Building Knowledge of Generative AI in Applied Linguistics
2025 Technology for Second Language Learning Conference
|
 |
Abstracts
(Chronologically ordered by presentation day)
| Wednesday, Nov. 5th ⎋ | Thursday, Nov. 6th ⎋ | Friday, Nov. 7th ⎋ | Saturday, Nov. 8th ⎋ |
Wednesday, November 5th
11-11:30am | AI Usage Dependency and Pattern in English Language Learning among University Students (TSLL25-82)
Naiem Al Amin 
(State University of Bangladesh, Bangladesh🇧🇩)
The rapid integration of Artificial Intelligence (AI) into education has reshaped English language learning, particularly among university students. Although AI tools are increasingly used in English language learning, there remains a lack of in-depth understanding regarding the specific types of language-related information students seek, their motivations for favoring AI over traditional resources, and the extent of their reliance on these tools. This study addresses this gap by investigating the motivations, behaviors, and patterns of AI usage among Bangladeshi public university students in their English learning journey. Grounded in Wilson’s Information Seeking Behavior (ISB) theory, the research employs a mixed-methods approach combining surveys from students and in-depth interviews with participants. Results reveal that AI tools are primarily used for academic writing, grammar clarification, vocabulary development, and real-time support. Students cite accessibility, efficiency, and structured responses as major reasons for choosing AI. However, concerns persist regarding the accuracy, credibility, and contextual relevance of AI-generated content. The findings suggest a rising dependence on AI tools, though tempered by critical awareness of their limitations. This study highlights the urgent need for improving the reliability of AI outputs and underscores the importance of fostering AI literacy to ensure meaningful and responsible integration into English language education.
11:30-Noon |GenAI for Pre-Service English Language Teachers in EFL Contexts: Review of Empirical Research Since 2022 (TSLL25-91)
Presentation withdrawn
1-1:30pm | Mitigating Over-Reliance on Generative AI Language Models: Strengthening Critical Thinking through Language Classrooms (TSLL25-148)
Mythiri B. (Xavier Institute of Management & Entrepreneurship, India🇮🇳)
This study explores the over-reliance on generative AI language model (GenAI) and the decline in learner engagement brought on by the ease of access to information, which may hinder students’ capacity for critical thinking and creativity. Grounded in Paul and Elder’s (2006) framework for critical thinking, the research investigates the feasibility of implementing a pedagogy that supports self-regulated learning while using GenAI as a tool to enhance, rather than replace, student input. The study was conducted with approximately 75 students enrolled in a Post Graduate Diploma in Management (PGDM) program, for whom business communication is a core course component. Although these students received their entire formal education in English many continued to struggle with articulating original ideas in academic writing. This linguistic limitation often led to passive dependence on AI-generated content, used as a substitute for cognitive effort rather than a support mechanism. However, when integrated within a critical thinking framework, GenAI served as both a linguistic and cognitive scaffold, helping students gradually move from surface-level content replication to more reflective, evaluative, and independent speaking and writing practices. Data was collected through interviews with both students and educators, exploring how students performed and how AI over-reliance was affecting their academic development. Thematic analysis was employed to identify patterns in responses and behavior. Findings indicate a shift from passive to more strategic use of AI tools, with students beginning to critically assess, adapt, and supplement AI outputs with their own reasoning. This study contributes to the discourse on AI in higher education by demonstrating that a structured pedagogical approach can mitigate overreliance on GenAI and promote sustained critical thinking skills.
1:30-2pm | Generative AI and the automation of language teaching: Implications for language-teacher identity construction (TSLL25-159)
Dmitri Detwyler , Sydney Stone , Kadhir Profit , & Hyunhwa Kim (Georgia State University🇺🇸)
The wave of generative AI hype since the emergence of ChatGPT in 2023 has also swept through the fields of applied linguistics and language teaching. Attempts to understand generative AI systems as straightforward “tools” for linguistic applications are fraught with power relations (McKnight and Shipp, 2024), and often neglect that such systems also have agency (Godwin-Jones, 2024) in relation to their human users—distinct from the emergence of specialized LLMs also called “AI agents.” The prospects for automating various domains of teaching activity are now being mapped out (Bond et al., 2024), and the corresponding expansion of teacher capabilities has been theorized with new metaphors like the “centaur” (Fassbender, 2024) that can go farther and faster. Language-teacher perspectives on these developments show a mix of excitement and existential concern (Barnes and Tour, 2025). What has become clear is that generative AI systems are shifting language-teacher knowledge, practices, and identities (Ghiasvand and Seyri, 2025). Yet language teachers themselves remain cagey about their use of such systems (Barnes and Tour, 2025), suggesting unresolved tensions with respect to professional identity. Our study in progress contributes to the emerging literature about the evolution of what it means to be a language teacher in the era of generative AI. We use survey data, interviews, and think-aloud protocols with language teachers to explore AI-assisted lesson planning as an example case of teaching activity increasingly exposed to automation. Preliminary results suggest that new forms of technical expertise (Mishra et al., 2023) and critical AI literacy (Leander and Burriss, 2020), while important, may not be sufficient for identity construction when language teaching itself is increasingly automated and de-socialized (Selwyn et al., 2023). This work charts a course for educational technologists, language-teacher educators, and language teachers themselves to collaboratively question the “presumptions and promises” (Selwyn et al., 20223, p. 16) of educational automation in language teaching.
2-2:30pm | Research Integrity Challenges of Commercial GenAI Services in Applied Linguistics (TSLL25-146)
Emma Caputo (University of Barcelona, Spain🇪🇸)
This presentation addresses critical methodological issues arising from commercial generative AI (GenAI) services in applied linguistics research. As these technologies become increasingly common in language assessment, classroom studies, and linguistic analysis, they introduce challenges to research integrity that require technical understanding and examination. We examine how commercial GenAI services operate as a “Service as a Software Substitute” (SaaSS), with computational processes occurring on third-party servers beyond researcher control. This external computing model creates fundamental methodological problems: proprietary algorithms and inaccessible training data prevent reproducibility, while unannounced system updates and unpredictable model behaviors compromise experimental consistency. These limitations undermine researchers’ ability to verify computational processes or ensure reliable analytical methods. Additionally, transferring participant data to third parties with unclear retention policies raises confidentiality concerns. Based on our development of free and open-source GenAI research tools, we demonstrate practical alternatives to common commercial services. These locally deployable solutions restore computational control to researchers, ensuring both methodological transparency and data sovereignty. By implementing version-controlled research environments, these approaches maintain research integrity while keeping participant data private. This research contributes to applied linguistics by analyzing commercial GenAI services’ black-box limitations and providing principled alternatives. We offer researchers concrete approaches for ethically integrating GenAI technologies in language teaching and assessment while advancing scientific standards through free and open-source solutions that restore computational transparency and research integrity.
2:45-5:30pm | [Poster 1] Ethical Considerations for the Use of ChatGPT in the Algerian Higher Education Setting
Abdelhamid Djeghoubbi (Pázmány Péter Catholic University, Hungary🇭🇺)
Adopting artificial intelligence (AI) tools like ChatGPT in higher education raises ethical, privacy, and academic integrity concerns (Eden et al., 2024). This is a replication study of Huallpa’s (2023) research, which sought to investigate the prospect and challenges of using ChatGPT in higher education. The present study focuses on the English Department at Kasdi Merbah University in Ouargla, Algeria. It seeks to gather perspectives and opinions from university teachers and students on the use of ChatGPT in their educational processes, emphasising ethical considerations and its possible impact on the integrity of academia. This study uses a mixed-methods approach within the pragmatism paradigm. 47 participants from the English Department, using random sampling, participated in this study. Data were collected through a structured questionnaire consisting of both open and closed-ended questions and analysed through descriptive statistics and thematic analysis. The demographic findings indicate a young participant base, mostly female, with different educational levels. Key findings include concerns over AI leading to academic misconduct, with 85percent of the participants acknowledging its capacity to improve learning if used properly. Data elicited from the open-ended questions emphasise the need for universities to establish policies and ethical guidelines that regulate the use of AI technologies like ChatGPT. This study suggests a balanced approach that makes use of AI to support rather than replace human interaction in classrooms.
2:45-5:30pm | [Poster 2] The Impact of Kuki Chatbot on Developing EFL Learners’ Speaking Skills at El Oued University
Aissa Berregui & Mohammed Naoua (University of El Oued, Algeria🇩🇿)
The remarkable advancement in digital communication technologies has led to the emergence of Generative Artificial Intelligence (GenAI), which offers innovative solutions for addressing challenges in traditional language education. This study investigates the effectiveness of the GenAI-powered chatbot, Kuki, in enhancing the speaking skills of undergraduate English as a Foreign Language (EFL) students at the University of El Oued, Algeria. A mixed-methods approach was adopted, combining both quantitative and qualitative data to evaluate improvements in learners’ spoken language proficiency. The research involved a cohort of 60 undergraduate EFL learners, divided into two groups: a control group of 30 students who received traditional classroom instruction, and an experimental group of 30 students who interacted with the Kuki chatbot over a 10-week period. Both groups were assessed using pre-tests and post-tests designed to measure specific aspects of speaking proficiency, including accuracy, fluency, discourse complexity, and organizational skills. The pre- and post-tests consisted of structured oral tasks such as picture description, role-playing scenarios, and impromptu speaking exercises, which allowed for a comprehensive evaluation of the learners’ progress. Data analysis techniques included both thematic and statistical methods. Quantitative data from the pre- and post-tests were analyzed using paired t-tests and ANOVA to determine the significance of improvement in speaking skills. Qualitative data, obtained through semi-structured interviews, underwent thematic analysis to identify recurring themes and insights into the learners’ experiences with Kuki. The findings revealed that the experimental group demonstrated notable improvements in spoken fluency, self-confidence, and error correction compared to the control group. The rapid and personalized feedback provided by Kuki facilitated error identification and correction, enabling more effective language practice. Additionally, qualitative data highlighted that students appreciated Kuki’s neutral and patient demeanor, which reduced anxiety and encouraged more frequent practice. This study underscores the potential of GenAI chatbots like Kuki to revolutionize EFL education by offering flexible, accessible, and adaptive learning opportunities. It recommends the integration of such tools into language curricula to enhance learner engagement and outcomes. Future research should explore the long-term effects of GenAI chatbots on language acquisition and their applicability across diverse educational contexts. By providing dynamic and interactive learning experiences, GenAI technologies hold significant promise for transforming language instruction.
2:45-5:30pm | [Poster 3] Autonomy or Dependency? TESOL Learners’ Use of GenAI for Language Learning Tasks
Anastasia Shikanova (Ohio State University🇺🇸)
An Evidence-Based Perspective from One-on-One Online Tutoring As generative AI (GenAI) tools like ChatGPT become increasingly accessible, adult TESOL learners are integrating them into their independent study routines. While these tools offer significant potential for enhancing language practice—through instant feedback, vocabulary expansion, and modelled writing—there is growing concern that overreliance may hinder learner autonomy and critical engagement. This presentation draws on real-world tutoring experiences with adult learners in an online, one-on-one TESOL context to explore how GenAI use can support, rather than replace, intentional language learning. Grounded in classroom practice, I will share how I guide learners to use GenAI ethically and effectively as a supplementary resource. This includes teaching prompt-writing for language development, using AI to rehearse conversations or generate vocabulary-rich contexts, and encouraging learners to critically assess AI-generated responses. Case examples will illustrate how learners initially approach GenAI, the challenges they encounter (e.g., overly polished AI writing, factual inaccuracies, lack of cultural nuance), and how structured tutor support can help shift their use from passive consumption to active engagement. The talk highlights specific strategies I use to promote learner agency, such as collaborative task design, reflective discussion of AI outputs, and scaffolding towards metalinguistic awareness. By positioning GenAI as a tool for practice—rather than a shortcut to task completion—I aim to cultivate both linguistic competence and digital literacy. This session contributes to a growing conversation about the role of GenAI in applied linguistics, offering practical insights for tutors and educators working in one-on-one or small-group TESOL settings. Attendees will leave with adaptable strategies to help learners navigate AI ethically and productively.
2:45-5:30pm | [Poster 4] Enhancing Second Language Speaking Performance at the C1 level with Generative AI
Anastasiia Petrenko (University of Cambridge, UK🇬🇧)
Despite the diversity of textbooks for second language acquisition and the importance of developing fluent and coherent speech on various topics at the C1 level, opportunities to enhance this skill remain limited. Even communicative materials, such as SpeakOut, provide limited resources for practicing speaking skills. Traditionally, a lesson starts with a lead-in and finishes with the discussion based on the text or grammar rule. However, at the C1 level, it is crucial to develop spontaneous speech and to elaborate on the topic. In this talk, using a design-based research approach, I show how ChatGPT can be used to accommodate these needs. I developed several prompts to design paraphrasing exercises, ides continuation exercises, debate simulation, and further discussion questions. I argue that embedding such exercises is beneficial for students and boosts their confidence in using new grammar material and vocabulary. To justify this approach, I conducted a pilot study in which 8 Russian-speaking people with the C1 level of English took part. They were divided into two groups. The first group followed the standard SpeakOut plan, while the second group received additional AI-generated exercises. The 90-minute lessons were delivered online, via Zoom, and were on the topic of inversion. Results from the first session revealed that the SpeakOut materials alone were insufficient for full acquisition, and students expressed a lack of confidence in using the structures. The second group showed better performance and greater engagement benefiting from additional AI-generated exercises.
2:45-5:30pm | [Poster 5] Innovative Approaches to Supporting SLIFE: Generative AI for Language Acquisition and Motivation
Anthony Guzman & Hoa Nguyen (Teachers College, Columbia University🇺🇸)
Following an extensive period of research into the fields of artificial intelligence and language acquisition, a notable lacuna in research concerning language learners in the K-12 population was identified. This lacuna represents a significant gap in the evaluation of artificial intelligence (AI) effectiveness in the context of language learning and teaching for ELLs in impoverished communities. A prevailing concern among educators is the potential of AI to function as a mere substitute in the classroom. As a dual researcher and educator in K-12 contexts, I have employed generative AI in the English Language Arts and TESOL classroom for specific content and language objectives. My classroom report will present a range of lesson designs that incorporate content and language objectives, along with the customized adaptation of a conversational agent to function as a teacher-assistant during specific instructional segments. The report will include a detailed data analysis of the impact of the conversational agent on writing skills among English Language Learners (ELLs). The objective of this analysis is to provide insights into the use of generative AI by these students for specific purposes and to explore more sophisticated pedagogical approaches to integrating this technology with SLIFE.
2:45-5:30pm | [Poster 6] To what extent should university students be allowed to use AI in their language studies?
Aurelija Daukšaitė-Kolpakovienė (Vytautas Magnus University, Lithuania🇱🇹)
This case study explored Lithuanian university students’ opinions about using AI in their studies, especially in language-related courses. The university where the study was conducted requires its students to study English (as a second language) until level C1/ C2 proficiency (based on the Common European Framework of Reference for Languages) is achieved. 20 students taking an elective English for Specific Purposes course (English for Humanities, advanced level) wrote paragraphs on Moodle to answer the following question: to what extent should university students be allowed to use AI in their language studies? Through thematic analysis of the paragraphs, the following main themes emerged: advantages, disadvantages, dangers or challenges, assessment and recommendations regarding AI usage. The advantages mainly included grammar and vocabulary learning with the help of AI-based (interactive) tools. The identified dangers focused on non-linguistic issues related to creativity, problem-solving, and overreliance on or overuse of AI. The students also argued that AI should not be used in the assignments that are assessed (e.g., tests and examinations). Finally, some recommendations were provided. The students expressed a need for clear university guidelines indicating what exactly is allowed so that students can “maintain academic integrity”. Currently, the university has general guidelines but encourages separate departments to have their specific rules and regulations. Even teachers can prepare their own rules, which may confuse the students.
2:45-5:30pm | [Poster 7] Iceberg under the Surface: EFL Teachers’ Overt and Covert Challenges of Integrating Generative Artificial Intelligence into Teaching
Behrokh Abdoli (University of Tehran, Iran🇮🇷)
The various challenges that English as a foreign language (EFL) teachers encounter while incorporating Generative Artificial Intelligence (GenAI) into their teaching practice are significant, particularly considering the rapidly-evolving nature of technology in language education (Holmes et al., 2019). Therefore, a transcendental phenomenology study was conducted to explore EFL teachers’ challenges in integrating GenAI into teaching. To this end, narratives and interviews from 13 EFL teachers were analyzed using Giorgi’s (2009) five-step framework. The findings revealed four primary challenges EFL teachers face when integrating GenAI into teaching, including developing engaging AI-driven materials, managing skill stagnation in response to constantly developing AI tools, overcoming technical obstacles to AI implementation, and optimizing limited technological resources. It was concluded that EFL teachers face two distinct categories of challenges. The first category is overt challenges, caused by readily apparent reasons, such as financial barriers to Al integration. The second category is covert challenges which impact the emergence of overt challenges. While covert challenges may be less immediately apparent, they are of considerable significance, such as the need for ongoing professional development among EFL teachers to remain current with Al developments and effectively implement them in their classrooms. The implications of this study underscore the critical need for comprehensive EFL teacher education programs that address not only the overt logistical challenges of integrating Al into teaching but also explore the deeper, less obvious challenges that, if neglected, could result in more significant problems.
2:45-5:30pm | [Poster 8] Artificial Intelligence Literacy in Language Teacher Education: An Exploratory Study
Cíntia Regina Lacerda Rabello (Universidade Federal Fluminense, Brazil🇧🇷)
This study is situated within the contemporary context of the rapid emergence of Generative Artificial Intelligence (GAI) technologies, such as large language models (LLMs). The evolution and widespread adoption of these technologies have raised concerns among educators and administrators regarding their ethical implications in higher education and research. Believing in the potential for the critical and ethical use of these technologies in the development of writing skills through a co-authorship process (Söğüt, 2024), this research aims to investigate how pre-service language teachers are utilizing Artificial Intelligence technologies for language learning and academic writing in their undergraduate courses. The subsidiary objectives comprise the identification of potentialities and challenges of AI-mediated language teaching and learning, as well as how students and faculty are dealing with these technologies in the academic setting. The ongoing exploratory study was conducted with pre-service teachers in a Brazilian university and involved the use of an online questionnaire in the second semester of 2024. The questionnaire consisted of 20 open and closed-ended questions designed to understand the uses of these technologies in language teaching and learning. The main results evidence that most pre-service teachers use AI technologies for different academic purposes and believe that they should be included in the curriculum and that language educators should guide them in their ethical and purposeful use in language learning. These results may help language educators and administrators to understand the potential and implications of GAI technologies in higher education, exploring their possible contributions to language teaching and learning, as well as the development of academic writing skills. Finally, the study seeks to reflect on the need for AI literacy (Chan, 2024) in language teacher education to foster responsible use of GAI technologies in contemporary society.
