AIEd Knowledge BaseAIEd KB
On this page

Concept

Language Learning

created 2026-08-09 · updated 2026-11-06

Language Learning — the study of how AI supports second language (L2) acquisition, writing development, and linguistic diversity in educational settings. AI in education research in this knowledge base spans AI interlocutors for spoken dialogue, automated writing evaluation for L2 learners, reading support, and concerns about language bias in AI scoring systems.

Questions to Consider

  • Language is inherently interactive, which makes it well-suited to conversational AI — but AI's linguistic capabilities also raise risks of bias against non-native patterns. Where have you seen this tension between opportunity and risk play out?
  • One study found AI scoring systematically underestimates linguistically weaker students, while another proposed comparing students to their own prior work rather than native-speaker norms. How does the 'reference point' for evaluation change whether AI feedback helps or penalizes a learner?
  • AI interlocutors can extend communicative practice at scale, but the page warns they should pair with human interaction so fluency transfers to real conversation. What might you gain from practicing with an AI that a human partner can't give — and what would you lose?
  • Teacher support — not just the AI tool — was shown to drive engagement in AI-assisted language learning through students' achievement goals. How does the social and pedagogical context shape whether learners keep engaging with an AI practice tool?
  • A meta-analysis found small-to-moderate, level-dependent gains from emerging tech, with productive skills (speaking, writing) gaining more than receptive ones. Why might speaking and writing benefit more than listening and reading from AI tools?
  • If AI privileges standard English and can penalize non-native or diverse language patterns, how should language instructors design evaluation and feedback so AI supports linguistic diversity rather than erasing it?

Introduction

Language learning has emerged as a significant AI in education domain because language is inherently interactive — making it well-suited to conversational AI — and because AI's linguistic capabilities raise both opportunities (personalized language practice at scale) and risks (systematic bias against non-native language patterns). The articles in this knowledge base explore both sides of this equation. Where the target language is English specifically — especially English for Academic Purposes (EAP) and EFL/ESL/L2 English teaching — see the dedicated English Education (EAP / EFL / ESL) concept page, which distinguishes English-specific research from general L2 acquisition and general writing.

AI as language tutor and interlocutor is the most developed theme. What Changes When the Interlocutor Is an AI? examines interactional fluency and linguistic uptake when L2 learners converse with AI versus humans. TACT provides pedagogically adaptive ESL tutoring. Children's English reading story generation via supervised fine-tuning of compact LLMs with controllable difficulty and safety explores AI-generated stories for children's reading development. These connect to Intelligent Tutoring and Generative AI. Yang, Weng and Yang (2026) designed two Large Language Models (LLMs)-based agents — a conventional AI English teacher and one using the 5E framework (engage, explore, explain, elaborate, evaluate) for inquiry-based grammar learning. Across 37 ESL students in a randomized comparison, high-performing students responded positively to the AI teacher while low-performing students showed mixed attitudes, and the conditions differed in intrinsic motivation, cognitive change, and performance — indicating that LLM-agent design should be matched to learner proficiency.

For physically embodied tutors, a meta-analysis of 11 RALL studies (N = 595) found a large pooled effect on L2 learning (g = 0.83) with high heterogeneity, and only interaction format moderated it — group-based formats beat one-on-one while robot morphology, modality, autonomy, and social role did not (Wang, Zhang & Zou (2026)).

Knowledge component granularity is a modeling decision in school-based tutoring: El Hefny and Meurers (2026) compared 19 knowledge component (KC) models spanning 11 levels of the FeedBook domain hierarchy on 153,819 first-attempt responses from 1,059 seventh graders learning English in German schools. Fine-grained item-level models predicted best for new students on known content, while semantically grounded mid-level models generalized best to new content (Property ID + Exercise Type RMSE 0.4243 under AFM versus 0.4447 for Actionable Element ID). An annotation-coverage audit found strict property mapping initially removed 45.7% of records and a newer annotation tool reduced the exclusion to 30.6%, with the combined label positive|statement alone accounting for 74.9% of excluded records.

