PaperlayerView exact revision on arXiv

Source-linked AI summary

Distributed agency in second language learning and teaching through generative AI

Robert Godwin-Jones

arXiv:2403.20216v4cs.CYcs.AI

TL;DR

This paper considers where agency resides in AI–learner–teacher relationships and how AI’s expanding role affects language learning and teaching. It examines critical AI literacy and ecological perspectives, concluding that AI should be enlisted, despite ethical concerns, in support of plurilingual competence, social justice, and cross-cultural understanding.

  • Problem

    As generative AI becomes more integrated into education, questions arise about who controls AI-mediated interactions and how its use affects learners, teachers, and agency.

    Source: Introduction · pp. 1–2

  • Method

    The column considers AI in language learning through questions of human–AI agency, critical AI literacy, and the ethical and practical conditions of use.

    Source: Introduction · p. 2

  • Results

    The column argues that AI’s power should be enlisted toward plurilingual competence, social justice, and cross-cultural understanding despite concerns about its creation and use.

    Source: Conclusion: Emergent, Shared Agency in… · p. 18

  • Takeaways & Limitations

    Agentive literacy is needed to empower learners to enlist AI support toward more open, locally grounded education.

    Source: Conclusion: Emergent, Shared Agency in… · p. 18

    The prospect of ambient AI for informal language learning is qualified by unequal access, as the envisioned scenario is limited to socially and economically privileged groups.

    Source: Learner Agency Through AI is Tempered… · p. 14

Abstract

from arXiv · show

Generative AI offers significant opportunities for language learning. Tools like ChatGPT can provide informal second language practice through chats in written or voice forms, with the learner specifying through prompts conversational parameters such as proficiency level, language register, and discussion topics. AI can be instructed to give corrective feedback, create practice exercises, or develop an extended study plan. Instructors can use AI to build learning and assessment materials in a variety of media. AI is likely to make immersive technologies more powerful and versatile, moving away from scripted interactions. For both learners and teachers, it is important to understand the limitations of AI systems that arise from their purely statistical model of human language, which limits their ability to deal with nuanced social and cultural aspects of language use. Additionally, there are ethical concerns over how AI systems are created as well as practical constraints in their use, especially for less privileged populations. The power and versatility of AI tools are likely to turn them into valuable and constant companions in many peoples lives (akin to smartphones), creating a close connection that goes beyond simple tool use. Ecological theories such as sociomaterialism are helpful in examining the shared agency that develops through close user-AI interactions, as are the perspectives on human-object relations from Indigenous cultures.

Introduction

Generative AI is rapidly entering education and offers new capabilities for L2 learners and teachers, while raising questions about its effects and appropriate use. The column argues for critical AI literacy and ecological perspectives to understand AI’s power, ethical and practical constraints, and shared agency with users.

Introduction

  • Generative AI already supports L2 writing, translation, grammar correction, lesson planning, learning activities, and assessment.Students and teachers are exploring these capabilities as institutions and professional organizations issue guidance and workshops on AI integration.Source: Introduction · p. 2
  • Critical AI literacy should help learners and teachers understand generative AI’s nature, power dynamics, ethical concerns, and practical barriers to use.The column presents this as an appropriate initial response to AI in instructed second language acquisition settings.Source: Introduction · p. 2
  • AI’s expanding role in L2 learning and teaching may shift second language acquisition and teachers’ roles in the learning process.The column anticipates wider change as AI tools become ubiquitous, at least in privileged communities.Source: Introduction · p. 2
  • Ecological frameworks such as sociomaterialism can illuminate agency shared between human users and AI, while Indigenous and non-Western ontologies inform inclusive perspectives on linguistic plurality and control.The column also frames its inquiry around who controls interactions in the AI–learner–teacher relationship.Source: Introduction · p. 2

AI Asserts Control: “I’m sorry, Dave, I’m afraid I can’t do that.”

The section examines who controls AI-mediated interactions and how agency can emerge across users, tools, and their environments. It presents AI as a potentially powerful language-learning companion while emphasizing that user control, reciprocal agency, and ethical constraints vary by context.

