task. Their speech was analysed in terms of speed, breakdown, and repair fluency and was compared across tasks. The results of Generalised Linear Mixed-effects Modelling showed that conceptualising demands were reflected in the frequency of filled pauses, while formulation demands were associated with the articulation rate, mid-clause pause ratio, and mid-clause pause duration. These findings unveil the interrelationship between task characteristics, fluency measures, and how learners approach tasks.