2:45-5:30pm | [Poster 9] Using Language to Establish “Friend” and “Mentor” Personas: How Well Do GenAI Chatbots Do?
Febriana Lestari , Gi Jung Kim , Widya Kusumaningrum , Hwee Jean (Cindy) Lim , Duong Nguyen , Mostafa Ranjbar , Shuhui Yin , & Carol Chapelle (Iowa State University🇺🇸)
Most users of GenAI chatbots are immediately struck by the friendliness of the machine as it presents its readiness to help and encouragement in conversation with its human interlocutors (e.g., Brooks, 2023). Consequently, GenAI tools are prognosticated to be able to take over a range of communication-related human jobs including language teaching and assessment. However, this observation about the adeptness of GenAI tools at displaying human friend and mentor personas requires a closer examination. Urks for you…”). Such interpersonal language also expresses a measure of familiarity expected of a friend in contrast to the informational language (e.gsing methods for context-based qualitative discourse analysis (Martin and Rose, 2007), the study examined a sample of 93 communicative exchanges between the humans and the GenAI chatbots. Analysis focused on discovery of the language generated to create appropriate personas for interacting with humans who asked for help with questions about daily life and with learning. Conversations initiated by the humans for these purposes called upon the chatbot to adopt friend and mentor personas, so interpersonal language (Englebretson, 2007; Martin and White, 2005) is critical to express polite engagement (“Would you like…”), positive attitude (“Yes, absolutely!”), and alignment (“find what wo., with nominalizations and post nominal modifiers) signaling greater distance (Biber, 1986). Findings reveal instances in conversations with the GenAI chatbots where they succeed and where they fail to generate language that display their friend and mentor personas. Results are used to suggest linguistic features that can be used to increase the rigor of evaluations of the friend and mentor personas constructed by GenAI chatbots. References Biber, D. (1988). Variation across speech and writing. Cambridge University Press. Brooks, E. (2023, December 14). You can’t truly be friends with an AI: Just because a relationship with a chatbot feels real, that doesn’t mean it is.
2:45-5:30pm | [Poster 10] Redesigning L2 Writing Development through AI: A Multidimensional Approach Integrating Generative Feedback, Emotion Recognition, and Learner Autonomy in Adaptive Learning Environments
Hanieh Shafiee Rad (Shahrekord University, Iran🇮🇷)
This study presents a multidimensional framework for enhancing L2 writing development through artificial intelligence, focusing on integrating generative feedback, emotion recognition, and learner autonomy within adaptive learning environments. While previous research has demonstrated the value of automated feedback and AI-supported writing tools, few studies have examined how combining cognitive, affective, and behavioral dimensions can holistically transform the L2 writing experience. Addressing this gap, the current research investigates how AI-driven interventions, specifically large language models (LLMs) providing generative feedback, emotion-aware systems offering affective support, and learner-controlled pathways promoting autonomy, can jointly improve writing proficiency and learner engagement. The study employed a mixed-methods design involving 150 university-level EFL learners across three instructional conditions: traditional instruction, AI-generated feedback only, and the full multidimensional AI-enhanced model. Quantitative data were gathered through pre- and post-writing assessments, learner autonomy scales, and affective engagement surveys. Qualitative insights were collected via reflective journals and focus group interviews. Findings indicate that the multidimensional AI-enhanced group showed the most significant improvement in lexical richness, syntactic complexity, and overall coherence (p < .001). Learners in this group also reported increased emotional resilience, motivation, and a stronger sense of ownership over their writing process. The integration of emotion recognition enabled timely affective scaffolding, while customizable learning pathways empowered students to engage more deeply with writing tasks. These results underscore the transformative potential of AI when leveraged not as a replacement for human instruction but as an intelligent partner in personalized learning. The study offers a new direction for L2 pedagogy, advocating for systems that not only deliver feedback but also recognize learner emotion and foster autonomy. Implications highlight the need for designing adaptive ecosystems that are emotionally intelligent, cognitively supportive, and pedagogically flexible to meet the evolving demands of L2 writers in the age of AI.
2:45-5:30pm | [Poster 11] Agreement Between Large Language Models and Human Raters in Essay Scoring: A Research Synthesis
Hongli Li , Che Han Chen , Kevin Fan, Chiho Young-Johnson , Soyoung Lim , & Yali Feng (Georgia State University🇺🇸)
Modern transformer-based Large Language Models (LLMs) are increasingly used for automated essay scoring, but their reliability compared to human ratings remains unclear, with mixed findings. Shi and Aryadoust (2024) provided a systematic review on AI-based automatic written feedback using SSCI-indexed published between 1993 and 2022. However, since then, more advanced LLMs have emerged, and recent studies report varying results. For example, Yavuz et al. (2024) found high correlations (above 0.9) between scores from LLMs (e.g., ChatGPT, Bard) and human raters. In contrast, Kundun and Bardosa (2024) reported that LLMs (e.g., ChatGPT, Llama) gave lower scores and showed weak correlations with human ratings. This study aims to synthesize empirical studies comparing LLM-generated essay scores with human ratings. To ensure broad coverage, we will include published peer-reviewed articles, unpublished manuscripts, and other relevant empirical studies. We will summarize patterns of agreement between LLMs and human rates (as well as among different LLMs) and factors influencing such alignment. We will also assess whether a meta-analysis is feasible based on the number and quality of identified empirical studies. This synthesis will offer timely insights into the current capabilities of LLMs in essay scoring and inform their use in writing assessment.
2:45-5:30pm | [Poster 12] Uncovering Promotional Rhetoric: A Corpus-Based Move Analysis of Chinese Book Sales Copy
Hui-Hsien Feng & Tsai-Yuan Huang (National Kaohsiung University of Science and Technology, Taiwan🇹🇼)
Genre analysis is a widely used approach for understanding how language fulfills communicative purposes in specific discourse communities. A key methodological development within this approach is move analysis. It has been extended to various domains of English for Specific Purposes (ESP), including business communication (Park, Jeon and Shim, 2021). In business discourse communities, promotional genres such as book sales copy has emerged as a highly adaptable and rapidly evolving area of discourse (Bhatia, 2005). In addition, research has demonstrated that corpus-based discourse analysis is highly useful for analyzing various types of business texts (e.g., Handford, 2010; Li and Chen, 2022). However, while corpus-based move analysis has been effectively applied to various promotional texts in English-language contexts, Chinese book sales copy remains underexplored, particularly regarding their rhetorical structure and linguistic features (Zhu, 1997, 2000). Addressing this gap, the present study investigates the rhetorical move structure and associated linguistic characteristics of Chinese book sales copy in the finance and business management sectors. A corpus of 400 sales copy was compiled via web scraping from a leading Taiwanese online bookstore. An initial rhetorical structure was developed by identifying recurring moves and steps through iterative analysis. This structure was further refined in consultation with professors specializing in move analysis and sales copywriting. The finalized structure was subsequently analyzed using Feng’s (2015) rhetorical structure framework, which includes schema coverage, move diversity, and move density, to describe its rhetorical distribution and composition. Natural language processing techniques were applied to extract linguistic features of each move, such as frequent lexical items and their part of speech. By uncovering the rhetorical and linguistic patterns embedded in Chinese book sales discourse, this study aims to contribute to genre analysis, business communication, and Chinese for Specific Purposes (CSP), offering practical implications for marketing professionals and publishers.
2:45-5:30pm | [Poster 13] Profiling L2 Learner Proficiency for AI-Supported Writing Feedback: A Corpus-Based Study Using ICNALE
Hyunhwa Kim (Georgia State University🇺🇸)
This study explores how natural language processing (NLP) tools can inform second language (L2) writing assessment and instruction by analyzing learner texts through a generative AI-augmented lens. Argumentative essays from the International Corpus Network of Asian Learners of English (ICNALE) were analyzed to extract lexico-grammatical features—such as part-of-speech (POS) patterns and n-grams—using spaCy. These features were examined in relation to learners’ L1 backgrounds and proficiency levels to identify linguistic patterns across groups, with implications for more tailored, data-driven feedback. Drawing on corpus-based frameworks for academic writing complexity (Biber and Gray, 2016), the study highlights how such learner profiles can inform the development of AI-assisted writing tools that support instructional goals. Rather than replacing human evaluation, generative AI systems grounded in empirical learner data can serve as interactive support tools, offering individualized guidance while maintaining instructor oversight. This poster contributes to ongoing discussions surrounding the integration of GenAI in applied linguistics by proposing a hybrid model of human-AI interaction for L2 writing pedagogy.
2:45-5:30pm | [Poster 14] AI Integration in South Korean K-12 English Education and the Role of Communities of Practice
Hyunjoo Moon & Greg Kessler (Ohio University🇺🇸)
This case study explores the various ways AI is being utilized in K -12 South Korean schools, the challenges teachers encounter, and the support needed to facilitate successful implementation. We discovered that teachers leverage AI for lesson preparation, such as generating creative and engaging learning activities and worksheets. AI also serves as a teaching resource, providing content knowledge on various lesson topics and the English language. Additionally, AI-assisted learning analytics enable teachers to track student progress and assist teachers in providing prompt, personalized feedback to students. However, several challenges persist, including limited infrastructure, student digital distractions, over-reliance, age restrictions, and inaccuracies in AI-generated content. Notably, although AI ethics guidelines have been developed by provincial offices of education, many teachers remain unaware of them. To navigate AI integration effectively, teachers largely rely on communities of practice, such as research groups, open classes, teachers’ learning communities, training programs, and social media. This study provides valuable insights for educational practitioners seeking to understand how AI transforms English language teaching and learning. Also, by identifying the challenges and the gaps in teacher professional development, researchers provide a road map to facilitate AI integration in education in South Korea.
2:45-5:30pm | [Poster 15] Sora in the L2 Classroom: Harnessing Text-to-Video AI for Vocabulary Learning
Ibrahim Halil Topal (Gazi University, Türkiye🇹🇷)
This poster explores the pedagogical potential of OpenAI’s Sora, a generative AI model that converts natural language prompts into high-quality, dynamic video in the context of second and foreign-language vocabulary instruction. Sora’s ability to generate realistic visual scenes from descriptive prompts offers language educators a novel modality for presenting vocabulary in highly contextualized, multimodal formats. Drawing from principles of dual coding theory, comprehensible input, and contextual learning, the poster highlights how Sora can support the acquisition of concrete, thematic, and action-based vocabulary. Through curated examples, we demonstrate how Sora-generated videos can scaffold meaning, prompt learner production, and facilitate engagement with new lexical items across proficiency levels. Sample prompts, scene stills, and vocabulary mappings are presented to illustrate potential classroom applications. Focusing on a foundational skill in language acquisition, like vocabulary, contributes to ongoing conversations about AI integration in L2 pedagogy. The poster invites participants to consider the ethical and practical dimensions of AI-mediated instruction and discuss how generative tools like Sora can reshape input design in language education.
2:45-5:30pm | [Poster 16] Enhancing Supervisor-Instructor Dialogue with AI: Rethinking Post-Observation Feedback in Language Classroom
Irina Mikhaylova (Foreign Service Institute🇺🇸)
This poster highlights how generative AI tools like StateChat are being used to improve post-observation conferences in a professional foreign language training program. These tools help supervisors and instructors reflect more deeply on classroom challenges and generate targeted instructional responses. Supervisors can use AI to address common issues noted during class observations by generating practical suggestions, lesson tweaks, and example tasks. AI can also help create alternative instructional segments based on the lesson materials, offering revised activity sequences or task types to address specific learning gaps. For supervisors who don’t speak the language of instruction, AI can support post-observation discussions by summarizing or translating class content and helping generate informed feedback questions. This ensures that feedback remains pedagogically relevant even without full language proficiency. The use of AI in feedback sessions can also spark broader idea-sharing among instructors and supervisors. Informal exchanges of AI prompts and classroom applications can foster a culture of experimentation across the team. This poster shares practical examples, sample prompts, and takeaways for language educators and program leaders interested in using AI to strengthen observation and feedback processes in linguistically diverse settings.
2:45-5:30pm | [Poster 17] The Role of Generative AI in Institutional Policies and Pedagogical Practices
Isaac Ewuoso (Iowa State University🇺🇸)
Two months after its launch, ChatGPT had amassed approximately 100 million active users (Hu, 2023). This rapid adoption prompted educators, policymakers, and scholars in educational technology to propose various responses to what they perceived as disruptive innovation. In higher education, many of these responses have taken the form of evolving institutional policies on AI use, which continue to shift alongside advancements in AI and our understanding of its capabilities (Cummings et al., 2024, p. 2). This poster presents a qualitative study’s findings based on a thematic analysis of syllabi statements from multi-section composition classes and interviews with faculty (n = 13) at a large Midwestern University. Policies regulating generative AI (“Gen AI”) in higher education remain in flux at the university. The faculty in this study reveal how they enforce syllabi policies on Gen AI differently as they strive to uphold academic integrity. The study finds that many instructors’ attitudes toward AI usage in the classroom are on a continuum. Some raise pedagogical anxiety related to the authorship of students’ work, while others are optimistic about including Gen AI tools in basic courses. Findings of this study suggest some implications for institutional policies and pedagogical practices in English composition classes.
2:45-5:30pm | [Poster 18] Empowering Language Learners to Navigate Technology: Interactive Modules on AI, Machine Translation, and Online Dictionaries
Lillian Jones & Julia Gómez (University of California, Irvine🇺🇸)
This virtual poster introduces a set of interactive Canvas-based learning modules developed to guide early language learners in critically and ethically navigating key language technologies: generative AI, machine translation, and online dictionaries. The modules incorporate multimodal resources- videos, readings, and interactive activities such as tool comparisons and AI prompt experimentation– to develop digital literacy, support informed technological tool use, and foster ethical reflection. While not yet an empirical study, the project is grounded in applied linguistics pedagogy, the Successive Approximation Model of instructional design, and research– informed insights on artificial intelligence. Now in its second beta phase, revisions reflect feedback from students, faculty, and digital learning experts. Ultimately, the modules will become a required part of our Spanish language program curriculum and a language-agnostic module template will be released as an open education resource (OER) to support broader education, teacher training, and adoption. This presentation aims to share this resource, as well as gather feedback from topic experts on module design, ethical framing, and directions for future empirical research.
2:45-5:30pm | [Poster 19] Ethical Considerations in Legal Machine Translation
Marus Mkrtchyan (Charles University, Czech Republic🇨🇿)
Since the 1950s, people started scrutinizing Machine Translation. Employing various strategies and approaches, it evolved from Rule-based Machine Translation to Statistical and Neural Machine Translation. Machine Translation is nowadays widely utilized. It is applied by a broad range of users for translating everyday conversations and even professional texts. Previously, Machine Translation output was incomprehensible and it produced incoherent results, nonetheless, recently it achieved substantial improvement and enhanced its capacities. However, this may lead to far more serious issues than people can imagine. Despite profound advancements in this area, the issue of Machine Translation ethics is still a critical challenge that requires considerable attention. Depending on the domain of the translation, we can encounter various ethical issues. The aim of the essay is to study the main ethical issues that humanity can face due to the integration of Machine Translation tools into the field of translation and their misapplication. In the legal domain, we can observe such issues as transparency, privacy, adequacy, algorithmic bias, accuracy, and fidelity. For example: 1) Rigid structure of legal texts 2) Accuracy and precision 3) Usage of specialized legal terminology 4) Specificities of legal systems and legal language dependency on the system 5) Confidentiality and Privacy (non-disclosure duty/responsibility) 6) Legal standards and conventions. The essay seeks to analyze the ethical implications of Machine Translation with a special focus on challenges arising specifically while translating legal texts. It covers such aspects of Machine Translation of legal texts as confidentiality, accuracy and precision. In the essay, a comparative analysis between human and machine translations will be conducted as well. We will discuss the differences between the two types and the associated risks.
2:45-5:30pm | [Poster 20] Task-Based Language Teaching Meets AI: ChatGPT as a Peer Review Tool
Meysa Acar (Vrije University Brussels, Belgium🇧🇪)
This presentation reports about the results of the implementation of ChatGPT integrated Technology-Mediated Task Based Language Teaching (TMTBLT) writing materials in an English as a Foreign Language (EFL) classroom. The developed writing materials are in accordance with the Task-Based Language Teaching (TBLT) framework of Ellis (2003) and have been integrated into TMTBLT accordingly. This research explores the link between the integration of ChatGPT and Interactionist Theory within the framework of Foreign Language Acquisition (FLA) notably in relation to the negotiation of meaning and/or form. Furthermore, this study examines the effectiveness of ChatGPT in terms of a peer feedback tool, based on qualitative data with a focus group of English language teachers and students’ written outcomes from the ChatGPT-integrated TMTBLT task. The data are complimented by the observation of the student’s interactions with ChatGPT and individual interviews with the English language teachers. The findings seem to indicate that whereas ChatGPT is time efficient for both teachers and students, it provides explicit error correction which doesn’t align with Interactionist Theory to foreign language teaching. Although, the teachers acknowledge the advantages of ChatGPT, they highlight the challenges such as curricular constrains and classroom conditions.
2:45-5:30pm | [Poster 21] Evaluating LLM Performance on Translated vs. Non-Translated Queries in Low-Resource West African Languages
Ndiana Obot (University of Chicago🇺🇸)
Machine Translation has transformed cross-lingual communication through tools like Google Translate, mitigating language barriers for its over 200 million active users (Ngak, 2012). Similarly, Large Language Models (LLMs) have advanced communication, education, and knowledge access globally (Peláez-Sánchez et al., 2024). However, a growing disparity in performance between high-resource and low-resource languages threatens to marginalize over 3 billion speakers worldwide (Song et al., 2025; Kshetri, 2024). This divide is especially pronounced in sub-Saharan Africa, where all 2,123 native languages are considered low-resource, resulting in significantly lower LLM performance (Hammerstrom, 2015; Alhanai et al., 2024). As a workaround, many users translate queries into English, interact with LLMs, and then translate responses back; however, this process often reduces clarity and accuracy (Alhanai et al., 2024). Previous studies have also shown that cumulative errors from translation and back-translation can be particularly severe for African languages (Benjamin, 2019). Yet, while both machine translation and LLMs have improved, it remains unclear which languages suffer more from cumulative translation-induced errors and those for which translation improves LLM results. This study investigates this gap by evaluating LLM performance across several West African languages, including Igbo, Yoruba, Hausa, Twi, and Ewe, comparing native-language queries to their translated equivalents.