AI Asserts Control: “I’m sorry, Dave, I’m afraid I can’t do that.”

  • AI capabilities continue to expand, but the section distinguishes such development from achieving general AI or the predicted singularity.It points to unexpected functions users have uncovered and improvements across ChatGPT versions as reasons to expect further capabilities.Source: AI Asserts Control: “I’m sorry, Dave,… · p. 3
  • AI’s growing role as an adaptive companion in education raises questions about trust, influence, and how teachers’ roles may need to be reconsidered.The section connects those concerns to adaptive AI tutors and other AI-powered services.Source: AI Asserts Control: “I’m sorry, Dave,… · p. 3
  • AI chatbots offer informal and structured L2 learning through written or voice interaction, with prompts allowing users to set conversational parameters.The section describes them as potentially always-available, non-judgmental partners and argues that AI should be considered within a broader learner–tool–environment ecosystem.Source: AI Asserts Control: “I’m sorry, Dave,… · pp. 3–4
  • Affordances emerge through situated relationships between learners and tools, and users’ prompts both repurpose AI and become data that can inform machine learning.The section describes this as a two-way process between users and technology.Source: AI Asserts Control: “I’m sorry, Dave,… · p. 4
  • AI chatbots offer users more control over content and conversational parameters than algorithmically driven social media, but that control remains limited.Users can direct conversational flow and modify output through subsequent prompts; the section also notes that AI systems have built-in biases and control mechanisms.Source: AI Asserts Control: “I’m sorry, Dave,… · p. 4
  • Ecological accounts treat agency as residing in users and AI, with reciprocal but not necessarily symmetrical action constrained by rights, responsibilities, and ethics.This account rejects a view of humans as standalone agents with unique control over the material world.Source: AI Asserts Control: “I’m sorry, Dave,… · p. 4
  • Learner–AI exchanges are described as circular, with distributed agency emerging through iterative interaction rather than simply replacing learner agency.The section says it is too early to know whether AI use will follow patterns found in studies of machine translation; for now, humans remain “in the loop.”Source: AI Asserts Control: “I’m sorry, Dave,… · p. 4

Through Critical AI Literacy, Learners and Teachers Gain Agency

Critical AI literacy requires understanding both how generative systems produce language and where their capabilities fall short. The section connects that understanding to ethical, effective use in teaching and learning, including attention to access and power.

Through Critical AI Literacy, Learners and Teachers Gain Agency

  • Generative AI’s language ability comes from statistical patterns in large volumes of text, which enable language models to predict and generate sequences of words.The section describes tokens being represented as vectors and combined into parameters, with transformer attention helping select contextually likely words.Source: Through Critical AI Literacy, Learners… · p. 5
  • Fluent, authoritative-sounding output does not mean that AI understands meaning or verifies truth, and its lack of understanding can produce inaccurate or inappropriate text.The section notes that early users reported hallucinations, including misplaced or offensive comments and judgments.Source: Through Critical AI Literacy, Learners… · pp. 5–6
  • AI can mimic social-language patterns such as politeness, but it has less success adapting output to context and interlocutor characteristics.Speakers may know cultural norms yet choose to modify or reject them for personal, interpersonal, or political reasons.Source: Through Critical AI Literacy, Learners… · p. 6
  • Although AI systems handle multiple languages, they lack bilinguals’ lived experience of code-switching shaped by setting, identity, intent, and sociocultural factors.The section presents translanguaging as the continual co-presence of all the languages a person speaks.Source: Through Critical AI Literacy, Learners… · p. 6
  • AI’s limited access to embodied and sociocultural cues constrains human-like communication, especially where interpreting subjective states or sharing experiences is involved.The section reports that ChatGPT could use pragmatic language appropriately but had difficulty selecting language in contexts involving speaker subjectivity; performance improved in subsequent versions in the reported study.Source: Through Critical AI Literacy, Learners… · pp. 6–7
  • Learners and teachers can use AI for text creation, tailored feedback, and classroom activities, while instruction should model effective, ethical use and critical evaluation of output.The section recommends peer collaboration and teacher modeling so learners can integrate AI output into personal narratives that retain individual voice and creativity.Source: Through Critical AI Literacy, Learners… · p. 7
  • AI access is shaped by companies’ decisions about cost, geography, delivery, code availability, and user conditions, while advanced functionality is likely to depend on commercial cloud services.The section says access mode affects functionality and the degree and nature of user control, making differentiated user experiences and learning outcomes worth studying.Source: Through Critical AI Literacy, Learners… · p. 7

