Background. AI-hallucination poses a threat to second language (L2) scientific writing because of misinformation produced by generative AI (GAI) tools, which affects students’ criticality and self-efficacy in producing quality writing. Human-AI collaborative writing may address this problem, yet few studies have examined its role in fostering students’ awareness of AI-hallucination, particularly within an integrated framework involving critical thinking and self-efficacy. Purpose. To examine the effectiveness of a human-AI collaborative writing strategy in improving students’ awareness of AI-hallucination, critical thinking skills, and self-efficacy in producing valid and accurate content for scientific writing. Method. This study used a quasi-experimental design combining quantitative and qualitative approaches. Thirty-five students in two intact L2 scientific writing classes were assigned to the control (n = 17) and experimental (n = 18) conditions. The control class received unscaffolded AI-assisted instruction, and the experimental class was taught with the human-AI collaborative writing strategy. Data were collected using the Awareness of AI-Hallucination Scale (AAHS), Facione’s Holistic Critical Thinking Scoring Rubric (HCTSR), the Self-Efficacy Scale for Content Validity and Accuracy (SESCoVA), and a semi-structured interview guide. Results. Under the conditions of this study, post-test scores were significantly higher in the experimental class than in the control class for awareness of AI-hallucination (t(33) = −27.76, p < .001, η² = .96), critical thinking skills (t(33) = −5.28, p < .001, η² = .46), and self-efficacy in producing valid and accurate content for scientific writing (t(33) = −12.44, p < .001, η² = .82). Interviewed students reported that the strategy gave them practice in searching for, analyzing, and verifying GAI-generated content to ensure its validity and accuracy. Conclusion. The human-AI collaborative writing strategy, extending prior work, appears to be associated with greater awareness of AI-hallucination, stronger critical thinking skills, and higher self-efficacy among L2 students. It may be useful in L2 scientific writing courses that aim to scaffold students’ critical verification of AI-generated content. In addition, the study contributes two preliminary instruments, the AAHS and SESCoVA, for measuring awareness of AI-hallucination and self-efficacy in L2 scientific writing.
Background. Experimental evidence that cohesion facilitates comprehension comes almost entirely from English and shows that the effect depends on readers' age and prior knowledge. Research on Russian cohesion remains largely descriptive: its effect on schoolchildren's comprehension of instructional texts has not been tested experimentally, leaving textbook design and computational models of text complexity without validated cohesion indicators. Purpose. To test whether increased cohesion improves 11–12-year-olds' comprehension of a Russian instructional text, and to compare recall of main ideas and supporting details from low- and high-cohesion versions of the same text. Method. Of 118 fifth-graders, 65 with mid-range scores (13–16) on the WISC General Knowledge subtest were selected to control for background knowledge. Group A (n = 31) read an unaltered 211-word textbook excerpt; Group B (n = 34) read a version revised for cohesion (e.g., content word overlap, anaphora, determiners, temporal markers, connectives, sentence splitting) with identical propositional content. Comprehension was measured by immediate recall, scored by two independent experts against the text's 143 propositions, and by a six-item cloze test. Result. Readers of the high-cohesion version recalled 40.6% of propositions versus 30.1% for the original version, about one-third more; the difference was significant for main ideas, details and total recall (p ≤ 0.003). Gains were nearly identical for main ideas (44.1% vs. 32.3%) and details (39.3% vs. 27.7%), so cohesion raised recall without altering the advantage of main ideas over details. Cloze scores showed the same trend (3.0 vs. 2.4 of 6 items) but did not reach significance (p = 0.073). Conclusion. Making referential and connective links explicit, without adding content, substantially improved recall, indicating that part of the difficulty of Russian instructional texts lies in how their content is linked rather than in the content itself. Because the devices were introduced jointly and tested on one text with immediate recall, the effect cannot yet be attributed to individual devices or assumed to persist. Cohesion revision is therefore a promising low-cost tool for textbook authors and test developers; studies isolating individual devices across genres, ages and delayed recall should determine which matter most.
Introduction. The concept of “academic culture” has explanatory value in higher education research, yet its analytical status remains uncertain. Existing typologies describe other objects: organisational typologies classify the university as an institution, and disciplinary typologies describe fields of knowledge, whereas the normative order of the academic profession, here treated as academic culture, lacks a typology of its own. Purpose. To substantiate a metacultural framework for analysing academic culture and to identify the regulative dimensions through which stable and misaligned configurations of the academic profession can be described. Method. The study is conceptual and relies on ideal-typical analysis. Academic culture is treated not as a separate entity alongside organisational, research and learning cultures but at an analytical level that examines the rules of normative interrelation between the organisational order, the order of knowledge production and the order of learning. The regulative dimensions are relatively autonomous from institutional, disciplinary and national contexts but do not prescribe the content of academic culture. Results. The analysis substantiates four regulative dimensions of academic culture: the axiology of knowledge, the regime of professional legitimation, the locus of professional identity and the normative coordination of academic activities. Their combinations yield a typology of epistemic-disciplinary, formative-pedagogical, public-orientated and pragmatic-applied configurations. Normative coherence is defined by a single dominant orientation across all dimensions, as well as by a stable principle that makes the relation between the value of academic work, its recognition, professional identity and the structure of activities intelligible. Two forms of normative misalignment are distinguished: (1) normative colonisation, in which an external evaluative logic displaces the substantive foundations of academic work; (2) normative disarticulation, in which no shared principle links values, recognition criteria, identities and academic activities. Conclusions. The framework shifts the analysis of academic culture from listing its values and the conditions of its formation to the normative coherence of the academic profession. It does not replace research on organisational,disciplinary, research and learning cultures but allows analysis of how they are aligned, hybridised, held in tension or misaligned, and prepares the ground for empirical operationalisation and configurational analysis across national and institutional contexts.