2:45-5:30pm | [Poster 22] Unveiling the Negative-Side Effects and Limitations of AI-Powered Tools in English Language Teaching
Özge Gümüş (Adıyaman University, Türkiye🇹🇷) & Emel Kulaksiz (National Defence University, Türkiye🇹🇷)
Although AI-powered tools (e.g., ChatGPT, Grammarly, QuillBot, and ELSA Speak, etc.) have increasingly found their way into English as Foreign Language (EFL) classrooms, yet due to the lack of clear pedagogical frameworks, their use often remains unguided. This mixed- methods study aims to uncover the negative-side effects and limitations confronted within the use of AI based language learning tools (e.g., ChatGPT, Grammarly, QuillBot, Elsa Speak, etc.) from the perspectives of English language teachers and learners. To gather data, the study employed semi-structured focus group interviews and open-ended questionnaires to tertiary level English language teachers and learners of English (n=28) at a state university. Findings reveal a range of concerns including 1) excessive reliance on AI tools, potentially undermining independent learning; (2) mismatches in language proficiency levels; (3) content inaccuracy and misleading feedback facts; (4) privacy and data security threats; (5) reduced teacher agency over instructional choices; and (6) cultural and contextual insensitivity in AI outputs. These findings highlight the nuanced challenges of AI adoption in EFL context and thus underscore the need for a balanced AI adoption, focusing on controlled implementation and as well as ethical concerns. By pointing to these pitfalls, this study adds to the existing body of literature on AI in language learning, providing practical suggestions for educators, policymakers, and researchers to maximize AI’s benefit while minimizing its risks. In moving forward, additional research needs to investigate ways to promote responsible AI use that augments—and not displaces—human instruction to support pedagogical integrity and learner development.
2:45-5:30pm | [Poster 23] Boosting Non-Native Students’ Confidence in Pronunciation Through AI-Assisted Teaching
Patricia Ibiapina (British Council ESAT-J, UK🇬🇧)
As English continues to function as a global lingua franca, the ability to communicate clearly and confidently has become more important than ever, particularly for non-native speakers. Yet, pronunciation instruction remains underemphasized in many curricula, often leaving learners feeling insecure and unintelligible in real-world contexts. This presentation explores how AI-assisted teaching can provide an innovative and scalable solution to enhance pronunciation and build learner confidence, especially in contexts with limited access to native-speaker input or expert phonological instruction. Drawing on classroom-based action research and practitioner experience across diverse educational settings in Africa and India, this presentation outlines how AI technologies can provide real-time, individualized feedback to learners. Unlike traditional methods, these tools allow students to practice autonomously in a low-pressure environment, receive targeted correction, and gain exposure to various English accents, thus promoting both accuracy and intelligibility. The presentation also situates this practice within the broader discourse of English as an International Language (EIL) and English as a Lingua Franca (ELF), challenging native-speaker norms and advocating for intelligibility-focused teaching. Case studies will highlight learner progress and engagement, with a focus on how these tools can democratize access to quality pronunciation instruction. The session will conclude by offering pedagogical implications for teacher training, material development, and classroom implementation. It will be of interest to educators, teacher trainers, and researchers seeking to integrate technology meaningfully into language instruction, particularly in under-resourced or multilingual contexts.
2:45-5:30pm | [Poster 24] In-Service English Teachers’ Attitudes Toward Artificial Intelligence in Education and their Reflections to the Students
Pelin Derinalp (Gaziantep University, Türkiye🇹🇷)
Teacher Attitudes Toward Artificial Intelligence: A Qualitative Study with In-Service EFL Teachers Artificial intelligence (AI), which has been discussed in academic and industrial circles since the 1950s, has been rapidly spreading in the field of education in recent years. Especially in the context of language education, AI has been shown to have many potential benefits such as increasing student achievement (Song and Song, 2023; Wei, 2023), providing personalized learning experiences (Amin, 2023) and enriching teaching materials (Hsiao and Chang, 2023). However, the extent to which AI is effective in real classroom settings and how teachers perceive this technology has not yet been adequately researched. This study aims to examine in-service EFL teachers’ attitudes towards AI. Within the scope of the study, three focus group discussions were conducted and data were collected from a total of seven teachers. Using thematic analysis, the findings were categorized around three main themes: (1) teachers’ perceptions of AI, (2) the potential role of AI in foreign language teaching, and (3) the challenges faced in integrating AI into the teaching process. By revealing how teachers view AI as a pedagogical tool and their opportunities and concerns, the findings point to important implications for the future use of AI in foreign language teaching.
2:45-5:30pm | [Poster 25] Examining the Ethical Implications and Stakeholder Perceptions of Integrating GenAI into Language Teaching and Learning
Pooria Barzan, Mohammad Mahdi Maadikhah , & Reza Khany (Ilam University, Iran🇮🇷)
Despite the increasing adoption of Generative Artificial Intelligence (GenAI) in language teaching and learning, there is a pressing need to explore its ethical implications and stakeholder perceptions. This mixed-methods study investigates the attitudes of teachers and learners toward GenAI integration in language education, focusing on ethical concerns such as data privacy, equity of access, and the potential displacement of human instructors. Data were gathered through surveys, focus groups, and case studies of institutional policies. Surveys revealed a widespread acknowledgment of GenAI’s potential to enhance language learning, coupled with significant apprehensions about privacy and fairness in access. Focus group discussions provided deeper insights, emphasizing the demand for transparent ethical guidelines and the preservation of human-centered pedagogy. Case studies illustrated a range of institutional responses, with some prioritizing ethical frameworks more robustly than others. The findings highlight the critical need for well-defined institutional policies and sustained stakeholder engagement to responsibly integrate GenAI into education. Furthermore, the study underscores broader societal implications, particularly regarding the evolving role of educators in an AI-enhanced landscape. By addressing these challenges, this research contributes to the discourse on ethical AI use in education and offers practical recommendations for policymakers and practitioners to navigate this transformative shift.
2:45-5:30pm | [Poster 26] A Battle of Bots: Comparing ChatGPT and DeepSeek’s Translation Strategies for English Idioms into Persian
Roghaieh Moslehpour (Shiraz University, Iran🇮🇷)
This study, drawing on Baker’s (1992) framework of idiom translation strategies, examines the performance of ChatGPT and DeepSeek in translating English idiomatic expressions into Persian. Adopting a mixed-methods design, the research compares AI-generated translations with human-produced equivalents. To that aim, seventy-five idioms across five thematic categories—age, beauty, family, food, and clothes—were selected from Rafatbakhsh and Ahmadi’s (2019) frequency-based list of idioms, derived from the Oxford Dictionary of Idioms. Human translations were compiled from established bilingual resources, including Farhang Moaser (2014), Aryanpur Progressive Dictionary (2005), Abadis online dictionary and 1001 English Idioms with Persian Translation (2012), and validated by experts in English studies and Persian literature, five MA and PhD holders of English studies and a PhD student of Persian Literature. AI-generated translations were assessed for accuracy and later analyzed using Baker’s (1992) taxonomy of translation strategies. The results of data analysis reveal that ChatGPT outperformed DeepSeek with a 46.9percent accuracy rate, compared to DeepSeek’s 32.6percent. Across all AI-translation outputs, the most frequently employed strategy was using an idiom of similar meaning but dissimilar form, followed by paraphrasing and, less commonly, using an idiom of similar meaning and form. In short, while both DeepSeek and ChatGPT exceed in using the translation by paraphrase strategy in comparison to human translators, ChatGPT’s strategy use most closely resembled that of human translators, demonstrating higher fidelity and a stronger alignment with human translation norms.
2:45-5:30pm | [Poster 27] Ethical use of AI in language assessment: a review of ILTA alignment
Rurik Tywoniw (University of Illinois Urbana-Champaign🇺🇸)
Generative Artificial Intelligence (GenAI) has become ubiquitous in academia as both teachers and students innovate with GenAI-produced text. Educators have experimented with GenAI for item generation (Runge et al., 2024) and grading and scoring (Kaldaras et al., 2024), while GenAI allows students to expand their skillsets by using chat tools for editing, planning, summarizing, and other tasks not related to idea generation. All the while, the use of GenAI in education has generated ethical controversies related to students’ overuse of AI and lack of critical engagement with assignments (Hou et al., 2025), and the GenAI industry faces controversies related to data privacy, providence of Large Language Model data, and energy consumption. However, the use of GenAI tools by educators and assessors remains relatively unexplored from an ethical perspective. This synthesis reviews the current landscape of GenAI tools, their promises, and their controversies, and notes considerations for applying GenAI tools to various aspects of assessment, such as item generation, automated scoring, and feedback creation. Various uses of GenAI are scrutinized through the lens of the ILTA Code of Ethics, and discussion of each of the nine principles of ethics are used to make prescriptions for use of AI in language testing.
2:45-5:30pm | [Poster 28] Creating an L2 Pronunciation Unit with GenAI
Sebastian Leal-Arenas (University of Pittsburgh🇺🇸)
As generative AI (GenAI) tools become increasingly accessible, language educators are exploring their potential in enhancing second language (L2) instruction. One area of particular need is pronunciation, which, despite its importance for communicative competence, is often overlooked in the curricula of Spanish language programs. This study examines how GenAI can support the development of a pronunciation-focused unit for university-level L2 Spanish learners. Using tools such as ChatGPT and AI-powered voice applications, the unit was designed to offer learners structured exposure to key pronunciation features and scaffolded practice through AI-generated scripts, dialogues, and listening tasks. While the unit does not provide personalized or interactive feedback, it demonstrates how GenAI can be used creatively by instructors to design accessible and engaging pronunciation materials. The study also addresses ethical and pedagogical considerations, including accent representation, authenticity of input, and the evolving role of the instructor.
2:45-5:30pm | [Poster 29] Novice English Language Educators’ Technological Pedagogical and Content Knowledge in Materials Development: A Case Study
Shahzadi Kulsoom & Sadia Irshad (Air University, Pakistan🇵🇰)
This study uses Mishra and Koehler’s TPACK (Technological Pedagogical Content Knowledge) framework (2006) to investigate how new English language teachers who are enrolled in a teacher training course called Technology Enhance Language Learning and Teaching (TELLT) develop prompts for the creation of AI-driven materials. The research uses a mixed-methods approach, incorporating both qualitative and quantitative analysis. Prompts have only been assessed for frequency of use using descriptive statistics in the quantitative phase. The structure of these prompts and the way in which teachers employ AI tools to create materials have been examined. Through the examination of AI tool prompts, the study also investigates teachers’ technical pedagogical and content knowledge on seven TPACK constructs. Structure of the prompts provides information on (a) content i.e teaching objectives such as speaking, writing, listening, vocabulary, and grammar (b) context i.e. the learners’ levels (e.g., beginner, intermediate, advanced); and (c) task i.e. materials developed such as lesson plan, assessments, activities, texts, videos etc. The AI applications employed in the materials development (e.g., ChatGPT, InVideo, Gemini) have also been examined. In the qualitative phase, semi-structured interviews were used to investigate the advantages and difficulties of using AI to the creation of content. The study identifies ChatGPT as the most popular AI tool and emphasizes the value of teacher preparation programs that integrate technology to help new teachers become TPACK literate in using contemporary tools like AI to create engaging and useful ELT materials. The results are intended to advance knowledge of quick structure and its function in maximizing AI-generated material in modern teacher preparation and language education initiatives.
2:45-5:30pm | [Poster 30] Examining the Effectiveness of CoT Prompting in ESL Writing Context: An Experiment with Select GenAI-powered Chatbots
Shruti Das, Jaipal Choudhary (VNIT, India🇮🇳), & Rajakumar Guduru (IIT, India🇮🇳)
The increasing integration of GenAI in academia is suggesting a crucial transformation in the treatment of teaching and learning process. The NLP landscape in ESL context has been revolutionized by LLMs in the form of chatbots. The recent research focused on the vital link between Chain of Thought (CoT) prompting and the multi-steps reasoning ability of AI-powered chatbots. However, this practice is limited to solving Math problems. To date, there is hardly any research that has systematically investigated the application of CoT prompting in ESL writing context. Therefore, the present study is an attempt to bridge the gap by critically examining the effectiveness of CoT prompting on three widely recognized chatbots – ChatGPT, Claude AI, Deepseek – for generating short stories. Self-designed few-shot prompt is applied to generate ideas for creating short stories. Qualitative analysis will be employed to evaluate the generated outcomes by these chatbots. Key findings advocate the importance of CoT prompting in creative writing. Additionally, empirical evidences recommend that human-AI collaboration can provide better results. This research will contribute valuable insights for students, educators, software developers and AI researchers. However, the limitation of the current study is that only one genre of short story with three chatbots are explored here qualitatively. Further research on the same theme of different genres and with other popular chatbots are therefore recommended.
2:45-5:30pm | [Poster 31] A Systematic Review of the Impact of Generative AI on Second Language Acquisition in Educational Settings: Focusing on Student Motivations
Sun Kun (University of Malaya, Malaysia🇲🇾)
This systematic review, following PRISMA (Preferred Reporting Items for Systematic Reviews and Meta – Analyses), examines how generative AI, like ChatGPT, affects second language acquisition (SLA) in educational settings. Since ChatGPT emerged, SLA in campus has faced challenges, with some universities banning its use in foreign language learning. However, this paper deems such bans outdated, as generative AI is an unstoppable technological trend that should be embraced. Generative AI can boost students’ after – class assignment efficiency, yet its impact on learning outcomes raises concerns. This review focuses on the link between students’ second – language learning motivation and their use of generative AI. Existing studies show higher motivation often leads to more use of generative AI, but the nature of this motivation needs clarification. In academic research, motivation, as an educational psychological variable, is understood differently in various theories. Thus, identifying dominant motivation types and their relationship with generative AI use intention is crucial. Extrinsic motivation, especially instrumental motivation, may prompt students to use generative AI for utilitarian ends like getting good grades or quickly finishing assignments. In contrast, intrinsic motivation, driven by a love for the second language, may lead students to use it to enhance learning experience. Figuring out which motivation type mainly influences students’ use of generative AI in SLA is vital. If instrumental motivation dominates, adjusting the second – language learning structure may be needed to help students use this powerful tool for real language acquisition.
2:45-5:30pm | [Poster 32] AI-generated Texts for Reading Comprehension in EAP Learners: An Empirical Study
Tam Thi Thanh Nguyen (Ho Chi Minh City University of Forestry and Agriculture, Vietnam🇻🇳)
The advent of AI tools has made it more convenient for language teachers to prepare their lessons and teaching materials efficiently, particularly when they have to deal with large-sized classes. In line with this, AI tools play an important role in assisting teachers with reading materials preparation in English for Academic Purposes (EAP). This study aims to explore the effectiveness of AI-generated texts in supporting reading comprehension for EAP learners. A total of 108 undergraduate students participated in this study in which they were assigned into control group and experimental group. They completed reading comprehension tasks based on both human-written and AI-gerenated reading materials. Their scores were then compared using paired samples t-tests to evaluate their reading performance across both text types. The preliminary findings suggest that AI-generated texts can provide comparable support to human-written texts and the complexity of lexcial resources and of AI-generated texts surpasses that of human-written ones. The study indicates the potential use of AI tools as a supplementary resource in EAP reading instruction and pedagogical implications for language educators.
2:45-5:30pm | [Poster 33] Talk Back: Student Perception on AI Chatbot Feedback and Correction Strategies for English Language Learning
Valentina De Brasi & Anna Mongibello (University of Napoli “L’Orientale”, Italy🇮🇹)
The use of Large Language Models, particularly chatbots, has grown exponentially in recent years. Language learning contexts are greatly affected by these conversational tools, which have broadened the scope of computer-assisted language learning (CALL) systems (Bibauw et al., 2022). Their integration implements innovative language learning environments, where AI-powered tools are intended to offer learners tailored input, feedback and a more comfortable learning setting (Huang et al., 2022). This case study focuses on the interactions between AI-chatbots and Italian learners of English as a foreign language. As part of the “UNiversally Inclusive Technologies (UNITE) to practice English” PRIN project, we conducted experimental sessions and collected 72 interactions between two chatbots (ChatGPT and PiAI) and L1 Italian university students learning English. These interactions took place within a two-part activity for language practicing, comprising a task-oriented roleplay and a small talk exchange. At the end of each session, students completed a mandatory post-hoc questionnaire on their perceptions about the experience. This contribution pays particular attention to the post-hoc questionnaire. Through an analysis of the student-chatbot conversations and the subsequent surveys responses, we identify the correction strategies most frequently employed by the chatbots. Furthermore, we examine students’ perceptions of these strategies and appraise whether learners themselves consider the feedback provided by the AI-powered tools useful for effective language learning. Therefore, the findings offer valuable insights into the impact of chatbot-assisted learning on student motivation and its potential influence on language learning experiences.
2:45-5:30pm | [Poster 34] Can using Artificial Intelligence (ChatGPT) enhance medical students’ academic writing?
Zahra Shahsavar (Shiraz University of Medical Sciences, Iran🇮🇷)
This paper tries to explore if the use of ChatGPT as an Artificial Intelligence (AI) tool impacts different components of medical students’ academic writing. Eighty-three students, who randomly separated into the control group (n = 41) and the experimental group (n = 42), took part in this study. In the control group, the instructor applied the traditional writing approach by providing written feedback, focusing on lexical and grammatical aspects. The same process-oriented approach was implemented in the experimental group to teach writing; however, in these classes, the students were trained to use ChatGPT in writing. Four writing-skill components (i.e., content, organization, vocabulary, and mechanics) were examined between the two groups through ANCOVA analysis. The results indicate that the experimental group significantly outperformed the control group in all writing components except for the mechanics. This may imply that AI tools like ChatGPT can be valuable in assisting with certain aspects of writing, but they should not be regarded as a one-size-fits-all solution for enhancing students’ writing skills. The findings of this research can be beneficial for educators, particularly those interested in teaching writing.