AI in language assessment is emerging as LLMs support item generation and evaluation. Aryadoust and Wong (2026) compared prompt engineering against fine-tuning for automatic item generation in L2 listening assessment: iterative prompt refinement improved item quality but plateaued, while fine-tuning GPT-4.1 on the optimized prompt (holding prompt design constant) yielded further gains — a template for when assessment developers should invest in model adaptation over prompt iteration.

A comparison of 52 EFL assessment tasks rated by 20 experienced teachers found no significant quality difference overall between AI-generated and human-developed items, but a clear division of labor: AI was preferred for grammar and vocabulary (69%) and human development for reading, writing, listening and speaking (75–83%) (Nourashrafi, Alavinia and Darvishi (2026)).

Automated writing evaluation for L2 learners evaluates AI's ability to assess non-native writing. Bannò et al. proposed a self-referential approach comparing student writing to their own prior work rather than native-speaker norms. Feser & Tschisgale found AI scoring systematically underestimates linguistically weak students — a finding that connects to Assessment Validity and Bias Mitigation concerns. Generative AI and linguistic diversity in academic writing and publishing: Perspectives from World Englishes explores how AI affects linguistic diversity in academic contexts.

A discourse-level system goes further by naming the defect rather than scoring the text: sentence-pair relation classification localized coherence breaks - logical jumps, missing connectives, ambiguous reference - and teacher-judged adoption of its generated feedback ran 71.2%-88.4% (Wang et al., 2026).

Accessibility for language learners connects to Inclusive Learning: DysLexLens analyzed how dyslexic learners use AI for literacy support, and AI tools in Arab University English classrooms: Looking back and forward explored AI tools in Arabic-English classroom contexts. These studies connect language learning to Equity and Special Education.

Motivational mechanisms in AI-assisted language learning examine why learners engage with AI for language practice. Wang & Wang (2026) used goal-setting theory with 758 Chinese university English learners to show that teacher support enhances engagement in AI-assisted learning through students' mastery-approach and performance-approach goals (not avoidance goals) — evidence that the pedagogical and social context, not just the AI tool, determines whether learners stay engaged with AI-assisted language practice. This connects language learning to Motivation and Student Engagement.

Annamalai et al. (2026) add a qualitative self-determination case: in interviews with 25 Malaysian undergraduates, ChatGPT supported competence and autonomy, and its conversational responsiveness produced a sense of being heard — an "AI-mediated motivational ecology" in which relatedness is partly met by the tool, though inaccurate references demanded verification and human complement.

Design quality, not usage frequency, carried the motivational effect in an instructor-built tutor: across 74 undergraduates using a course-scoped Japanese GPT, out-of-class usage frequency showed no significant correlation with autonomy, competence, or relatedness, while learners rated the tutor highly for self-directed learning (M = 4.39) and cited affective safety (60.8%) (Lee & Kwon, 2026).

GenAI-supported writing at the primary level. Lu et al. (2026) ran a nine-week opinion-writing program with 301 Grade 5 and 6 learners in Eastern China, with eight intact classes randomly assigned to the program or to conventional instruction. The program raised learners' ideal L2 writing self (adjusted mean difference 0.20) and academic buoyancy (0.17), and lifted behavioral and emotional engagement, but it did not move growth mindset, cognitive or metacognitive engagement, or rubric-scored organization — among writing dimensions only language use improved. Two features of the design matter for language teachers: prompting was taught explicitly, through a categorized bank of prompts tied to specific writing goals, and GenAI feedback was used alongside comparison with teacher feedback and repeated revision. The authorship gains learners reported rested on that instructional structure rather than on the tool by itself, and the authors name reduced self-monitoring and shortcut-oriented strategies as the standing risks.

GenAI grammar instruction works through revision, not commentary. In a ten-week quasi-experiment with 88 Ethiopian EFL students, draft–prompt–reflect–revise cycles raised grammar (η² = 0.499) and overall writing, and content scores rose although the prompt barred content feedback (Wordofa & Abo, 2026).

Implications for language instructors

  • Emerging technologies yield small-to-moderate, level-dependent gains. A meta-analysis of 33 TEFL studies (N = 3,181) finds an overall effect of Hedges' g = 0.38 that rises with educational level (primary 0.29, secondary 0.35, tertiary 0.44), with VR/AR yielding the largest effects and productive skills (speaking, writing) gaining more than receptive skills — supporting the use of emerging tech, especially at tertiary level, while keeping expectations realistic.