Varied Reach and Power of AI: Narrow and General Use Implementations

General-purpose and specialized AI tools offer varied options for language learning and teaching, from tailored writing support to customized apps and ambient devices. Their range may improve the match between tool functionality and user needs, while also introducing uncertainty for learners and teachers.

Varied Reach and Power of AI: Narrow and General Use Implementations

  • General-purpose AI systems can perform functions previously associated with narrow tools and are adding image, audio, and video generation capabilities.The section contrasts broadly capable systems with tools designed for specific language or media tasks.Source: Varied Reach and Power of AI: Narrow… · p. 8
  • General AI can tailor writing analysis and corrective feedback to prompts or rubrics, including priorities for L2 texts and particular programs.The section presents customization and avoiding multiple separate platforms as advantages over specialized writing tools.Source: Varied Reach and Power of AI: Narrow… · p. 8
  • LLMs can support data-driven learning by generating keyword-in-context displays, hybrid corpora, and examples of multiword units in context.The section describes these options as supporting vocabulary learning beyond one-to-one L1-to-L2 correspondences.Source: Varied Reach and Power of AI: Narrow… · p. 8
  • No-code services let educators create customized AI apps by saving instructions and adding external documents to shape functionality.The section gives an example of a pedagogical agent for English learners focused on ordering and transition phrases in narratives.Source: Varied Reach and Power of AI: Narrow… · pp. 8–9
  • AI-enabled apps and devices may become everyday companions, offering independent learners opportunities to integrate language use into daily experiences.The section says varied tools may create uncertainty but can also let users better match functionality to need.Source: Varied Reach and Power of AI: Narrow… · p. 9

L2 Writers’ Shared Agency

L2 writers may work with generative AI as a thinking partner, using it iteratively while retaining their own voice. The author argues that this shared work should be balanced with human interaction, critical reflection, and attention to linguistic and cultural diversity.

L2 Writers’ Shared Agency

  • AI-assisted writing assignments can include tool-use tasks and tasks that avoid AI, paired with reflection; AI detectors are unreliable for evaluating non-native writing (Liang et al., 2023).Suggested activities include comparing machine-translation versions with drafts and evaluating register or formality.Source: L2 Writers’ Shared Agency · p. 9
  • Writers use ChatGPT at different stages, including brainstorming, drafting, and revision, while some maintain their own voice and reject suggestions that do not fit their style (Warschauer et al., 2023; Baek et al., 2023; Jacob et al., 2023).Kailing refined prompts iteratively to address specific needs.Source: L2 Writers’ Shared Agency · pp. 9–10
  • Prompt instruction and hands-on experimentation can help learners revise prompts against previous output, shifting writing away from a transactional view and toward a process orientation.The passage calls this process “chain-of-thought” prompting (Wei et al., 2022, p. 1).Source: L2 Writers’ Shared Agency · p. 10
  • Kailing described ChatGPT as a thinking partner that helped her build on her ideas, while other studies report support for persuasive reasoning and writing self-efficacy [Jacob et al., 2023; Guo et al. (2022); Liu et al. (2021)].Liu et al. (2021) also reported improved self-regulated learning.Source: L2 Writers’ Shared Agency · p. 10
  • AI can assist writing mechanics, while human-to-human tasks such as peer editing and virtual exchange address writing’s social communicative function and pragmatic competencies.Real-world exchanges are more likely than chatbot conversations to develop adjustment to social contexts and strategies for communication breakdowns.Source: L2 Writers’ Shared Agency · pp. 10–11
  • AI voice recognition may struggle with dialects, accented speech, and regional variation, creating opportunities to discuss AI limitations and foster metalinguistic awareness and positive attitudes toward plurilingualism (Ramesh et al., 2023).Bilingual digital storytelling has been used to encourage collaboration and creativity across languages and cultures.Source: L2 Writers’ Shared Agency · p. 11
  • AI may support culturally grounded storytelling through brainstorming, revision, and multimodal images or video, but professional writers and translators use digital tools selectively as drafting aids.Storytelling can connect learners with cultural traditions, family relationships, communities, and heritage languages.Source: L2 Writers’ Shared Agency · p. 11
  • Sharing AI-assisted student writing online can connect classroom work to real-world uses and personal or professional applicability (Blyth, 2023; Dooly and Darwin, 2022).Examples include public journals, wikis, and renewable assignments.Source: L2 Writers’ Shared Agency · p. 11