2:45-5:30pm | [Poster 35] Generative AI in Language Assessment: Mapping the Current Landscape
Zeynep Arslan & Mega Wulandari (The Ohio State University🇺🇸)
This poster shares findings from a scoping review of empirical studies on the use of Generative AI in language assessment. Guided by four questions, how GenAI is used, what benefits and concerns exist, and what gaps remain, the review draws on 13 peer-reviewed studies published between 2022 and 2025. A structured search was conducted across major databases (Scopus, Web of Science, Google Scholar, and others), followed by title/abstract screening and full-text review using Covidence. Inclusion criteria limited studies to empirical research focused on GenAI-based assessment tools. The selection process followed PRISMA-ScR guidelines. Most studies addressed automated scoring, written feedback, or diagnostic applications, often involving tools like GPT-4. While some reported consistency with human scoring, others raised concerns related to fairness, feedback quality, and applicability across diverse learner populations. English dominated the research context, with limited attention to other languages or multimodal assessment. The review calls for further empirical work that explores human-AI collaboration in assessment and investigates how GenAI can be integrated without reinforcing dominant norms. By mapping existing research, the poster provides a foundation for future inquiry into responsible and inclusive use of GenAI in language assessment.
2:45-5:30pm | [Poster 36] AI Support in Intercultural Communicative Competence: A Study of EFL Teachers and Students
Zeynep Saka Lloyd (Syracuse University🇺🇸)
This study investigates the state of Intercultural Communicative Competence (ICC) in an English as a Foreign Language (EFL) context by examining (1) how EFL learners and teachers perceive their own ICC, (2) what their ICC levels are as measured by Discourse Completion Tasks (DCTs), and (3) how artificial intelligence (AI) may influence these measured levels and be perceived as a tool for ICC development. Using a mixed-methods design, data was collected from 89 students and 46 teachers across nine state universities in Turkey. Participants first completed a 25-item Likert-scale questionnaire on ICC self-perception. Then, they responded to two DCT sets: one independently and one with optional generative AI support. Preliminary findings suggest improved ICC performance in some AI-assisted DCT responses, particularly among students, highlighting AI’s potential as a scaffolding tool. However, concerns emerged around the cultural authenticity of AI-generated responses, as well as its limitations in conveying emotional understanding. Ongoing analysis explores performance patterns by focusing on emerging themes in DCT responses and intergroup differences. By examining the interplay between support of AI and human interaction in ICC development, this study may contribute to broader conversations about the pedagogical implications and future directions of generative AI in second language education.
2:45-5:30pm | [Poster 37] Generative AI for Multimodal Lecture Comprehension Assessment: An Ongoing Study
Ziteng Wang & Vahid Aryadoust (Nanyang Technological University, Singapore🇸🇬)
As generative AI technologies rapidly expand the possibilities of input design in language testing, this study investigates how synthetic multimodal input featuring AI-generated speech and AI personas can be integrated into academic lecture listening assessment. Specifically, we examine the potential of large language models, neural text-to-speech, and text-to-video platforms (e.g., Synthesia) to produce test materials that are not only scalable and realistic but also cognitively principled. The study focuses on how core multimedia principles of design—such as the image principle and redundancy principle—may need to be re-evaluated in the context of synthetic multimodal input. The target task type is lecture listening comprehension, which poses high cognitive demands and is central to academic listening assessment. A set of listening tasks was developed using text-to-video tools to create lecture-like scripts delivered by AI personas. Topics were selected for their general academic relevance and optimized for accessibility and linguistic balance. The comprehension task required test-takers to produce written summaries, allowing for an integrative assessment of understanding beyond surface-level recall. The primary goal is to examine how different input modalities (audio-only vs. AI persona video) influence L2 learners’ comprehension performance in academic listening. This ongoing empirical study employs a within-subjects quantitative design with university EAP students being the main participants of the study. By combining technological innovation with principled test development, the study contributes to the design of next-generation listening assessments. It also invites critical reflection on how traditional multimedia learning principles apply—or evolve—when learners encounter AI-mediated input in assessment settings.
2:45-5:30pm | [Poster 38] PCK-T with Generative AI: A Pedagogy-Driven Experiential Approach to AI Literacy
Zoe Handley (The University of York, UK🇬🇧)
This paper reports the evaluation of “PCK-T with Generative AI”, a pedagogy-driven experiential approach to AI literacy development, within a master’s module on educational technology. “PCK-T with Generative AI” is based on PCK-T or “PaCKed with Technology” (Author, DATE) a broader approach to technology professional development that emphasizes the importance of adopting a pedagogy-driven approach that enables student teachers to make connections between the affordances of new technologies and pedagogical principles and providing student teachers opportunities to experience and reflect on learning through technology. As such, there are three phases to PCK-T with Generative AI: (1) Comprehension and connection, (2) Experience, and (3) Reflection. During the comprehension and connection phase, conceptions of literacy and generative AI are introduced and students are encouraged to explore the possible impact of generative AI on currency conceptions of literacy. During the experience phase, following training in prompting, students were asked to produce an essay using AI, recording what tools they used, how and why, and the outputs they produced. During the reflection phase, following a de Bono-style approach to reflection, students were asked to reflect on what worked, what didn’t, and what they would do differently if they were to write a future assignment with the support of generative AI as well as what advice they would give to teachers and policymakers regarding student use of generative AI and designing effective assignments in the age of generative AI. The impact of “PCK-T with generative AI” on students’ AI literacy was evaluated by comparing students’ responses to a pre-intervention questionnaire exploring their prior understanding of generative AI and the advice they would have given to students and teachers before completing the training with their reflections on their experience writing the summative assessment for the module with the help of generative AI, their responses to a modified version of Ma, Crossthwaite, Sun and Zu’s (2024) ChatGPT literacy questionnaire and their feedback on the value of the training. Data is currently being collected. Preliminary results and reflections from the tutor and markers will be shared in the presentation.
Thursday, November 6th
8-8:30am | Iranian MA TEFL Students’ Perceptions of AI-Generated Feedback Versus Teacher Feedback in Writing Tasks (TSLL25-76)
Reza Khany & Behnaz Alijani (Ilam University, Iran🇮🇷)
The integration of generative artificial intelligence (GenAI) in language learning has sparked growing interest in comparing its effectiveness with traditional feedback methods. This qualitative study explores the perceptions of Iranian MA TEFL students regarding AI-generated and teacher-provided feedback in academic writing tasks. Fifteen MA TEFL students from a public university participated in semi-structured interviews, offering insights into their experiences with both feedback types. Thematic analysis was employed to examine their views on the clarity, usefulness, and emotional impact of each. The results indicate that students appreciated the speed and consistency of AI-generated feedback, particularly for grammar and structural suggestions. However, they felt that teacher feedback offered more personalized, context-sensitive guidance and greater motivational impact, with a stronger ability to address individual learning needs. Some participants also expressed concerns about the impersonal nature of AI feedback, particularly in complex writing tasks that require deeper understanding and engagement. Despite these concerns, AI feedback was valued as a complementary tool, especially during the revision stage. The study contributes to the ongoing discourse on the role of AI in language education, offering practical insights for teachers, curriculum designers, and developers of AI tools aimed at enhancing writing instruction in EFL contexts.
8-8:30am | Does ChatGPT have register awareness? A cross-modality analysis on stance and complexity (TSLL25-174)
Eunmi Kim (Michigan State University🇺🇸)
Previous studies have shown that both spoken and written registers have distinct syntactic complexity features: speakers rely on clausal structures with long dependent clauses, whereas writers rely on phrasal structures that compressed meanings (Biber et al., 2011). Drawing on the Register-Functional approach to grammatical complexity (Biber et al., 2022), this study investigates how register-sensitive stance and complexity features are used across modalities in human and ChatGPT generated texts. The data consist of spoken monologues and written essays produced by English L1 users from the ICNALE corpus and ChatGPT-generated responses in spoken and written forms to the same argumentative prompts. I extracted all spoken-oriented stance and complexity features, including finite complement clauses, finite adverbial clauses, and stance adverbials. They were coded and compared across modalities and sources using normalized frequency. Findings revealed that human-generated spoken texts have richer and more various spoken-oriented stance and complexity features than written essays, demonstrating register awareness of humans. In contrast, ChatGPT relied on phrasal structures in both written and spoken texts with much less stance marking compared to humans, maintaining consistent features across modality. This indicates that ChatGPT has limited register awareness and generates texts that lack register specific features. The study contributes to the growing research on AI language production by providing a register sensitive evaluation of ChatGPT generated texts. It provides better understanding of the extent to which ChatGPT mimics human-like stance-taking, syntactic complexity across modalities.
8:30-9am | Human Touch or Machine Precision? Exploring the Emotional Dimensions of AI Feedback (TSLL25-94)
Johanathan Woodworth (Mount Saint Vincent University, Canada🇨🇦)
As generative AI (GenAI) systems increasingly supplement or replace traditional teacher feedback in educational settings, critical questions arise about their ability to linguistically convey emotional intelligence and empathy in ways that support student growth and well-being. This study investigates the “empathy gap” between human teacher feedback and GenAI-generated feedback on student assignments through a detailed linguistic analysis. Employing a mixed-methods approach, we collected paired feedback samples (human and GenAI) on identical student work across disciplines; each sample was analyzed using a multi-dimensional framework that combined quantitative coding of linguistic markers (such as frequency of motivational language and technical accuracy) with qualitative thematic analysis of empathetic expression and personalization, integrating both data types to triangulate findings. Preliminary findings reveal distinct linguistic patterns: GenAI feedback is characterized by formulaic structures, high consistency, and frequent use of positive but generic language, while human teacher feedback demonstrates greater contextual sensitivity, directness, and authentic expressions of empathy. Notably, students’ engagement with feedback was lower for AI-generated responses—even when they could not distinguish the source, which indicates that perceptions and understanding of AI’s affordances shape willingness to act on feedback. Despite advances in natural language processing and affective computing, GenAI systems still struggle to authentically replicate the linguistic subtleties that convey genuine emotional intelligence. Our findings highlight the need for student education regarding the affordances of AI feedback systems to mitigate bias against AI-generated feedback, and propose a hybrid feedback model that leverages the complementary strengths of both human and AI-generated responses. This research contributes to the discourse on social AI by identifying specific linguistic dimensions of emotional intelligence that remain challenging for AI to authentically manifest in educational contexts.
8:30-9am | Peer Pressure and the Use of GenAI in English Writing: A Qualitative Study of EAP Students (TSLL25-96)
Mehran Hosseinkhani & Musa Nushi (Shahid Beheshti University, Iran🇮🇷)
The integration of Generative AI (GenAI) tools like ChatGPT into academic English writing instruction has transformed academic practices, offering both opportunities and challenges (Barrot, 2023). While these tools support learners in language learning, their use is entangled with social influences on technology usage (Kandoth and Kushe Shekhar, 2022). Thus, this qualitative phenomenological study investigates how peer pressure shapes English for Academic Purposes (EAP) students’ decisions to engage with or resist GenAI technologies in writing tasks. Utilizing in-depth interviews with 22 undergraduate students enrolled in EAP courses at four state universities in Tehran, Iran, the study examines peer pressure (whether explicit or implicit) and how students make sense of their decisions against social, ethical, and institutional expectations. Participants were recruited using a purposive sampling, informed by the review of course materials and syllabi from the universities to maximize the likelihood of participants’ exposure to GenAI tools within the EAP courses. Hybrid thematic analysis of data uncovered complex tensions between individual agency and collective norms, the emotional and ethical dimensions associated with using these tools. These tensions often arose from students’ internal negotiations between academic integrity, peer influence, and their desire for linguistic improvement through GenAI support. Findings provide key insights for educators and policymakers who intend to promote fair and reflective practice in the changing landscape of AI-supported language learning.
9-9:30am | Engagement strategies in human-written and AI-generated academic essays: A corpus-based study (TSLL25-88)
Sharif Alghazo , Ghaleb Rababah (University of Sharjah, UAE🇦🇪), & Dina El-Dakhs (Prince Sultan University, Saudi Arabia🇸🇦)
Based on an appraisal theory framework, this study explores the use and functions of engagement strategies in human-written and AI-generated academic essays. The study analyses 80 essays (40 human-written and 40 AI-generated) for the use of Expansion and Contraction engagement strategies. The human-written essays were collected from the Louvain Corpus of Native English Essays (LOCNESS), which includes essays written by university-level native English writers, while the AI texts were generated by ChatGPT. Analysis shows that both Expansion and Contraction strategies occur more significantly in human-written texts than in AI-generated texts. Native English writers utilise a more significant proportion of Entertain markers, with a sensitive regard for alternative standpoints, and utilise Disclaim markers in actively opposing counterarguments. AI-generated texts, in contrast, utilise a high proportion of objective citing and hedging, with little objective use of strong Proclaim markers and a virtual lack of Concur dialogistic positions. There is a striking contrast in engagement functions, with humans utilising a more significant proportion of complex rhetoric and more profound argumentation supported through statistical analysis. The findings provide implications for educators and writing instructors aiming to enhance students’ argumentative skills and for developers of AI writing tools seeking to improve rhetorical complexity and engagement in generated texts.
9-9:30am | Empowering Literary Genre Pedagogy through Artificial Intelligence: Using ChatGPT to Enhance Students’ Critical Writing Skills (TSLL25-62)
Fizza Malik & Farzana Masroor (Air University, Pakistan🇵🇰)
In educational settings, the integration of Artificial Intelligence (AI) has generated immense opportunities at various stages of genre production such as idea generation, theme development, to drafting and producing text types to fulfil various academic tasks. This requires a controlled use of AI technologies to empower students instead of undermining their critical abilities. Critical writing is an important skill for university students where the ability to critique and analyze poems is essential. This research aims to explore the potential of GenAI to enhance critical writing skills of undergraduate students within the framework of literary genre pedagogy. The research investigates the ChatGPT’s impact on undergraduate students’ ability to write literary critiques by conducting a move analysis of critiques produced before and after AI intervention. The data for this study consists of 20 literary critiques written by university students and 20 critiques generated by ChatGPT. Using Swales’ (1990; 2004) genre analysis framework and Hyland’s (1990) Model of Argumentative essay, the research identifies key rhetorical moves and steps present in both critiques. A qualitative genre analysis of student-written and ChatGPT-generated critiques, supplemented by focused group interview, provides insights into AI’s role in supporting traditional literary instruction. The present study proposes a structured model of literary critique tailored to enhance both students’ critical engagement with poetry. In addition to this, the focus group interviews give qualitative insights into students’ experiences with AI-assisted writing. By comparing student-written and ChatGPT generated critiques, the study provides insights into the role AI can play in supporting traditional literary instruction. The implications of this study contribute to the ongoing discourse on AI in education to improve students’ critical writing skills and can be beneficial for teachers as an outcome of literary genre pedagogy.
10-10:30am | Proofreading with ChatGPT: A comparative study of human and AI feedback on L2 expert academic writing (TSLL25-75)
Marie-Aude Lefer (UCLouvain, Belgium🇧🇪) & Magali Paquot (FNRS – UCLouvain, Belgium🇧🇪)
In an increasingly globalized academic world, the imperative to “publish in English or perish” has led many non-native English-speaking scholars to seek language support services. A common strategy is to draft their articles in English and submit them to professional editors for proofreading—an approach that, while effective, is often costly and inaccessible to many. With the rapid development of generative AI, tools such as ChatGPT are increasingly being used as alternatives to traditional proofreading services (Giglio and Costa, 2023). This shift raises a key concern: To what extent can AI tools provide appropriate and useful language feedback to second language (L2) researchers? To address this question, this study analyses a newly compiled corpus of 100 excerpts from research articles representing a variety of disciplines and submitted to the [anonymized] Translation Center for professional proofreading. The expert feedback was double annotated using an adapted version of the Louvain Error tagging Manual (Granger et al., 2022), which was expanded to capture a wider range of revision recommendations, including not only error corrections but also stylistic improvements, clarifications, and other suggestions aimed at enhancing the overall quality of the text. Each excerpt was also submitted to the free version of ChatGPT, using a carefully crafted prompt designed to provide relevant context. ChatGPT’s feedback was annotated following the same procedure as for the human proofreading. We then compared the number, types, focus, and appropriateness of the revisions provided by the human expert and ChatGPT, identifying points of convergence and divergence. This comparison sheds light on the current capabilities and limitations of AI-assisted academic editing and provides insights into the evolving contributions of human expertise in a writing landscape increasingly shaped by AI.
10:30-11am | Who’s Holding the Pen? Tracing AI in Undergraduate EFL Writing (TSLL25-133)
Febriana Lestari (Iowa State University🇺🇸) & Agustinus Hardi Prasetyo (Sanata Dharma University, Indonesia🇮🇩)
The emergence of GenAI these days has brought some debate among university writing instructors. In EFL contexts, this concern is amplified, as students often have limited exposure to writing beyond the classroom, with writing courses serving as their primary academic support. Meanwhile, AI tools like ChatGPT offer a shortcut, allowing students to complete assignments eloquently without the practice they need. This raises critical questions: What does “eloquent” writing actually mean? Do teachers understand their expectations of “academic” language within their contexts? And what parameters guide their judgments? To operationalize these parameters, the present study examines student writing quality before and during the GenAI era. Data come from three sub corpora: (1) pre-2022 student texts (pre-GenAI), (2) student texts from 2022 onward (during GenAI), and (3) ChatGPT-generated texts written in response to the same prompts. All texts are drawn from the Indonesian Corpus of Undergraduate Writing (ICReW; under development). Two rubrics are used: one for rating writing quality based on writing criteria and lexicogrammatical features, and one for assigning overall essay scores. Scores are then analyzed using many-facet Rasch measurement (MFRM) to account for rater severity. A comprehensive linguistic analysis using Multi-Dimensional Analysis (Biber, 1998; Staples and Egbert, 2019) is conducted to identify linguistic patterns across the three subcorpora. Comparative findings from both analyses will inform possible operationalizations of human- vs. AI-generated writing. Comparative findings from both raters’ perceptions and MDA analysis will inform possible operationalizations of human vs. AI-generated writing. This study is a work-in-progress and part of an on-going large-scale project developing ICReW, motivated by teacher concerns about AI use in writing, low English proficiency for the respective context (Lestari, 2024; EF EPI Index, 2024), the need to research-informed pedagogical decisions, and future learner corpus study in EFL contexts.
11-Noon | Corpus analysis of AI-generated language
Tony Berber Sardinha (Pontifical Catholic University of Sao Paulo (PUC-SP), Brazil🇧🇷)
In this talk, I present a corpus-based account of research on the language produced by Large Language Models (LLMs). I report on a set of studies comparing human and AI-generated texts across different contexts, using Multi-Dimensional (MD) Analysis (Berber Sardinha, in press-b; Berber Sardinha & Veirano Pinto, 2014, 2019; Biber, 1988; Veirano Pinto et al., in press) to identify the patterns of language use associated with each. I argue that Corpus Linguistics (Berber Sardinha, 2004, in press-a; Berber Sardinha & Barbara, 2009; Biber et al., 1998) is well-suited to investigate LLM output, as it enables the analysis of large volumes of textual data while attending to key linguistic attributes such as register, text, discourse, and situation of use (Berber Sardinha, 2024; Goulart et al., 2024; Mizumoto et al., 2024; Raffloer & Green, 2025).