  • A systematic review maps where elementary AI language education is thin. Across 31 studies (2013-2025), work in elementary language classrooms clustered on speaking, literacy and vocabulary while grammar, listening comprehension and sign language were barely studied, and most designs lacked grade-level specificity (Hamasha et al., 2026).

  • Use AI to extend communicative practice, not replace it. AI interlocutors and adaptive ESL tutors expand interactional practice at scale — pair them with human interaction so fluency and uptake transfer to real conversation.

  • Extend AI Scaffolding to collaborative reading. With 63 Chinese English majors in a GenAI-assisted social annotation study, Pan et al. (2026) found that the chatbot raised cognitive and emotional engagement and reading performance and tied social presence to integration and resolution, though high-achieving groups used the scaffold far more.

  • Readiness, not observed performance. Across 708 learners, greater AI-assisted speaking practice tracked higher human-directed willingness to communicate (β = 0.384) and lower foreign-language speaking anxiety (β = −0.416), while AI voice-assistant scenarios lowered anticipated anxiety further; neither study observed actual human speaking (Wang & Li (2026)).

  • Watch AI output for pragmatic, not just grammatical, error. Teachers across five online language-teaching methods named pragmatic blindness, where AI output is grammatically correct but wrong in tone, formality or culture (Baoyi and Khan (2026)).

  • Prioritize feedback quality over quantity in ASR-supported speaking. Chen et al. (2026) find that accurate error correction and structured reflection tasks improve Feedback internalization and reflective behavior in college English speaking, while frequent ASR use and recognition accuracy boost motivation or reflection only partially — technical precision alone does not drive deeper cognitive engagement, and language proficiency moderates the gains (stronger learners internalize feedback more effectively). This argues for pedagogically sound feedback (e.g., articulatory explanations over simple error flags), scaffolded reflection, and proficiency-differentiated support.

  • Give clues before corrections. A 'clue before correction' ChatGPT task, in which learners infer fixes from guided hints rather than receiving direct corrections, reduced cognitive load and supported personalized revision — but the benefit held mainly for learners who already had enough prior knowledge (58 students, CEFR A1–B1) (Lukešová & Jennings (2026)).

  • Deliver pronunciation feedback that localises the difference to close. Profy learns proficiency from largely unannotated speech and shows where a learner diverges from native-speaker distributions; its pre/post intelligibility confidence intervals did not overlap, unlike an elicited-imitation baseline — evidence that imitation practice can be supported without expert raters (Kawamura (2026)).

  • Use the transcript channel as a second signal, not a substitute. A WavLM audio classifier beat a zero-shot GPT-4o classifier reading transcripts on a CEFR-balanced learner corpus, so lexical traces of a first language are real but weaker than pronunciation, rhythm and intonation (Uluslu et al. (2026)).

  • Support learners' psychological adaptation to AI-assisted study. Wu (2026) tracks learners of Japanese over a semester and finds they sort into maladaptive, moderate, and positive adaptation profiles driven by the balance of technostress and resilience, with most learners gradually shifting toward positive adaptation and reporting higher Self-Efficacy and lower burnout — a signal to design AI-mediated language practice that manages technological strain, not just tool access.

  • Be alert to scoring and feedback bias against learners. AI scoring can penalize non-native patterns; linguistic-diversity research warns AI privileges standard English — use self-referential or human-moderated evaluation.

  • Support the full spectrum of learners. Dyslexia and accessibility studies and culturally responsive design (Arab-English contexts) show AI must be adapted to diverse learner needs, not assumed universal.

  • Infrastructure, not just scoring, excludes languages. Bengali holds under 0.5% of global web content against a 67:1 English-to-Bengali training-token deficit and a rural–urban connectivity gap (36.5% vs 71.4% internet penetration), so native-language, offline-first design is a prerequisite for access rather than a convenience (Roy and Roy (2026)).