Background. The increasing use of generative artificial intelligence (GenAI) in professional communication raises a new question for English for Specific Purposes (ESP) pedagogy: how should learners specify the linguistic and pragmatic conditions under which AI-generated texts are produced? Prompt engineering can be examined as a form of applied-pragmatic activity, yet the sociopragmatic content of learners’ prompts has received limited direct assessment. Purpose. This study aims to develop and preliminarily evaluate the Pragmatic Constraint Index (PCI), a rubric for coding how explicitly learners’ prompts specify pragmatic and genre-related conditions, and to describe how management ESP students’ prompts differ before and after explicit instruction. The construct examined is prompt-focused specification of pragmatic conditions, not the acquisition of autonomous pragmatic competence. Method. Forty-two management students in a C1-targeted ESP course at one Russian university completed two business correspondence tasks with GPT-5.2 in a one-group pre-test/post-test study without a control group; the two tasks used different scenarios. A two-week instructional intervention introduced the P.A.R.T. framework (Persona, Act, Register, Tone) and taught students to express metalinguistic directives. The 84 prompt chains (42 pre-test, 42 post-test; iterative prompts concatenated) were coded with the exploratory PCI (four categories, 0–12 points) and analyzed for word count, lexical density, and syntactic complexity; paired comparisons used Wilcoxon signed-rank tests with Holm correction. In a secondary analysis, two experts who were not told the phase or the PCI scores rated 20 AI-generated business messages (10 pre-test, 10 post-test) written by 20 different students. Model-level reproducibility is not claimed. Results. Median prompt length increased from 23 to 85 words and median PCI from 2 to 10 points (Z = 5.65, Holm-adjusted p < .001, r = .87). The metalinguistic-directives category showed the largest mean increase (0.1 to 2.5). In the 20-output subsample, post-test messages received higher expert ratings for mitigation of face-threatening acts (U = 5.0, Holm-adjusted p = .002, r = .76). Pooled across the two phases, PCI and composite output ratings were correlated (Spearman’s r_s = .78), but neither within-phase correlation was statistically significant (pre-test r_s = .38; post-test r_s = .44; n = 10 each); the pooled coefficient therefore largely reflects the difference between phases and is not evidence of independent predictive validity. Qualitative examples showed more explicit sociopragmatic instructions in post-test prompts, alongside chains containing conflicting directives. Conclusion. After instruction, students produced longer prompts with more explicit metalinguistic and genre-related constraints, which indicates observed uptake of the taught directives. The single-group design, the different test scenarios, and the immediate post-test do not allow causal or longitudinal claims, and observed uptake of taught directives should not be equated with demonstrated sociopragmatic competence. The findings provide preliminary evidence that explicit prompting can be taught and coded in management ESP; the PCI is best regarded as a coding rubric for observable instructions rather than a validated measure. The proposed ‘Editor-in-Chief’ assessment idea remains a proposal for future testing with controlled designs and delayed transfer tasks.
Introduction. Academic-language support in teacher education may be distributed across courses, thesis work, and teaching practice without an explicit account of how these components connect. Little is known about whether the connections within such distributed provision are explicit in programme documents and recognisable to stakeholders. Purpose. To examine how visible academic-language provision was in one anonymised Hungarian English teacher education programme, and how participants reported using it, in order to identify where connections among courses, thesis work, and teaching practice need to be made explicit; and to develop and apply a case-based procedure for evaluating such distributed provision. Method. A mixed-methods case study of provision during 2020–2024 combined 12 central documents, questionnaire responses from 90 trainees and graduates, and 21 interviews across four stakeholder groups. One interview conducted in July 2025 contributed retrospective evidence only. Twenty-two questionnaire items formed four case-specific descriptive indices. Documents, questionnaire responses, and interviews were analysed separately and integrated through a joint display, with greater interpretive weight assigned to documents and interviews. Results. The 12 central documents identified provision across courses, thesis requirements, and practice but did not present these components as one developmental pathway. Questionnaire index means were close to the scale midpoint (2.98–3.09 on a 1–5 scale) and concealed heterogeneous responses: across the four indices, 38.9%–47.8% of respondents scored below and 42.2%–50.0% above the midpoint. Interview accounts linked usable support to models, explicit criteria, guided practice, revision, and feedback, while also identifying inconsistent access to these opportunities. Academic contribution was more clearly articulated in interviews than transfer to school teaching; no Product item had more than 36.7% of responses in the two highest categories. Conclusion. The case identifies a problem of making connections within formally distributed provision explicit, rather than demonstrating the absence of support. A document register with explicit evidence-status categories and source-specific integration can guide further evaluation, but neither programme effectiveness nor professional transfer was established.