The first studies look at the grammar of AI-generated texts. LLMs acquire grammar by processing examples from large collections of text, which is a remarkable achievement in itself. However, in doing so, they also absorb the typical characteristics of the texts included in their training data, most of which come from the web. These sources comprise an unbalanced range of registers, and the register properties of the data are rarely identified. As a result, LLMs tend to be register-unaware. When prompted to produce language in a particular register, they will deliver an output that looks like a good exemplar, yet they often fail to reproduce the grammatical patterns that define that register in actual human usage.
The remaining studies focus on the lexical patterns found in AI-generated texts, using Lexical Multi-Dimensional Analysis (Berber Sardinha & Fitzsimmons-Doolan, 2025). More precisely, the attention shifts to discourse analysis. Just as with grammar, LLMs acquire discursive patterns from the texts included in their training data. These patterns are later reinforced or suppressed during the fine-tuning phase. We will look at instances of AI-generated discourse from various sources, in addition to attempting to steer LLM reasoning through prompting.
1-2:30pm | A Critical Evaluation of GenAI as a Research Assistant: Three Case Studies from Corpus Linguistics
Nergis Danis , Tom Elliott , Andrea Flinn, Bethany Gray , Shangyu Jiang, Gi-Jung Kim , Shaya Kraut , Febriana Lestari , Chris Nuttall , Duong Nguyen , & Junghun Yang (Iowa State University🇺🇸)
Colloquium Overview
The prevalence of large language models (LLMs) and generative AI tools has increasingly led to proposals for (and sometimes evaluations of) the application of such tools to facilitate resource-intensive research tasks, including within corpus-based approaches to linguistic and text analysis (Zappavigna, 2023; Uchida, 2024; Curry et al., 2024). Assumptions about the ability of GenAI tools to facilitate language analysis likely stem from the fact that applications which are themselves built on LLMs and which can produce extensive quantities of human-like language are likewise able to analyze language. However, the extent to which generative AI tools can effectively and accurately take on research-based tasks related to language and text analysis that have traditionally required extensive human/manual effort or which are subjective/functional in nature have not been evaluated. In this colloquium, we critically evaluate the ability of a range of GenAI tools and LLMs to serve as a sort of ‘research assistant’ to corpus linguists. Each paper within this colloquium takes on a core task common in corpus-based research, develops prompts to engage a GenAI tool to assist with that task, and then evaluates the accuracy and/or reliability of the tool. After a brief introduction (10 minutes), three case studies (20 min each) will describe the core corpus-based task and how GenAI tools were prompted to assist in carrying out the task and detail the methods and results of the evaluation of the tool’s performance compared to a human gold standard. The colloquium concludes with a brief synthesis (10 minutes) and time for questions and discussion (10 minutes).
Introduction
Bethany Gray (Iowa State University)
Case Study 1: Fixtagging POS Tags
Andrea Flinn, Shaya Kraut, Febriana Lestari, & Duong Nguyen (Iowa State University)
Grammatical complexity features are common in specialized informational writing (Biber et al., 2011; Gray, 2011). They can be tagged automatically with tools such as the Biber tagger (Biber, 1998); however, errors occur. To improve tagging accuracy, Gray (2011) developed a tool (fixtagging) that coders can use to check features known to be error-prone (e.g., complement clause, relative clause, etc.). This case study explores ChatGPT 4.0’s potential for fixtagging eight categories of that-complement clauses. Twenty texts from the two-million word Corpus of Graduate Student Writing (CorGrad; Becker, 2022) were tagged using the Biber Tagger. Texts were then fixtagged in two ways: (1) manually using Gray’s software, and (2) automatically using ChatGPT 4.0. For the manual approach, two coders independently fixtagged the texts. Near-perfect intercoder reliability was calculated using Cohen’s Kappa. In the second approach, ChatGPT was trained using the that-complement clause framework. After several iterations, ChatGPT showed limited potential in detecting and correcting that tags (e.g., tht+vcmp+++, tht+rel+subj++). It was able to identify only a fraction of that-instances, pulled in random order, and it did not have high accuracy. Furthermore, the output often included that-instances not in the original text.
Case Study 2: Functional Coding of Key-Word-in-Context (KWIC) Lines
Nergis Danis, Tom Elliott, & Gi-Jung Kim (Iowa State University)
Motivated by the ongoing efforts to test the capability of AI tools for conducting corpus linguistic (CL) tasks (Curry et al., 2024), this case study aims to evaluate the ability of ChatGPT to code KWIC lines into functional groups. To this end, it replicates an existing analysis reported in Danis (2022) that coded KWIC lines consisting of instances of the first-person pronoun I and their co-occurring verb phrases. These verb phrases were coded into process type categories following the Systemic Functional Linguistic approach that divides verb phrases into six functional categories: material, mental, verbal, relational, existential, and behavioral. Known for its fuzzy nature and indeterminacy issues, coding process types is a demanding task, even among trained linguists (Gwilliams & Fontaine, 2015). One way to reduce the researcher’s workload for functional coding could be to use ChatGPT as a coding assistant or even as a second coder instead of another human researcher. Therefore, investigating ChatGPT’s capability for carrying out such coding offers significant potential for not just corpus linguists but other researchers wanting to use a functional linguistic framework in their investigations. The findings from this study can also be valuable for any KWIC line coding for functional purposes.
Case Study 3: Annotation of Text Excerpts for Target Linguistic Features
Shanyu Jiang, Chris Nuttall, & Junghun Yang (Iowa State University)
One task commonly performed in conjunction with corpus-based analysis involves the annotation of excerpts to highlight specific linguistic features in context. Depending on the type of analysis, this task can be a difficult process. Multi-dimensional analysis (MDA) is one such analysis. MDA often requires researchers to annotate a variety of features in a single text excerpt to demonstrate how they work in context to perform a specific function. Manually detecting these features demands substantial effort and time from researchers. Generative AI carries the potential to automate this task, reducing this time and effort. The purpose of this case study is to determine the viability of generative AI, specifically, ChatGPT, with respect to annotating text excerpts for multiple linguistic features. If AI proves to be a viable option for text annotation, researchers could devote more valuable focus to other analytical tasks. The outline below delineates the proposed methodological steps for determining the viability of Generative AI for annotating linguistic features in text excerpts.
3-3:30pm | ReflectGPT: Exploring ChatGPT-4’s potential for rhetorical analysis of written critical reflections (TSLL25-190)
Aysel Saricaoğlu (Social Sciences University of Ankara, Türkiye🇹🇷) & Selahattin Yilmaz (Yıldız Technical University, Türkiye🇹🇷)
The genre analysis capacity of ChatGPT has been explored for research articles, corporate social responsibility reports, and narrative short stories (e.g., Kim and Lu, 2024; Yu, 2025). However, little is known about its effectiveness in analyzing genres that are more widely used in higher education, such as critical reflections. Manual analysis of this genre poses unique challenges for researchers due to its highly context-dependent, personalized, and subjective nature and the lack of a standardized move-step model. This study explores the potential of a custom GPT model for rhetorically analyzing 140 critical reflections written by 70 undergraduate students for two reflective tasks in a counterbalanced design: source-based and non-source-based. We investigate (a) ReflectGPT’s accuracy in identifying the rhetorical moves in student-written critical reflections compared to human annotations and (b) the effect of task type on ReflectGPT’s accuracy. We manually annotated the reflections using four moves from Ryan and Ryan’s (2013) reflective scale. We then trained a custom version of OpenAI’s GPT4, ReflectGPT, using 80percent of our dataset for training purposes, 10percent for validation, and 10percent for test, a common practice for small domain-specific datasets (e.g., Yu, 2025). To evaluate ReflectGPT’s performance, we calculate precision, recall, and F1 score for all moves across the tasks. Data analysis is in progress. We expect the findings will contribute to our understanding of AI use for rhetorical analysis, particularly for complex genres such as critical reflections in the current study.
3-3:30pm | Register Appropriateness of ChatGPT-generated Academic Texts (TSLL25-79)
Yağmur Demir & Jesse Egbert (Northern Arizona University🇺🇸)
The rise of generative Artificial Intelligence (GenAI) tools such as ChatGPT has transformed language pedagogy (Baskara and Mukarto, 2023; El Shazly, 2021) and assessment (Shin and Li, 2023; Wodzak, 2022). Despite their growing use in academic contexts—from classroom materials to standardized testing—questions remain about the register appropriateness of the texts they produce. The humanlikeness of AI language must be defined not only by fluency or coherence (see…), but by register appropriateness—functional language use that aligns with the situational characteristics of registers. Building on the work of Biber (1995, 2014), Berber Sardinha (2024), and Myntti et al. (2025) this study investigates whether ChatGPT-generated academic texts mimic human-authored writing in two academic genres (journal articles and textbooks) across two disciplines (biology and history). Using multi-dimensional analysis, we analyzed 200 texts (100 AI-generated and 100 human-authored) along three linguistic dimensions: (1) specialized information density vs. non-technical synthesis, (2) definition/evaluation of new concepts, and (3) author-centered stance. Our results reveal a mixed picture: while ChatGPT exhibits moderate success in mimicking register distinctions found in journal article registers, its performance is notably less aligned with textbooks. ChatGPT-generated textbook excerpts in biology, for instance, often resemble the dense, technical style of journal articles, as a result failing to match the simplified, pedagogically oriented discourse found in human-authored textbooks. Our findings challenge the assumption that register-appropriate, human-like language naturally emerges from GenAI. Overall, they indicate that ChatGPT texts lack adequate functional appropriateness. Therefore, we urge further quantitative linguistic analyses of AI-generated language and emphasize the need for educators and test developers to exercise caution when using ChatGPT for content creation. This study also offers implications for language pedagogy and AI training.
3:30-4pm | A systematic analysis of the linguistic features that inform text detection by human reviewers (TSLL25-103)
Michelle Richter & Tove Larsson (Northern Arizona University🇺🇸)
Since its advent, ChatGPT has caused some apprehension due to its unprecedented impact on published academic writing, particularly when it comes to the use of generative AI (gen-AI) for writing research articles (Bisi et al. 2023). In fact, journal reviewers have been found to demonstrate negative bias against gen-AI content. Specifically, reviewers associate poor writing quality and vagueness with gen-AI abstracts, and good writing quality with human-authored abstracts (Gao et al., 2023). However, it remains unclear which linguistic features are reliable predictors of authorship and thus how reliably humans can actually detect gen-AI use. Without this piece of the puzzle, humans cannot be expected to uphold the expectations of originality in scholarly writing. The present study aims to identify linguistic features that are indicative of human versus gen-AI writing and test how accurate human journal reviewers are at detecting gen-AI. We answer the following research questions: (1) Which linguistic and stylistic features are reliable predictors of human/AI abstracts? (2) To what degree can reviewers of Applied Linguistics journals accurately detect AI abstracts? (3) To what extent are the features that are correctly identified by human reviewers predictive of authorship detection? We use key feature analysis (Egbert and Biber, 2023) on a corpus of 40 human-authored journal abstracts and 40 gen-AI-authored abstracts, along with survey methods. Early findings indicate that contractions and noun phrase complexity are strong predictors of authorship. Additionally, the reviewers from our pilot study (n=36) exhibited a relatively low degree of accuracy in detecting a gen-AI-produced abstract. They associate a low degree of specificity and unexpected vocabulary choices with AI abstracts and non-US spelling conventions with human-produced ones. However, these cues do not always align with accurate detection. Consequently, the study’s findings provide insights into how reviewers can make informed judgments when differentiating between human and AI writing.
3:30-4pm | The Impact of AI Literacy on Scholarly Skill Development (TSLL25-182)
Maryam Saneie Moghadam , Betsy Araujo Grando , Kristin Terrill , Lily Compton & Elena Cotos (Iowa State University🇺🇸)
Scholarly literature review for graduate thesis and dissertation writing is one of the most challenging and linguistically complex tasks novice scholars face, drawing on multiple elements of literacy from both the reading and writing sides. Emerging generative artificial intelligence (GenAI) technologies appear poised to revolutionize literature review processes by changing how students and scholars engage with published research. There is growing concern about incipient learning loss among the current cohort of rising scholars and fear that novices’ reliance on automated approaches could prevent the development of reading, argumentation, and critical thinking skills. Yet optimistic perspectives have also emerged, suggesting that integrating GenAI into literature review workflows could scaffold and, thus, promote deeper engagement. AI literacy is bound to be a deciding factor between these contrasting outlooks. Learners lacking AI literacy will likely fall into the trap of overreliance and credulity, mistaking automated re-mixing of source texts for insightful interpretation. On the other hand, learners with strong AI literacy could, potentially, leverage their technological competence to augment their communicative competence in academic English. In this talk, we will present an in-progress framework defining three AI literacy domains: algorithmic awareness, procedural competence, and evaluative judgment, along a developmental continuum from novice to expert. We will also present data from our 12-month project developing materials to teach AI-Facilitated Literature Review at the graduate level. Data collected from participants in two pilot workshops and two finalized workshops will be interpreted within this AI literacy framework. Evidence of participants’ learning will be presented, along with lessons learned from the project.
4-4:30pm | Comparative analysis of linguistic features and argument structures in AI-generated vs. L2 learner generated argumentative writing (TSLL25-167)
Ruiyang Dai & Cecilia Guanfang Zhao (University of Macau, China🇲🇴)
As generative AI tools become increasingly more prevalent in education, understanding how AI-generated writing compares to human writing becomes crucial. While previous studies have examined linguistic features and argumentative structures separately, few have explored their interaction in shaping argumentative essays. The present study addressed this gap by comparing 125 essays written by L2 English learners with 125 essays generated by GPT-4o, all in response to the same prompt. Sentences were first classified into Toulmin elements (e.g., claim, data, rebuttal) using rule-based pattern matching in Python, followed by manual verification. Each essay was also labeled with an overall argument structure type (e.g., two-sided complete argument, one-sided complete argument) using rule-based classification and human verification. Linguistic features, such as lexical complexity, syntactic complexity, and lexical bundles, were extracted using corpus tools including Lu’s (2010) Syntactic Complexity Analyzer and AntConc (Antony, 2024). Features that cannot be reliably detected through automated scripts—such as stance expressions, evaluative adjectives, and hedging or boosting devices—were manually annotated following functional coding guidelines drawn from prior research. Analysis showed that GPT-4o essays more frequently employed two-sided argument structures, often incorporating counterarguments and rebuttals within recurrent Toulmin patterns exemplified by sequences like claim, counterargument, data, rebuttal claim, and warrant. These patterns were realized through high syntactic complexity and repeated discourse bundles, suggesting a templated approach to argument construction. L2 argumentative essays typically followed one-sided structures and repeatedly used simple patterns such as claim, data, and warrant. However, they showed more rhetorical flexibility by incorporating stance markers and evaluative language, especially within claims and warrants. This contrast highlights how structural choices in both groups are closely linked to distinct rhetorical and linguistic strategies. The findings may inform writing pedagogy, AI-assisted feedback design, and the evaluation of AI-generated texts in second language education.
4-4:30pm | Generative AI in Dissertation Writing: L2 Doctoral Students’ Attitudes, Self-reported Use, and Perceived Training Needs (TSLL25-178)
MohammadHamed Hoomanfard & Yaser Shamsi (Oklahoma State University🇺🇸)
Generative Artificial Intelligence (GenAI) has been extensively employed by L2 doctoral students to assist with their dissertation writing. However, little is known about how these students engage with GenAI in this process. To address this gap, we conducted a qualitative study exploring L2 doctoral students’ attitudes, self-reported use, and perceived training needs. We conducted interviews with 54 doctoral students from various departments at a public university in the Central Southern United States, all of whom were in the process of writing the initial draft of their dissertation manuscripts. To examine students’ perceptions, we applied both thematic and sentiment analysis to the interview data. Our analysis revealed several key themes. Participants used both positive and negative descriptors when discussing GenAI in the context of dissertation writing, yet the overall sentiment toward its use was positive. One prominent theme that emerged was the variety of ways in which L2 doctoral students engaged with GenAI tools, with 19 distinct purposes identified across three broad functions: exploration, confirmation, and execution. A significant theme centered around AI-giarism, particularly the ambiguity surrounding the boundary between legitimate GenAI use and plagiarism, as well as the ethical dilemma students faced regarding whether to disclose their use of GenAI in their dissertations. Another theme that surfaced was the perceived need for training, with students expressing a strong preference for learning about the available tools and their functionalities, mastering effective prompting techniques, addressing concerns about plagiarism, and navigating issues related to data privacy.
Friday, November 7th
8-8:30am | Artificial Intelligence-Enhanced L2 Writing Development: Investigating the Impact of Cognitive Load, Learner Engagement, and Feedback Personalization (TSLL25-86)
Hanieh Shafiee Rad (Shahrekord University, Iran🇮🇷)
This study explores the impact of artificial intelligence (AI)-enhanced instruction on second language (L2) writing skills, focusing on the integration of new variables such as cognitive load, learner engagement, and feedback personalization. While previous research has highlighted the potential of AI tools in language learning, limited attention has been paid to how specific learner-centered variables mediate the effectiveness of AI-driven writing support. Addressing this gap, the current study investigates how AI-enhanced platforms featuring adaptive feedback, emotion-aware interaction, and multimodal support affect L2 writing development in university-level EFL learners. A quasi-experimental design was employed involving 120 participants from two intact classes. The experimental group received writing instruction through an AI-supported platform integrating real-time, personalized feedback and cognitive-emotion tracking, while the control group engaged in traditional writing activities without AI assistance. Data were collected through pre- and posttests, surveys measuring cognitive load and engagement, and semi-structured interviews. Results revealed that the experimental group significantly outperformed the control group in writing fluency, coherence, and lexical diversity (p < .01). Furthermore, learners exposed to AI-enhanced instruction reported lower cognitive load and higher engagement, particularly in relation to the responsiveness and personalization of the feedback. Interview responses further confirmed that students perceived the AI tools as supportive in guiding their writing process and improving their confidence. The findings suggest that incorporating cognitive-affective variables into AI-mediated writing instruction can enhance the effectiveness of L2 writing development. This study contributes to the growing body of research on AI in language education by emphasizing the importance of personalized, learner-responsive environments. Pedagogically, it calls for the thoughtful integration of AI tools that not only automate feedback but also adapt to learners’ emotional and cognitive states, offering a promising direction for more inclusive and effective L2 writing instruction.