  • Educators value GenAI for preparatory work, not live classroom use. A PRISMA systematic review of 23 studies (Li et al. 2026) finds language educators most value GenAI for behind-the-scenes preparation — lesson planning, materials creation, and writing support/feedback — yet remain hesitant about direct, classroom-facing implementation, reflecting a theory–practice gap between approving AI in principle and using it live. Adoption is shaped by professional-identity, pedagogical, technical, institutional, and Academic Integrity factors, with educators falling on a spectrum from non-adoption to comprehensive integration; attitudes tend to evolve from initial insecurity toward confident, selective use with exposure.

  • Prepare language teachers' AI literacy. Systematic reviews find AI literacy among language teachers is a key gap — invest in teacher professional development alongside tool adoption. As AI reshapes language education, AI literacy is also crucial for teachers to engage critically with the technology: the Teachers' AI Literacy Scale (TAILS) was developed for language teacher education, operationalizing the six-dimension ED-AI framework (knowledge, evaluation, collaboration, contextualization, autonomy, Ethics) and validated with preservice English language teachers.

  • Four interaction profiles in a high-pressure bilingual task. Kuang, Li and Weng (2026) used eye-tracking, pen-recording and voice-recording with 22 interpreting trainees to show that students divide attention between AI output and their own note-taking in four distinct ways — Intensive Engagers, Fast Scanners, Traditionalists and Frequent Switchers — and that 58.3% of stage-level observations changed profile between the comprehension and production stages of the same task. Only comprehension-stage patterns predicted product quality, and the AI-heaviest cluster scored lowest on fluency of delivery and target language quality, which makes the case for teaching learners to describe and reflect on their own strategy rather than prescribing one way of working with the tool.

  • Route the full four-skill loop around cost and connectivity. Guo et al. (2026) release LLMersion-1, a local-first prototype that works listening, reading, speaking and writing over the learner's own document on consumer hardware — a 1B tutor quantizes to 808 MB and the whole resident stack stays under 4 GB — and prices five years of daily practice at roughly $18 of electricity, against $1,200 for a cloud subscription. No learner outcomes are reported and pronunciation feedback is segmental only.

  • Evidence on chatbot speaking gains is uneven by feature. In the only systematic synthesis of the question, ten studies from 2021 to 2024 improved range and intelligibility most reliably, while coherence was examined once and interaction almost never; four of the ten reported a single holistic score (Soriano-Gonzalez & Belda-Medina (2026)).

Translation students can be moved off surface correction. In a news translation course, Xian et al. (2026) found search-engine consultation rising from a median of 15.6% to 27.7% of session time (p = 0.020, Holm-corrected) and knowledge-based searching from 7.2% to 28.8% (p = 0.010). Machine translation use fell from 28.5% to 5.1% without reaching significance (p = 0.338), and interactive AI use stayed at a median of zero. The lever was showing students their own process rather than banning a tool.

Common Misconceptions

  • "Embodied robot tutors work because of the embodiment." A meta-analysis of 11 RALL studies (N = 595) found a large pooled effect on L2 learning (g = 0.83) with high heterogeneity, and — importantly — only interaction-level features moderated it. Morphology and modality did not, so the hardware is not the active ingredient.
  • "AI-generated assessment items are worse than teacher-written ones." A comparison of 52 EFL assessment tasks rated by 20 experienced teachers found no significant quality difference overall between AI-generated and human-developed items, but divergence on particular dimensions. Overall parity with specific weaknesses is a different claim from improvement or deficiency.
  • "LLM feedback can only score writing." A discourse-level system goes further by naming the defect rather than scoring the text, with sentence-pair relation classification localising coherence breaks such as logical jumps and missing links. The useful output is diagnostic, not numeric.

Connected Concepts

Connected Articles

Connected FAQs

Connected Resources

  • mglearn Classroom Resources
    A large set of free, browser-based classroom activity suites and print-ready materials from TCEA, including Gen AI literacy and digital citizenship modules.

Metadata

DisciplineLanguage Learning, Writing Education
LevelHigher Ed, K 12
Technologies and techniquesGenerative AI
Equity, ethics, and responsible useEquity
Foundations of AI in educationAI in Education
SourceHistoryEditChat

Embed this page

Copy the code below to embed this page in a learning management system or other website. The embedded version shows only the page content — it hides the site header, navigation, and footer.