Framing Joint Human-Tool Agency

The paper frames language learning as an ecological relationship among learners, tools, teachers, and social contexts, with agency distributed across humans and nonhumans. It uses this perspective to consider AI’s changing roles, learner-specific pathways, and the cultural and practical conditions shaping human-tool relations.

Framing Joint Human-Tool Agency

  • The paper places these relationships within formal education, social learning, and AI assistance, rather than treating AI use as separate from learners’ social contexts.It also connects language learning and social justice to marginalized communities and equitable multilingualism.Source: Framing Joint Human-Tool Agency · pp. 11, 13
  • An ecological perspective accounts for changing learner-machine-world relationships, while relational pedagogy emphasizes how materials and technologies interact with social worlds and individual creativity.The approach also considers learners’ starting points, learning trajectories, and external factors.Source: Framing Joint Human-Tool Agency · p. 12
  • Sociomaterialism frames language learning as an entanglement in which agency is shared among human and nonhuman contributors, rather than humans acting autonomously on material resources.The paper situates this within an expanded ecological framework incorporating material culture, embedded communication, and distributed roles.Source: Framing Joint Human-Tool Agency · p. 12
  • The teacher–AI relationship can be a partnership: the English teacher designs learning interactions, tests revisions, and mentors students, while an AI tutor controls learning flow, personalized content delivery, instant feedback, and performance tracking.Source: Framing Joint Human-Tool Agency · p. 12
  • AI’s power and ubiquity can increase uncertainty in learning outcomes, while systems that recall prior encounters may support incremental, context-aware learning and adaptive tutors.Such personalization aligns with SLA approaches emphasizing individualized goals.Source: Framing Joint Human-Tool Agency · p. 12
  • Learner-AI pathways vary with learner profiles, prior resources, and changing human and nonhuman influences, so successful cases suggest possibilities rather than a general blueprint.The paper points to case studies and cluster analysis as ways to examine this variation (Woo et al., 2024; Wang et al., 2023).Source: Framing Joint Human-Tool Agency · pp. 12–13
  • Indigenous perspectives offer a nondual view of humans and nonhumans and an acceptance of ambiguity that the paper connects to working with AI’s opaque, unpredictable capabilities.An Ojibwe language lesson treated natural entities as agentive in Ojibwe conversation.Source: Framing Joint Human-Tool Agency · pp. 13–14
  • As with smartphones, AI may become an everyday companion and a site of shared agency, with user-tool relations ranging from impersonal to intimate and shaped by individual L2 trajectories.The paper notes users’ strong adverse reaction when intimacy filters were instituted in the Replika chatbot.Source: Framing Joint Human-Tool Agency · p. 14

Learner Agency Through AI is Tempered by Ethical Issues and Practical Concerns

AI may offer learning opportunities, but access, proficiency, privacy, ethics, and cultural bias constrain its use. The paper argues for critical AI literacy and qualified trust rather than using AI as a primary source of linguistic or cultural knowledge.