8-8:30am | English for Academic Purposes (EAP) Tutors’ Practical Pedagogical Knowledge for (Generative) Artificial Intelligence (AI) (TSLL25-41)
Zoe Handley (The University of York, UK🇬🇧)
This paper reports the results of an interview study which explored what English for Academic Purposes (EAP) tutors know about generative Artificial Intelligence (AI) and how it might be deployed to support teaching and assessment when working with international students for whom English is a second language. Eleven tutors (9 male, 2 female) working in the UK participated in the study. The tutors who participated in the interviews were curious about generative AI and had spent much time exploring its capabilities. They suggested that generative AI raises fundamental questions about assessment and what it means to study at an English medium university, particularly, the role of language proficiency. Concerning assessment, they reported that the focus of policies had shifted from plagiarism to authorship. While several tutors recognized some ways in which generative AI might be deployed to support students for whom English is a second language, it was felt that assessment policies needed to be clarified before these new practices could be adopted.
8:30-9am | Beyond Human-AI Agreement: Construct and Criterion Validity of GenAI Systems for Automated Essay Scoring (TSLL25-114)
Presentation withdrawn
8:30-9am | Pre-Service English Language Teachers’ Detection of AI-Generated Texts in L2 Writing (TSLL25-139)
Gülfem Büyükaltunel & Sibel Söğüt (Sinop University, Türkiye🇹🇷)
Misdetection of Artificial Intelligence (AI) generated content in student writing raises concerns for academic integrity among both students and teachers. The existing AI-detection tools may provide false positives and negatives. The misdetections and bias could lead to a climate of anxiety and distrust in academic settings (Giray, 2024). Thus, human agency remains crucial, and teachers have a particularly important role in distinguishing between texts written by students and generated by AI. This study addresses specific linguistic and content-related detection strategies used by pre-service English language teachers and aims to reveal sources of their false positive and negative detections. In this ongoing action research, the participants are a total of 25 pre-service English language teachers in the Turkish higher education setting. Data is collected through demographic information (background information and experiences in using GenAI tools), and open-ended questions about the linguistic, contextual, and content-related strategies in recognizing characteristics of student writing and AI-generated texts. The pre-service teachers are presented with a collection of argumentative essays written by students and generated by AI on the same prompt and asked to provide their detections along with underlying reasons. The data gathered from these sources is analyzed through quantitative and qualitative methods. As a result, we aim to present potential biases of prospective English teachers, their justifications and underlying reasons for their correct and incorrect detections. Based on the findings, practical implications for L2 writing and suggestions for teacher training programs will be presented.
9-10am | Using Generative AI technology in language education: Promise and challenges
Jing Xu , Yasin Karatay , & Hye-won Lee (Cambridge University Press & Assessment, UK🇬🇧)
Since the public release of ChatGPT in November 2022, Generative AI (GenAI) has become a buzzword in various professions. The technology has been widely deemed as a handy tool for reducing labour-intensive work done by humans. The promise of GenAI for language education has been recognised by applied linguists, too. Many view the technology as a “game changer” for second language learning and assessment but concerns around responsible and ethical use of AI remain (Chapelle, 2024, 2025; Xi, 2023). In this talk, we will discuss the promise and challenges of using GenAI in three areas of language education.
The first area is automated scoring of constructed responses. Developing automated scoring systems typically requires a sufficient amount of training data and expertise in machine learning (ML). To address the problem of data shortage, GenAI has been used to generate synthetic L2 data. Additionally, there is an increasing body of research on fine-tuning large language models (LLMs) for automated scoring. Compared to traditional ML techniques, this new approach is more accessible to nonexperts, has a lower demand on training data, and may increase the explainability of automated scores to some extent. However, there are outstanding validity and reliability issues related to LLMs’ lack of knowledge of learner language and the potential bias in LLMs.
The second area is using GenAI in spoken dialogue systems (SDSs) for speaking assessment and practice. GenAI has fuelled a desire to move beyond monologic tasks toward dynamic scenario-based oral interactions and context-sensitive conversations tailored to user profiles. However, challenges remain in ensuring construct validity and test fairness in SDS-mediated tasks and developing relevant ethical and validation frameworks.
The third area is using GenAI in integrated learning and assessment (ILA). A GenAI agent can provide scaffolds for learning, personalised content and practice, and learning-oriented assessment and feedback. It holds promise for enhancing learner engagement and motivation, fostering learner autonomy, and supporting multilingual and culturally inclusive learning. However, concerns remain that AI may generate inaccurate or outdated information, lack cultural nuance, and potentially hinder critical thinking and creativity. We will conclude our talk by identifying research priorities in each area.
10:30-11am | Textual Parameters of LLM-Generated Reading Passages for Assessment: A Computational Analysis of Readability Metrics (TSLL25-109)
Norazha Paiman (Universiti Kebangsaan Malaysia, Malaysia🇲🇾)
This study investigates the potential of Large Language Models (LLMs) as tools for automated item generation in reading comprehension assessments, with specific focus on textual complexity parameters and alignment with established proficiency frameworks. The study examines whether AI-generated texts demonstrate sufficient linguistic validity for educational assessment purposes. The research employed a quantitative analytical framework wherein two contemporary LLMs—ChatGPT-4 and Claude 3.5—were systematically prompted to generate a corpus of 30 reading passages (n=15 per model) targeting CEFR C2 proficiency level across 15 academic disciplines (themes). These texts underwent multidimensional computational analysis incorporating: (1) standardized readability metrics, (2) corpus-based lexical frequency profiling, (3) text statistics examination, and (4) CEFR level verification via Text Inspector. The analysis framework focused on cross-platform comparative evaluation of textual complexity parameters, consistency in academic language usage, and alignment with established proficiency benchmarks for advanced-level reading assessment materials. Results reveal that both LLMs consistently produced texts at advanced academic levels, with ChatGPT demonstrating higher complexity metrics compared to Claude. Quantitative analysis of lexical frequency profiles showed statistically significant correlations (r=.78, p<.001) between intended and actual CEFR levels, with 93.3percent of ChatGPT and 86.7percent of Claude outputs meeting C2+ standards. While both models met established academic benchmarks for vocabulary distribution, ChatGPT exhibited greater consistency in readability metrics than Claude. However, Claude demonstrated more natural language variation and balanced vocabulary distribution, offering potential benefits for assessment authenticity. These findings indicate that LLMs can reliably generate assessment-quality reading passages that align with standardized linguistic parameters. This research contributes to the nascent field of AI applications in language testing by establishing empirically-grounded parameters for the implementation of automated item generation in high-stakes assessment contexts. Future research will expand this preliminary investigation through psychometric validation of test items developed from these passages, including differential item functioning analysis across diverse learner populations.
10:30-11am | Enhancing EFL Students’ Writing Through ChatGPT: Utilizing Effective Prompts for Feedback and Revision (TSLL25-119)
Trang Ho & Duong Nguyen (Iowa State University🇺🇸)
This study explores how writing instructors can support A2-B1 CEFR-level EFL undergraduates in using a structured prompt to elicit useful feedback from ChatGPT for paragraph writing. Specifically, it examines how students are trained to apply a prompt that combines the delimiter prompting technique with the Constructivist framework (McGuire et al., 2024) to receive targeted and meaningful feedback. The aim is to offer evidence-based pedagogical insights for integrating ChatGPT into writing instruction. An explanatory sequential mixed methods design (Creswell and Clark, 2011) was employed. Twenty-six Vietnamese first-year students wrote five paragraphs over five weeks, used the structured prompt provided in their L1 to engage with ChatGPT, revised their work, and reflected on grammar and vocabulary gains. Writing performance was assessed through holistic and analytic scoring of pre- and post-tests, with a paired sample t-test conducted to measure progress. The quantitative results informed the selection of participants for follow-up semi-structured interviews. Qualitative data from interviews, screen recordings, and teacher observations were analyzed through inductive thematic coding (Avineri, 2017) to explore students’ experiences and perceptions of ChatGPT-generated feedback. Preliminary findings suggest improvements in grammar and vocabulary use, and generally positive attitudes toward ChatGPT feedback. From the teacher’s observation, students demonstrated increased awareness of common grammatical issues (e.g., subject-verb agreement, plural nouns), collocations (e.g., verb-noun and adjective-noun combinations), and paragraph structure, particularly in crafting topic and concluding sentences. However, their interaction with ChatGPT beyond the initial prompt remained limited despite instructor encouragement. This study offers practical strategies for designing prompts that support effective AI feedback and foster learner autonomy in EFL writing revision.
11-11:30am | Investigating AI Accuracy in Rating Korean Passage Difficulty for Language Material Development (TSLL25-116)
Jean Young Chun & Jameson Moore (Defense Language Institute Foreign Language Center🇺🇸)
Recent advancements in artificial intelligence (AI) have significantly enhanced the efficiency of developing language teaching and testing materials. One area that stands to benefit substantially is the evaluation of passage difficulty levels. Accurately selecting texts aligned with learners’ proficiency levels is a fundamental tenet of effective material development, traditionally requiring extensive teaching experience and familiarity with standardized rating scales. While AI tools hold great promise in facilitating this process, their ability to assess passage difficulty—especially in non-English languages—remains underexplored. To address this gap, the present study investigates the extent to which two AI systems—ChatGPT and Copilot—can accurately evaluate the difficulty levels of Korean passages. It further examines whether significant differences arise between the tools and across passage difficulty levels. Additionally, the study explores the extent to which each tool can provide relevant justifications to support its level assignments. ChatGPT-4o and Copilot were initially trained by the researchers using level descriptors from the Interagency Language Roundtable (ILR) scale and example texts, specifically focusing on Levels 1, 1+, 2, 2+, and 3— the five primary proficiency levels targeted in the current Korean language program. Following this training, the tools were tasked with assigning difficulty levels to 100 Korean passages (20 per level), whose ILR levels had been predetermined, and providing accompanying justifications. A mixed-effects logistic regression model was performed to compare the accuracy of level identification by AI tool and ILR level. The justifications were qualitatively analyzed to determine whether they reflected key linguistic features associated with each ILR level. The findings contribute to the growing body of research on AI-assisted language teaching and testing material development by offering insights into the potential of AI tools to support the creation of level-appropriate materials, particularly in less commonly taught languages such as Korean.
11-11:30am | Zero-shot and engineered prompts for ChatGPT-generated feedback in L2 writing development (TSLL25-136)
Duong Nguyen (Iowa State University🇺🇸)
Research on GPT-based feedback for L2 writing has grown rapidly, but its effectiveness remains contested due to issues like overload, inconsistency, inaccuracy, overreliance, and reduced critical thinking (Barrot, 2023). A central question is how to design prompts that mitigate these limitations while leveraging AI’s strengths. This study compares the effects of GPT-generated feedback using a zero-shot (role-play) prompt versus an engineered prompt. A convergent parallel mixed-methods design was employed. Forty first-year undergraduates (aged 18–20, A2–B1 level) were randomly assigned to one of two prompt conditions: a zero-shot prompt or an engineered prompt incorporating delimiter prompting and a structured framework (McGuire et al., 2024). Over five weeks, participants wrote five essays, used GPT-4 via BChat to generate feedback, engaged with it, and reflected on their learning. Student writing was assessed through pre- and post-tests rated with a rubric and analyzed using repeated-measures ANOVA (RM-ANOVA) to examine group differences over time in overall proficiency and sub-skills (e.g., grammar, vocabulary). Feedback quality was evaluated by coding ChatGPT-generated feedback using a 3-point Likert scale based on seven principles of effective feedback (Nicol and Macfarlane-Dick, 2006), with results analyzed via RM-ANOVAs. Interaction logs were reviewed to examine students’ engagement with feedback (i.e., accept, resist, neglect). Preliminary findings suggest both prompts produced comparable feedback, challenging concerns about the zero-shot prompt. Specifically, both offered useful feedback on logic and idea development. The zero-shot prompt was more encouraging, noting both strengths and weaknesses, while the engineered prompt focused more on errors. However, students engaged more actively with feedback from the engineered prompt. The results offer insights into how prompting strategies can shape GenAI-generated feedback and inform ways to enhance its pedagogical value for L2 learners.
11:30-Noon | Exploring the Impact of Student-Generated Questioning on Writing Development (TSLL25-92)
Angelie Ignacio , Hajung Kim , Han Lai , & Eunice Eunhee Jang (University of Toronto, Canada🇨🇦)
Student-generated questioning (SGQ) promotes critical thinking and learner engagement, yet its role in elementary writing development remains underexplored (Maplethorpe et al., 2022). This study investigates whether (1) the quality of student-generated questions (SGQs)—scored by both human raters and a generative AI model (ChatGPT)—predicts students’ writing performance, and (2) whether SGQ quality improves when video stimuli include subtitles. Elementary students completed two integrated writing tasks (on social media and global warming), each preceded by a short video—only one with subtitles. After viewing, students generated three questions, planned, wrote, and revised an essay. In Phase 1, 1,800 SGQs were rated using a 4-point typology (irrelevant to critical). Each SGQ was scored by two human raters and was also evaluated using OpenAI’s ChatGPT, leveraging generative AI through prompt engineering to produce machine scores aligned with the human rating rubric. This GenAI application is central to the study: it not only assists in scoring but is validated as a moderator—an impartial third scorer to resolve discrepancies between human raters. In Phase 2, SGQ scores will be used in a Path Analysis to examine whether SGQ quality predicts writing performance, controlling for working memory and non-verbal reasoning. A Wilcoxon signed-rank test will compare SGQ quality between the subtitle vs no-subtitle conditions. This study advances writing assessment by integrating SGQ into literacy education and by demonstrating a novel use of generative AI as a scoring and moderation tool. It showcases how GenAI can enhance consistency, scalability, and fairness in student evaluation—marking a significant step toward human-AI collaboration in educational research and practice.
11:30-Noon | Native and Non-Native Learners’ Engagement with ChatGPT Feedback: A Longitudinal, Comparative Study (TSLL25-154)
Inyoung Na , Mahdi Duris , & Volker Hegelheimer (Iowa State University🇺🇸)
Growing interest in generative AI (GenAI) as an Automated Writing Evaluation (AWE) feedback tool has demonstrated the immediate benefits of AI-generated feedback for writing support (e.g., Escalante et al., 2023; Song and Song, 2023). However, much of the existing research relies on short-term, pre/post-test designs and provides limited qualitative insight into how learner engagement with GenAI feedback evolves over time and across proficiency levels. Given that the effectiveness of such feedback depends heavily on how learners interact with it, closer examination of engagement patterns is essential for understanding its pedagogical potential (Stevenson and Phakiti, 2019). This study investigates longitudinal learner engagement with ChatGPT-4o over an eight-week period, involving three distinct learner groups: lower-level ESL learners (n = 5), higher-level ESL learners (n = 4), and native English-speaking undergraduates (n = 4). Data sources included screen recordings, weekly check-in questionnaires, reflective learning logs, and ChatGPT-learner conversation logs. The mixed-methods approach examined how learners prompted ChatGPT, applied its feedback, and developed revision strategies over time. Findings reveal clear differences in initial engagement. Lower-level ESL learners tended to rely heavily on ChatGPT for rewriting and correction, often accepting suggestions with minimal modification. While similar in their tendency to implement feedback, higher-level ESL learners demonstrated greater prompting specificity and selectivity, frequently adapting suggestions to match their intended tone and rhetorical goals. Native speakers, by contrast, tended to use ChatGPT diagnostically, asking for suggestions before independently revising. Over time, learners across all groups exhibited increased strategic awareness, gradually shifting from passive acceptance to more reflective and purposeful engagement with feedback. This study contributes to the field by mapping how AI literacy develops and how different forms of engagement ultimately impact writing development and learning outcomes. Implications are discussed for learner training and instructional design to support the effective integration of GenAI tools into L2 writing pedagogy.
1-1:30pm | How Do GenAI Chatbots Use Language to Construct Human-Like Personas? (TSLL25-126)
Şebnem Kurt , Agata Guskaroska , Ahmad Zubaidi Amrullah , Saime Esma Masca , Fatemeh Bordbarjavidi , Inyoung Na , Jeanne Beck , Jiwon Choi , Nergis Danis , & Carol Chapelle (Iowa State University🇺🇸)
Chatbot developers and prompt engineers learn to specify the persona or role that the chatbot is expected to play in conversations with humans (e.g., TRACI, 2025). The chatbot’s human-like persona is then constructed on the fly through the language it displays in interaction with a human, who contributes to a process of co-construction of the identities for both. Any analysis of the success of GenAI persona creation rests in the language presented by the chatbot in such conversations. This paper is part of a larger study investigating how GenAI chatbots use language to create their human-like personas in conversation with humans. From a collection of over 200 episodes in conversations with chatbots, we examined a sample of 67 communicative exchanges between a GenAI chatbot and a human who was attempting to increase their productivity by asking the chatbot to deliver information or produce text, for example. The goal was to identify the linguistic devices in these productivity-oriented conversations that the chatbots used to construct human-like personas. The study was conducted using methods for context-based qualitative discourse analysis (Martin and Rose, 2007) and constructs such as speech acts (Searle, 1976) and maxims of conversation (Grice, 1975). Findings indicated that the chatbots fulfilled their roles in such exchanges by not only displaying the requested result but also enacting pragmatic discourse functions that displayed their human-like character. For example, in productivity-related exchanges the chatbots used metadiscourse (Hyland, 2013) to refer to the language of the conversation text (“I can guide you through the next steps or suggest alternatives!”), interpersonal language to express an epistemic stance (“It seems like your message got cut off…”) or attitude (“I’ll be happy to help further! :)”). We will show how these linguistic devices helped to construct personas and identify these and other linguistic dimensions of GenAI chatbot performance with examples from the data. By providing an initial analysis of the pragmatic resources chatbots use to construct their personas while delivering requested information services, we show the importance of a functional linguistic perspective in building knowledge about GenAI personas.
1-1:30pm | Building Prompt Literacy: Empowering EFL Teachers with A Model-Based Approach to Prompt Writing (TSLL25-160)
Sevcan Bayraktar Çepni (Trabzon University, Türkiye🇹🇷)
Within the framework of a project sponsored by the Regional English Language Office (RELO) of the U.S. Department of State in Turkey, this paper reports one part of a teacher training initiative aiming to improve the prompt-writing skills of English as a Foreign Language (EFL) teachers using AI tools. 40 volunteer EFL teachers serving in different levels of education (from primary to university) participated in a thorough training course with an emphasis on artificial intelligence integration into language instruction. One component of the training program was “Model- Supported Prompt Writing Training” in which participants were exposed to the PARTS and TATOO models—frameworks meant to facilitate efficient prompt composition customized to language teaching environments. Teachers participated in hands-on exercises throughout the course where they produced AI prompts consistent with their teaching goals to generate educational materials. An open-ended questionnaire was used to gather participants’ opinions to shed light on how their perceptions on prompt writing were shaped after the training. The elicited qualitative data was subjected to content analysis, revealing several emergent themes including awareness of prompt specificity, confidence in integrating AI into pedagogy, and the transformative impact of structured prompt models. The results show a notable change in teachers’ perspective on artificial intelligence prompting, therefore stressing the need for specific training in prompt literacy. This study has implications for the next teacher training initiatives meant to include artificial intelligence into pedagogically significant contexts into language instruction.
1:30-2:30pm | Generative AI Feedback for Language Learning: Possibilities, Challenges, and Future Directions
Sanghee Kang (Carnegie Mellon University🇺🇸)
Traditionally, human assessors such as teachers and peers have been the primary providers of feedback in language classrooms (Zou et al., 2024). With the development of digital technologies, however, feedback provision has expanded to include automated systems capable of delivering corrective feedback (CF) on learner production. More recently, generative artificial intelligence (GAI) technologies have opened new possibilities for providing personalized and interactive feedback (Yan, 2024). However, despite growing interest in GAI feedback for language learning, important questions remain about how it can be effectively integrated into the language learning environment, particularly regarding its pedagogical effectiveness, learner perceptions, and implications for instructional design.
In this talk, I will explore the potential and challenges of GAI as a source of CF by drawing on my recent classroom-based studies that illustrate possible applications of GAI feedback in instructional contexts. The first project examined GAI as an asynchronous feedback provider in student writing, while the second project investigated GAI as a conversational partner that provided CF during text-chat interactions. I will share findings related to how learners engage with GAI feedback, the extent to which it may support language development, and how learners perceive its usefulness and reliability. I will conclude my talk by highlighting possibilities and challenges of GAI feedback in language classrooms and outlining future directions for research and implementation in classroom settings.
3-3:30pm | Investigating the effects of training and prompt contextualization on ChatGPT’s pragmatic performance (TSLL25-51)
Tetyana Sydorenko (Portland State University), Judit Dombi (University of Pécs, Hungary), Veronika Timpe-Laughlin (Educational Testing Service), Saerhim Oh (Educational Testing Service), & Rahul Divekar (Bentley University🇺🇸)
Although Generative AI (GenAI) tools like ChatGPT display advanced capabilities in text generation and reasoning, their less impressive performance in conversational tasks suggests that LLMs are primarily trained on written text and designed to follow certain parameters, such as being unfailingly polite, and overly formal and helpful. However, it may be possible to train GenAI tools so that they produce output more appropriate for a given task or context (Wei et al., 2021). In this paper, we explore how various prompt contextualizations and training methods affect ChatGPT’s output in terms of pragmatic appropriateness in specified scenarios. We compared ChatGPT’s pragmatic performance to human performance. Our baseline dataset included 49 human-human spoken role-play interactions, in which English learners needed to ask a researcher, acting as a workplace supervisor, to schedule a meeting and review their slides before a presentation. We then tested the performance of ChatGPT as the interlocutor in the same role-play scenario. We fed language learner turns (from the aforementioned dataset) into ChatGPT and then compared the human output with ChatGPT’s responses in terms of pragmatic phenomena. Additionally, we gradually contextualized the prompt to ChatGPT in various ways (via a dialog tree; additional information in text) and included local training data to explore whether the different steps resulted in more pragmatically appropriate responses from ChatGPT. We discuss how the iterative contextualizations changed ChatGPT’s output in terms of pragmatic phenomena. Although the contextualizations improved some aspects of ChatGPT’s pragmatic performance, other issues remained: ChatGPT’s unfailing enthusiasm, which is not always sociopragmatically appropriate, and its inconsistency. We discuss the implications of these findings for L2 pragmatics teaching and assessment.
3-3:30pm | Designing Feedback and Assessment for Gen-AI- Assisted Digital Multimodal Composition in L2 (TSLL25-135)
Sibel Söğüt , Ergün Cihat Çorbacı (Sinop University, Türkiye🇹🇷), & Volker Hegelheimer (Iowa State University🇺🇸)
The rapidly evolving use of GenAI leads to a crucial need for new approaches to instruction and assessment of language skills in digital multimodal composing tasks (DMC). There is a need for updating the conventional single mode L2 writing tasks and reconceptualizing feedback and assessment that reflect the evolving nature of writing practice and the effective integration of GenAI. In this action research project, we address this need and focus on the production of practical feedback and assessment applications that effectively reflect the unique characteristics of GenAI-assisted DMC tasks. Within a 12-week language instruction in a Turkish higher education context, we examine the integration of Gen-AI tools into DMC tasks. This instruction includes the design and implementation of a novel feedback and assessment framework designed specifically for Gen-AI-assisted DMC that incorporates activities such as simulated conversations, writing, and reading tasks and multimodal elements. Data is collected through a needs analysis, pre- and post-surveys (with critical reflection components), and interviews. The learners engage in L2 instruction that integrates co-authoring with Gen-AI for brainstorming and idea generation, using GenAI tools for L2 DMC tasks, such as simulated conversations, real-world writing projects (infographics, storyboards, scripting, etc.), multimodal essays, and social media posts. We conduct an initial design, expert consultation and pilot testing with learner work to gather feedback on clarity and applicability. The expected outcomes of this study include: (1) a practical framework for integrating Gen-AI tools into L2 DMC instruction; (2) a set of assessment tools (rubrics, checklists, e-portfolio design) that can be used by language learners and educators. This study will present a novel contribution to the current literature through a framework of assessment and feedback that integrates critical AI literacy, learner agency, and authentic language practice through GenAI-assisted DMC tasks.
3:30-4pm | Humanizing AI in ELT through Transperipheral Gaming Spaces (TSLL25-144)
Rodrigo Costa dos Santos (Universidade Federal Fluminense, Brazil🇧🇷)
This presentation draws on findings from my doctoral research (dos Santos, 2024), which examined digital game-mediated literacy practices and identity construction among residents of Salgueiro, a peripheral neighborhood in São Gonçalo, Brazil. Through technobiographies—autobiographical and semi-structured narratives—I analyzed how gaming intersects with material conditions, territory, and digital literacies to shape what I call transperipheral subjectivities (Windle et al., 2020). These subjectivities reflect how learners in marginal contexts not only access and adapt digital tools, but also re-signify them in ways that resist dominant discourses. Building on these findings, I propose a humanizing perspective on Artificial Intelligence (AI) in English Language Teaching (ELT). Rather than treating AI as a neutral instructional aid, I frame it as a sociolinguistic actor (Blommaert, 2010; Pennycook, 2006), whose effects are felt unevenly across sociotechnical geographies. I pay particular attention to gambiarras—creative improvisations rooted in Brazilian peripheral life—as semiotic strategies for interacting with AI-driven features like automated moderation or adaptive feedback in digital games. This approach is grounded in a decolonial framework (Mignolo and Walsh, 2018) and a critical, transgressive understanding of Applied Linguistics (Moita Lopes et al., 2014), which centers learner agency, social positioning, and the politics of access. Transperipheral gaming spaces illuminate how AI tools are not merely used but repurposed and recontextualized by learners from the margins. By drawing on this lived experience, I argue for AI designs in ELT that are more culturally responsive and socially situated. This work will contribute to generate insights into how peripheral practices can inform more equitable language technologies and how to engage learners as co-constructors of meaning, identity, and possibility within digitally mediated worlds.
3:30-4pm | Multiliteracies in action: GenAI and DMC for multilingual learners of English through technology-mediated TBLT (TSLL25-173)
Alexander Tang , David Honeycutt , & Nicole Ziegler (University of Hawaiʻi at Mānoa🇺🇸)
The GenAI era, exemplified by tools like DALL-E, is reshaping digital multimodal composing (DMC) in language education. DMC is the process of creating texts that combine multiple models of communication using digital tools, which foster learners’ agency, creativity, and collaborative practices (Kessler, 2024). While GenAI can offer rich semiotic resources—images, text, and sound—there remains a gap in scaffolding learners’ engagement in ways that support language development through process-oriented, student-centered, and multiliterate processes. Drawing on principles of technology-mediated TBLT, this study explores how communicative tasks using GenAI can foster learners’ linguistic and multimodal competence. This study investigates how multimodal GenAI can be used to support multilingual learners in developing critical multiliteracies and language engagement. Intermediate English learners (N = 15) completed a series of GenAI-assisted DMC tasks in which they collaboratively developed travel guides. Learners completed six real-world narrative tasks about campus life, during which learners co-created a storybook about their favorite place. Data include student artifacts (e.g., eBooks, documentaries), learners’ weekly reflection journals, and semi-structured interviews, and demonstrate that learners engaged in translanguaging and collaborative meaning-making. Language-related episodes (Torres and Yanguas, 2021) were used to assess cognitive, behavioral, affective, and social engagement, while thematic analysis identified patterns in learners’ use of GenAI for translanguaging and multimodal construction. Preliminary findings suggest complex patterns in terms of learners’ engagement, and demonstrate that task-supported DMC can transform learners’ literacy practices, linking pedagogy and real-world application (Kessler, 2024). The findings underscore the potential of GenAI to develop multilingual learners in negotiating meaning through TMTBLT, bridging linguistic and cultural divides while fostering critical engagement with DMC.
Saturday, November 8th
7:30-8am | Impact of Verbal Entries on Multimodal Pictures Generated by Selected Artificial Intelligence Art Websites (TSLL25-46)
Mohamed Abdelwahab Amer (Egyptian Russian University, Egypt🇪🇬)
The recent vast spread of Artificial Intelligence (AI) art generators has helped learners and educators – amongst a wide range of professionals – generate and recreate spectra of art products including – but not limited to – pictures, 3D designs, videos, animations, and paintings, etc. This is not to mention the solid educational programs and curricula these generators can create from scratch with simple entries by any user. The output differs from one user to another, and depends on the verbal nature of the entry itself. For example, one user can input a simply-articulated sentence constituting few words in order to generate a wholly new visual or audiovisual output which mainly depends on the lexico-morphological components of the input itself. The monomodal element hence can be converted into a multimodal system, yet the latter is not totally controlled by the former. This paper attempts to examine the impact of verbal entries on multimodal pictures generated by selected AI websites, and the verbal and social determinants that control some features of the output. It hypothesizes that the selection of some social and linguistic actors can intentionally change the multimodal nature of the output. The paper also manifests actual experiments on selected websites that generate multimodal AI outputs in which the results are controlled by the social intervention of the user.
8-8:30am | Piloting “LinguAI Coach”: An AI Chatbot for Training EFL Teachers to Use AI Tools for Linguistic Skills Development (TSLL25-168)
Maria Perifanou (University of Macedonia & Aristotle University of Thessaloniki, Greece🇬🇷)
As artificial intelligence (AI) reshapes education, there is an urgent need to prepare language teachers to engage critically and creatively with AI-enhanced pedagogy. While AI tools are increasingly accessible, few teacher education programs offer structured guidance on how these tools can meaningfully support second language instruction, particularly in the development of learners’ linguistic skills. This paper reports on the pilot testing of the “LinguAI Coach”, an AI-guided pedagogical chatbot designed to scaffold language teachers in developing AI literacy, pedagogical creativity, and critical task design. Grounded in Design Thinking and Task-based Language Teaching (Fleury et al., 2016; Murphy, 2003), the chatbot positions AI not as a content delivery mechanism, but as a co-designer that supports teachers in crafting classroom tasks across core linguistic skills (i.e., reading, writing, speaking, listening). The pilot study engages both pre-service English language teachers enrolled in a bachelor’s program and in-service teachers attending a master’s-level course. Participants interacted with the “LinguAI Coach”, hosted on the “Mizou” platform, in student mode and completed one or more micro-design cycles involving skill selection, AI tool experimentation, and structured pedagogical reflection. This exploratory work contributes to the growing body of research on AI in teacher education by offering a replicable model for embedding chatbot-based design support into language teachers’ training. Early findings suggest that the “LinguAI Coach” fosters teacher creativity and critical reflection on AI tool integration, while also highlighting the need for more adaptive scaffolding and personalized feedback.
8:30-9am | How does Generative AI IDLE speaking practice influence intrinsic motivation, self-efficacy and performance? (TSLL25-60)
Jing Zhang (Universiti Sains Malaysia, Malaysia🇲🇾)
Emerging discussions have been emerged with regard to recent trends and applications of chatbots in education, but few attention were paid to GenAI speech-recognition chatbots in terms of L2 speaking. To address such a research gap, the research draws on the theoretical framework of the ARCS model to investigate the impacts of GenAI-assisted Informal Digital Learning of English (IDLE) speaking practice on intrinsic motivation, speaking self-efficacy, and speaking performance among vocational college students. Based on a quasi-experimental design, results showed that statistical differences were found in participants’ intrinsic motivation, interest-enjoyment, and perceived value in particular. The speaking performance of the experimental group also witnessed significant progress after the intervention. However, GenAI-IDLE speaking practice did not have a significant influence on speaking self-efficacy. Moreover, students perceived that they could boost learning confidence by facilitating language skills and broadening the knowledge reserve from the GenAI-IDLE speaking practice from semi-structured interviews. To conclude, this study will shed light on the effectiveness of GenAI chatbots in L2 speaking, especially through the lens of speaking self-efficacy and motivation. Besides, these results also underscore the potential advantages and future improvements of GenAI chatbots in language education through semi-structured interviews.
8:30-9am | UDL in Practice: Leveraging Generative AI to Overcome Resource Constraints in ESL Teaching (TSLL25-165)
Muskan Chhaparia (The English and Foreign Languages University, India🇮🇳)
While Universal Design for Learning (UDL) principles advocate for multiple means of representation, engagement, and action/expression, implementation in resource-constrained settings remains challenging. This research addresses this gap by examining how AI-generated visuals integrated within a UDL framework can transform reading instruction for 6th-grade ESL students in a low-resource English-medium school in Hyderabad, India. Using a sequential explanatory mixed-methods design, I compared traditional instruction with UDL-instruction supplemented by AI-generated materials. Quantitative data included pre-post assessments (n=21) measuring literal, inferential, and critical thinking dimensions of reading comprehension, along with delayed recall assessments (n=5) that allowed students to demonstrate comprehension using a combination of their first language and English. Qualitative data from teacher and student interviews further explored perceptions and experiences of the intervention. Results demonstrated statistically significant improvements in overall reading comprehension following UDL instruction (M=17.29, SD=2.99) compared to traditional instruction (M=11.86, SD=3.57), with a large effect size (Cohen’s d=1.35). Delayed recall assessments revealed substantial increases in information retention, with students using 35.8percent more total words and 25.4percent more unique English words following UDL instruction. Qualitative analysis identified AI-generated visuals as a key mechanism supporting comprehension and retention, particularly for struggling learners. This research supports CAST’s (2025) assertion that “integrating what we know about learning science (UDL) with the power of personalization (AI)” creates tools that “support all learners from the beginning rather than retrofitting solutions later.” These findings demonstrate how generative AI can democratise access to UDL principles in resource-constrained settings, providing a pathway to bring inclusive design into everyday classroom practice. It also has significant implications for advancing equity in ESL instruction across diverse educational contexts.
9-9:30am | GenAI in automatic Finnish speaking task generation, assessment, explanation and personalized feedback (TSLL25-73)
Nhan Phan (Aalto University, Finland🇫🇮), Anna von Zansen (University of Helsinki, Finland🇫🇮), Maria Kautonen (University of Jyväskylä, Finland🇫🇮), Raili Hildén (University of Helsinki, Finland🇫🇮), Ekaterina Voskoboinik (Aalto University, Finland🇫🇮), Tamás Grósz (Aalto University, Finland🇫🇮), & Mikko Kurimo (Aalto University, Finland🇫🇮)
GenAI technologies show considerable promise yet remain under-explored in applied linguistics – especially for low-resource languages such as Finnish. We present our innovative framework that leverages GenAI to address key challenges in automatic speaking assessment (ASA) for Finnish L2 learners, including task generation, grading explanation, and personalized feedback. We evaluate the framework on a picture-description task with Finnish L2 learners (Phan et al., 2024a). By using Chain-of-Thoughts and few-shot learning, our framework achieves substantial agreement between the automated system and human experts in assessing speech content. Moreover, the system provides transparent explanations for its grading decisions and generates tailored corrective feedback targeting learners’ specific errors. Building on these results, we showcase a self-study mobile app that implements the framework. The app features an automatic picture generation component with GenAI, offering learners virtually unlimited practice opportunities (Phan et al., 2024b). These findings demonstrate the potential impact of GenAI in revolutionizing language learning and assessment, especially where resources are limited. Our presentation aims to share the findings and practical implementation details of GenAI, contributing to the broader understanding and knowledge of GenAI in applied linguistics.
9-9:30am | Enhancing Academic Reading Through AI-Integrated CALL for MS English Students (TSLL25-120)
Shahzadi Kulsoom & Amna Mansoor (Air University, Pakistan🇵🇰)
This study explores the potential of Artificial Intelligence (AI)-powered Computer Assisted Language Learning (CALL) tools in improving academic reading skills among first-semester MS English students in Pakistan. With rapid advancements in educational technology, AI-integrated platforms offer innovative pathways for enhancing learners’ engagement and comprehension, especially in higher education contexts. Adopting a quasi-experimental design, this research involved 15 participants who were assessed through a pretest to establish their baseline academic reading proficiency. The intervention spanned 15 days, during which AI-based CALL tools were systematically incorporated into the reading component of their English classes. These tools provided personalized learning support, instant feedback, and interactive reading tasks to develop critical reading strategies and comprehension skills. Following the intervention, participants underwent a posttest to evaluate their progress. Quantitative data from the pretest and posttest were analyzed descriptively. At the same time, qualitative insights were gathered from learners’ reflections and feedback, offering a holistic understanding of their experiences with AI-facilitated learning. The results revealed notable improvements in students’ academic reading abilities. Participants demonstrated enhanced comprehension of complex texts, greater vocabulary acquisition, and improved critical engagement with reading materials. The qualitative data supported these findings, with students reporting increased motivation, reduced reading anxiety, and appreciation for the adaptability and interactivity of the AI tools. This study provides empirical evidence supporting the integration of AI-infused CALL approaches in tertiary-level English education. It underscores the capacity of AI technologies to transform traditional reading instruction into a more dynamic, learner-centered experience. The findings have practical implications for curriculum designers, language educators, and policymakers seeking to modernize English language instruction in Pakistani higher education through effective technological integration.