Learner Agency Through AI is Tempered by Ethical Issues and Practical Concerns

  • Ambient AI access is limited by privilege, infrastructure, available time and space, and the cost of high-performing systems; effective use also requires some proficiency.In some regions, digital tools may only be available in school settings, while AI companies may have little incentive to support minority languages (Ramesh et al., 2023).Source: Learner Agency Through AI is Tempered… · pp. 14–15
  • AI’s use raises concerns about privacy and surveillance, including collection of learner activity and data, and commercial interests in AI companions.The passage cites concerns about biometric monitoring and an early AI school experiment involving electronic headbands (Wang et al., 2019).Source: Learner Agency Through AI is Tempered… · p. 15
  • LLM creation raises ethical concerns about energy use, low-paid data work, creators’ rights, and synthetic likenesses and voices that can contribute to disinformation.The paper says these problems are likely to remain and grow as voice synthesis and video manipulation advance.Source: Learner Agency Through AI is Tempered… · p. 15
  • AI outputs may reflect white Western cultural biases and reproduce social and racial hierarchies, while Indigenous languages are largely absent from current AI data (Atari et al., 2023; Ramesh et al., 2023; McKnight & Shipp, 2024).These concerns help explain some First Nation communities’ reluctance to use AI or include Indigenous languages in AI data.Source: Learner Agency Through AI is Tempered… · p. 15
  • Cultural limitations and ethical concerns argue against using AI as a primary vehicle for linguistic or cultural competency; critical AI literacy should support critical thinking and qualified trust in outputs (Enriquez et al., 2024).The cited example of AI imitating an African American author raises concerns about cultural appropriation, identity theft, and copyright infringement.Source: Learner Agency Through AI is Tempered… · pp. 15–16

Conclusion: Emergent, Shared Agency in Merged AI and Virtual Reality

The conclusion presents AI-enhanced VR as a possible language-learning partner, while noting that communication is embodied and that access, ethical, and practical constraints remain. It calls for AI integration that supports plurilingualism, social justice, and shared human–AI agency.

Conclusion: Emergent, Shared Agency in Merged AI and Virtual Reality

  • Because SLA views learning as embodied and socially situated, AI’s communicative ability remains limited, even as multimodal VR could mimic more human semiotic resources.Realistic avatars could incorporate gestures, movement, facial expressions, and paralanguage.Source: Conclusion: Emergent, Shared Agency in… · p. 16
  • AI could replace scripted VR role-play with unscripted exchanges, but current apps often constrain responses and underuse movement, facial recognition, and gesture.AI bots could act as companions, tutors, or game characters; existing apps commonly limit responses to multiple choice.Source: Conclusion: Emergent, Shared Agency in… · p. 16
  • Access, privacy, cost, and compatibility constrain educational adoption, especially in under-resourced communities, although educators have demonstrated lower-cost and collaborative workarounds.Examples include inexpensive Cardboard viewers and a student-designed chatbot built with access to a single school computer.Source: Conclusion: Emergent, Shared Agency in… · p. 17
  • Prompt-customized VR can create immersive L2 environments and storylines, with learning emerging through shared, asymmetrical agency between human and nonhuman actors.Learners can specify languages, feedback, reinforcement, and areas of study or interest.Source: Conclusion: Emergent, Shared Agency in… · pp. 17–18
  • Prompt-based multimedia creation gives learners or teachers control over simulations without dedicated coders, while raising questions about coding careers and higher education.Prompts can create branched gaming environments, video, and 3D images.Source: Conclusion: Emergent, Shared Agency in… · p. 18
  • The conclusion pairs concerns about opaque AI systems with efforts toward moral AI, while noting that machine morality remains an open question.The complexity of LLMs makes understanding how they interpret input and generate output difficult; the Delphi project aims for more inclusive and socially aware behavior.Source: Conclusion: Emergent, Shared Agency in… · p. 18
  • The author prioritizes plurilingual competence, social justice, and cross-cultural understanding, and calls for agentive literacy to support effective, ethical AI integration.AI is presented as a possible catalyst for more equitable and effective education.Source: Conclusion: Emergent, Shared Agency in… · pp. 18–19
Loading 2403.20216v4…