9:30-10:30am | New Conversations, New Evidence: Revisiting the Impact of Bots for Language Learning in the LLM Era
Serge Bibauw (UCLouvain, Belgium🇧🇪)
Generative AI and large language models (LLMs) have opened new opportunities to practice conversation meaningfully in a foreign language with a bot. I will revisit 40 years of research on dialogue-based computer-assisted language learning (CALL), before and after ChatGPT, to identify both how much it changed and what still holds up. I will explore the effectiveness of these “artificial” conversations to determine what works, for whom, for what and to what extent.
In particular, I will share preliminary findings from a new meta-analysis of chatbots for language learning, looking at both “technological eras”. In this systematic review, we observe that the field is moving beyond comparing the technology to comparing pedagogical and instructional conditions in their effects on language learning. I will call for more attention to task design and characteristics, and to scaffolding in conversation, particularly productive support for less proficient L2 speakers.
11-11:30am | AI-driven Automatic Speech Recognition Systems and mediated feedback: A mixed-methods study (TSLL25-157)
Daniel Murcia (The Pennsylvania State University🇺🇸)
This study explores the integration of mediated feedback from Computerized Dynamic Assessment (C-DA) principles into Generative AI-driven Automatic Speech Recognition (AI-ASR) systems to provide personalized, adaptive, and feedback for L2 pronunciation learners. The AI-ASR model employed in this study is an experimental prototype with an interface in English and Spanish. ASR systems predominantly deliver static, native-norm-based corrections, potentially disadvantaging learners with diverse regional accents (Escalante, Pack, and Barrett, 2023). The primary research question guiding this study is: How does the AI-ASR system provide and structure mediated feedback to support L2 learners’ pronunciation development, as indicated by both performance metrics and learner perceptions? Utilizing theories focused on learner-centered feedback and adaptive instructional design (Poehner, Zhang, and Lu, 2015), this mixed-methods research employs a sequential explanatory design to quantitatively evaluate the accuracy and reliability of AI-generated pronunciation feedback compared to expert human raters, followed by qualitative exploration of learner and instructor perceptions regarding the effectiveness, clarity, fairness, and adaptability of AI-ASR feedback to regional accents. Data from 20 undergraduate EFL learners at a Colombian university include AI-ASR performance metrics, human rating comparisons (Cohen’s Kappa), student focus groups, and instructor interviews. Preliminary findings indicate moderate AI-human alignment in phoneme accuracy but lower reliability in suprasegmental features. Learners valued immediate visual feedback but expressed the need for more adaptive, explanatory guidance, particularly for regionally influenced pronunciation features.Additionally, findings suggest that while AI-ASR serves effectively as a mediation tool, it should complement rather than replace teacher mediation and real-time interactional feedback in pronunciation development (Kurt and Kurt, 2024; Fathi,2024). The implications of this research inform stakeholders and AI-ASR system designers on enhancing personalized feedback mechanisms to improve the learning experience and pronunciation development for linguistically diverse learners (Zou et al., 2023).
11-11:30am | Virtual Reality Meets AI: Enhancing ESL Listening Through Technology (TSLL25-141)
Robin Couture-Matte (Université TÉLUQ, Canada🇨🇦), Djibril Dieng , & Erika Desjardins (Université Laval, Canada🇨🇦)
This study examines the combined potential of immersive virtual reality (IVR) and artificial intelligence (AI) in second language learning for young learners. IVR, particularly via headsets, has gained attention for its language learning affordances (Dhimolea et al., 2023), Simultaneously, advancements in AI offer new avenues for interaction and learner guidance within digital environments (Godwin-Jones, 2023). A basic yet effective form of AI, reactive-machine AI (RMAI) responds to predetermined cues and shapes learner experience within IVR. Although promising results have emerged regarding vocabulary and listening comprehension (Tai, 2022), few studies focus on younger populations (Dooly et al., 2023). To address this gap, this study explores how RMAI within an IVR setting supports listening skills among 11- to 12-year-olds. Conducted in Canada in 2024, the study involved 17 students in an ESL program who interacted with Elixir, a task-based VR experience delivered via the Meta Quest 3 headset. Using Tai’s (2022) framework, immediate and delayed recall activities were analyzed through idea unit analysis, complemented by questionnaires and interviews. Findings showed gains in vocabulary and listening comprehension, supported by high learner engagement. The interactive features of the IVR environment, enhanced by RMAI, supported learner understanding and sustained attention. Results highlight the potential of IVR and AI to enhance language learning and inform future pedagogical innovations.
11:30-Noon | Nuances of English Lenition: Decrypting the Perceptual Challenges Faced by ESL Speakers Using Gen AI- Chatbot (TSLL25-63)
Aneeqa Zafar (University of Education, Pakistan🇵🇰), Mahwish Farooq (University of Central Punjab, Pakistan), & Khalid Ahmed (INTI International University, Malaysia🇲🇾)
The study aims to problematize the challenges faced by ESL speakers in comprehending English lenited consonants during listening practices on the Gen Al-chatbot, ELSA SPEAK. It also explores the difference between the phonetic context of English and Urdu languages that cause perceptual challenges for L1 Urdu speakers. Fledge’s speech learning model renders a theoretical lens to this study as it implies that the phonetic perception of ESL learners has some bearing on the phonetics of L1 and L2. A sample of 50 L1 Urdu speakers aged 18 – 22 years who use English as a second language and have adequate exposure to phonetic transcription have been selected. To analyze the garnered data from the results of listening tasks performed on Gen AI-chatbot, ELSA SPEAK, and the open-ended interviews, thematic analysis was used in this piece of research on the perceptual difficulties of L1 Urdu language speakers while perceiving lenited consonants in English as a second language. The findings reveal several errors in the perception of English lenited consonants at word-initial, word-medial, and word-final positions, and the impact of the various phonetic environments of the English language on the ability of Urdu speakers to perceive the English lenited consonants. This study probes into perceptual difficulties faced by L1 speakers of the Urdu language in decoding lenited consonants in English as a second language to fill a gap in the available research and equally renders practical acumens to academics as well as language teaching experts. The result of this study contributes substantial insights into the avenues of English language teaching and learning and effectual strategies to augment listening and speaking skills including the pronunciation of L1 speakers of the Urdu language who intend to acquire the English language as a second language.
11:30-Noon | Exploring Positive Emotions and Emotion Regulation Strategies of Language Teacher Educators in AI-Enhanced Classrooms (TSLL25-164)
Ramazan Yetkin (Niğde Ömer Halisdemir University, Türkiye🇹🇷)
With the advent of generative AI platforms such as ChatGPT in 2022, a significant shift has been observed in educational domains, particularly in language classrooms. Both learners and educators have increasingly adopted these tools to enhance the language learning process—learners for personalized support and immediate feedback, and teachers to ease the burden of preparation and facilitate instruction. While existing research has highlighted the pedagogical affordances of AI tools in language learning, such as tailored learning experiences and feedback, there remains a noticeable gap in exploring the emotional dimensions of AI integration, especially among language teacher educators.This qualitative study investigates the positive emotions and emotion regulation strategies of two experienced language teacher educators who have extensively incorporated AI tools into their teaching practices. Data were collected through reflective journals and semi-structured interviews. The findings indicate that the participants predominantly experienced positive emotions related to AI use, such as increased motivation and reduced teaching anxiety. To manage potential negative emotions—particularly concerns about overreliance on technology and ethical implications—they employed a blend of emotion regulation strategies. These included soliciting learner feedback, engaging in personal reflection, and comparing AI-supported courses with traditional approaches to maintain a balanced perspective. This study aims to contribute to the evolving discourse on AI in applied linguistics by foregrounding the emotional experiences of educators—key agents in the adoption and implementation of AI tools. The findings have implications for future research and classroom practices, highlighting the importance of supporting educators emotionally as they navigate the integration of generative AI in language teaching contexts. Keywords: generative AI, language teacher educator, emotion regulation, applied linguistics, AI in education
1-1:30pm | Incorporating Generative AI into task-based English instruction for adult refugees in Canada: A mixed-methods study (TSLL25-98)
Ji-young Shin & Liz Coulson (University of Toronto Mississauga, Canada🇨🇦)
Generative artificial intelligence (AI) chatbots have been increasingly used in teaching ESL (Hwang et al., 2023; Jeon et al., 2023; Joen and Lee, 2022), but research has mostly focused on the perceptions of conventional L2 students (and teachers). Less is known about its integration into specific ESL instruction models, moderation of ESL proficiency, or non-conventional ESL learners. The current project conducted a mixed-methods, quasi-experimental study that examines the impact of using AI for teaching refugees of varied English proficiency levels within the task-based language teaching (TBLT) framework, as well as their perceptions. Data were collected from 38 students enrolled in a seven week-long ESL program in a non-profit organization for refugees in Canada. Participants’ proficiency ranged from high-beginner to low-advanced with Arabic as the largest first language group. The participants were randomly assigned to experimental and control groups. The experimental group learned English conversations on everyday topics using ChatGPT-integrated pre-, during-, and post-tasks, while traditional tasks were used for the control group with assistance from volunteering tutors in addition to the instructor. As pre- and post-tests, elicited imitation (EI) was utilized. To examine pre-post differences are statistically significant between the groups, controlled for proficiency, multilevel modeling, particularly random intercept model, was used, analyzing 1,241 EI responses, where forty item scores (level 1) were clustered within each individual test taker/participant (level 2). The best-fitting model indicated that pre-post score changes across the individuals and groups were significant, even when controlling for pre-existing L2 English proficiency. The group differences were insignificant, meaning that AI chatbot-based activities were effective as interactions with human tutors. Follow-up interviews with participants also demonstrated the positive impact of AI chatbot-based activities for their engagement, motivation, and attitude. The study findings support the positive potential of AI-enhanced TBLT, particularly for ESL learners in non-school settings such as refugees.
1-1:30pm | Examining AI-Mediated Linguistic Alignment in L2 Catalan: Task Effects on Learner-ChatGPT Interactions (TSLL25-158)
Mireia Toda-Cosi (University of Maryland, College Park🇺🇸)
This research examines how intermediate L2 Catalan learners adapt their language during interactions with AI language models in synchronous computer-mediated communication (SCMC). Despite growing interest in AI applications for language learning, minimal research exists on the linguistic alignment phenomena that occur when learners engage with AI conversational partners. Our investigation builds upon initial findings suggesting potential alignment between the linguistic production of Catalan learners and ChatGPT in complexity and fluency dimensions (Toda Cosi and Poole, in press). The current paper will analyze lexical and syntactic alignment patterns across varied task-based interactions, as well as learners’ integration of AI-provided recasts following non-target-like forms. The study is framed within dual perspectives on linguistic convergence: as automatic priming processes (Pickering and Ferreira, 2008) and as strategic socio-cognitive adaptations (Costa et al., 2008). We anticipate that the SCMC environment will enhance linguistic alignment due to reduced processing demands and heightened form awareness (Kim et al., 2020), with task characteristics influencing alignment patterns. Preliminary interview data suggests learners notice and incorporate these recasts, potentially enhancing form-meaning connections. This work advances our understanding of technology-mediated language learning processes in less commonly taught language (LCTL) contexts, where authentic interaction opportunities remain scarce (Heidrich Uebel et al., 2023). Findings will contribute to the emerging field of AI-assisted language pedagogy by providing empirical evidence of how interactive alignment influences L2 development and informing the design of effective AI-supported learning tasks.
1:30-2pm | Task Engagement with Different Types of Interlocutors: Humans vs. AI (TSLL25-104)
Hadeel Arqbi (Georgetown University🇺🇸)
Technological advances have made AI assistants integral to everyday life, offering language learners conversational practice opportunities outside the classroom. Research has shown that engaging in language tasks positively impacts language development (Hiver and Wu, 2023). Though task engagement has gained interest in task-based language teaching (TBLT) literature (Philp and Duchesne, 2016), studies on its effects in interactions with AI agents are still limited ( Bear et al., 2024; Kim et al., 2022). This study explores task engagement among Arabic language learners when interacting with human interlocutors versus AI agents. It addresses a gap in TBLT literature regarding AI’s impact on learner engagement and efficiency as language partners. The study utilizes the Chat-GPT4o voice feature to investigate the learners’ interaction and engagement with AI agents. Employing a mixed-method approach with qualitative (stimulated recall and semi-structured interviews) and quantitative (Likert scale questionnaires and observational data) methods, the study surveys L2 Arabic learners aged 20-30. The study recorded instances of learner engagement with AI agents across cognitive, emotional, and behavioral aspects. Behavioral engagement is assessed through self-reports, verbal report, and observed behaviors during task interaction, such as time on task and questions asked. Emotional engagement is measured using self-reports, verbal report, and non-verbal behaviors such as laughter. Cognitive engagement is evaluated similarly through self-reports, verbal report, and language-related episodes analysis. Preliminary results suggest that participants experienced greater engagement with human interlocutors than with AI agents, with some feeling emotionally disconnected from the AI. However, all participants showcased cognitive engagement with different degrees when interacting with the AI, sometimes treating it similarly to human interlocutors. The study seeks to improve teaching methods and guide future research on effective tasks in technology-mediated learning, contributing to the knowledge of AI in language education and human-AI interaction.
1:30-2pm | Using ChatGPT in a Spanish Language Course for Heritage Speakers (TSLL25-52)
Sebastian Leal-Arenas (University of Pittsburgh🇺🇸)
The public release of Artificial Intelligence (AI) tools has sparked debate and discomfort in academia, particularly regarding ethical use and questions of responsibility (Mhlanga, 2023; Susnjak, 2022). However, students’ perspectives have received far less attention. This study has a twofold focus: it presents how ChatGPT was integrated into the writing process of a final project in a heritage writing course at a U.S. university, and it explores students’ perceptions of its use. Throughout the semester, students used ChatGPT to support specific stages of essay development, brainstorming, thesis creation, generating argumentative ideas, and crafting titles. For each stage, students submitted reflective journals describing their interactions with the tool. At the end of the semester, they completed a survey evaluating their overall experience. Results show that students appreciated the platform’s interactive nature (80percent) and the breadth of information provided (65percent). Reported drawbacks included repetitive content (70percent) and concerns about the reliability of the information (85percent). Importantly, all students (100percent) expressed concern about AI-detection tools potentially misclassifying their work, even in cases where no AI assistance was used. These findings are discussed in the context of recent research on writing classification by both AI and second-language speakers (Corizzo and Leal-Arenas, 2023; Liang et al., 2023).
2-2:30pm | GenAI in Task-Based Language Teaching: Learning Spoken Collocations Across Contexts (TSLL25-161)
Valentina Morgana & Francesca Poli (Università Cattolica del Sacro Cuore-Milano, Italy🇮🇹)
This study investigates the role of generative AI (GenAI) in enhancing spoken English collocational competence through task-based language teaching (TBLT). Using a mixed-methods design, the study compares two sets of AI-mediated tasks designed to promote fluent and context-appropriate use of spoken collocations in everyday conversation and academic presentation settings. Participants (N = 40) were assigned to two groups and completed five guided or unguided tasks with AI support, while a control group completed the same tasks without AI assistance. Ten high-frequency collocations for the movie-based tasks were selected from the American Movie Corpus (AMC; Forchini and Palmero Aprosio, 2025), and ten academic collocations were drawn from the Academic Spoken Collocation List (ASCL; Li et al., 2024). Collocations were embedded into AI-generated dialogues and task prompts to provide authentic input. Quantitative analysis examined group-level and task-condition effects on collocation use. Results indicated significant gains in the guided AI group, followed by the unguided group, with minimal improvement in the control group. The mixed-effects model confirmed a significant interaction between task type and AI support condition, highlighting the pedagogical value of structured AI interaction. Qualitative analysis of learners-AI interactions revealed that metalinguistic reflection and lexical reformulation were more frequent in the guided group than in the unguided tasks (Lee, 2023). Findings demonstrate that GenAI can play a key role in helping learners notice, process, and use high-frequency spoken collocations effectively, particularly when tasks are scaffolded and register-sensitive to promote (meta)linguistic reflection. The study highlights the benefits of combining traditional statistical and mixed-effects modeling in applied linguistics research.
2-2:30pm | Leveraging ChatGPT for Unit Design: Integrating Multilingual Standards into Project-Based Learning (TSLL25-170)
Jeanne Beck , Hwee Jean (Cindy) Lim , & Gulbahar Beckett (Iowa State University🇺🇸)
As generative AI (GenAI) becomes increasingly integrated into educational workflows, language educators are exploring how tools like ChatGPT can support complex instructional design tasks. This presentation examines how ChatGPT was used to co-develop a standards-aligned Project-Based Learning (PBL) unit for multilingual elementary students engaged in language revitalization efforts. Specifically, the study focuses on how prompt engineering enabled the integration of three distinct frameworks: the Missouri Learning Standards for English Language Arts, the WIDA English Language Development Standards, and the Gold Standard PBL: Essential Project Design Elements. Through multi-step prompt refinement, ChatGPT was guided to generate daily learning targets, scaffolded activities, and performance assessments that aligned with both language and content goals. The iterative process highlights the tool’s ability to synthesize standards from multiple sources into coherent, day-by-day instructional sequences, offering significant time-saving potential for educators who desire to create PBL units while navigating multiple state and consortium standards. The use of GenAI, however, also revealed notable limitations. While ChatGPT performed well in planning and standards alignment, both tools struggled to provide helpful information regarding Chuukese and Pohnpeian language and culture. Oversight and additional prompting was also needed to ensure the standards language remained consistent. This gap underscores the need for human oversight, particularly in minority language contexts. This session contributes to the growing conversation around human-GenAI collaboration in applied linguistics by demonstrating a pragmatic use case: standards integration for multilingual PBL unit planning. Attendees will gain insight into effective prompt engineering strategies, view annotated examples of GenAI-assisted PBL unit materials, and discuss implications for using GenAI to support multilingual, standards-aligned instruction. The presentation ultimately argues for a balanced approach, recognizing GenAI as a powerful planning partner that still requires human expertise to ensure cultural responsiveness and accuracy.