AI ASSISTANTS AS DIALOGUE PRACTICE PARTNERS: EXPERIENCE OF APPLICATION IN THE EFL CLASSROOM

Yirik til modellari (LLM) asosidagi suhbatvositalarining – xususan, ChatGPT va uning ixtisoslashganhosilalarining – tobora kengroq tarqalishi chet til sifatida ingliz tili (EFL)pedagogikasiga sifat jihatidan yangi imkoniyatlar olib kirmoqda. Ushbumaqola EFL muhitida sun‘iy intellekt yordamchilaridan muloqotamaliyoti sherigi sifatida foydalanishga oid zamonaviy bilimlarningnazariy asoslangan va empirik ma‘lumotlarga tayanuvchi sintezinitaqdim etadi. Asosan 2022–2025 yillarda chop etilgan ilmiy tekshirilgantadqiqotlarga tayangan holda, maqola sun‘iy intellekt vositachiligidagimuloqot amaliyotining uch o‘lchovini ko‘rib chiqadi: ijtimoiy-madaniyo‘qitish nazariyasi va kommunikativ til o‘qitish yondashuvi doirasidaginazariy asoslar; chet tilda nutq tashvishi, muloqotga tayyorlik, og‘zakiravonlik va o‘quvchi mustaqilligiga ta‘sirining hujjatlashtirilgannatijalari; hamda mas‘uliyatli integratsiya uchun bartaraf etilishi zarur  bo‘lgan tanqidiy pedagogik cheklovlar. Maqolada sun‘iy intellekt muloqotsheriklari – o‘qituvchi rahbarligidagi pedagogik tizim doirasida maqsadliqo‘llanilganda – qimmatli qo‘shimcha kommunikativ amaliyotnita‘minlashi ta‘kidlanadi. Bu ayniqsa O‘zbekiston va keng Markaziy Osiyokabi haqiqiy ingliz tilida muloqot imkoniyatlari tarkibiy jihatdancheklangan kontekstlar uchun ahamiyatlidir. Maqola kommunikativ tilrivojlanishining insoniy o‘lchamlarini siqib chiqarmagan holda sun‘iyintellekt vositalarini EFL gapirishni o‘rgatishga integratsiya qilishga oiddalillarga asoslangan tavsiyalar bilan yakunlanadi.

AI ASSISTANTS AS DIALOGUE PRACTICE PARTNERS: EXPERIENCE OF APPLICATION IN THE EFL CLASSROOM

Abdurakhimova Nigora Bakhtiyarovna Renaissance Educational University ESL Teacher nigora81@inbox.ru

Abstract. The growing availability of large language model (LLM)-based conversational tools – most notably ChatGPT and specialised derivatives – has introduced a genuinely transformative resource into English as a Foreign Language (EFL) pedagogy. This article offers a theoretically grounded and empirically informed synthesis of current knowledge concerning the use of AI assistants as dialogue practice partners in EFL settings. Drawing on peer-reviewed research published primarily between 2022 and 2025, the paper examines three dimensions of AI-assisted dialogue practice: its theoretical foundations in sociocultural learning theory and communicative language teaching; its documented effects on foreign language speaking anxiety, willingness to communicate, oral fluency, and learner autonomy; and the critical pedagogical limitations that must be addressed for responsible integration. The article argues that AI dialogue partners, deployed purposefully within a teacher-guided framework, offer meaningful supplementary communicative practice – particularly in contexts, such as Uzbekistan and broader Central Asia, where learners have structurally limited access to authentic English interaction. The paper concludes with evidence-based recommendations for integrating AI dialogue tools into EFL speaking instruction without displacing the irreducibly human dimensions of communicative language development. Keywords: AI conversational agents, EFL speaking practice, foreign language speaking anxiety, willingness to communicate, communicative competence, ChatGPT, Zone of Proximal Development, language acquisition.

ИИ-АССИСТЕНТЫ КАК ПАРТНЁРЫ ДЛЯ ДИАЛОГОВОЙ ПРАКТИКИ: ОПЫТ ПРИМЕНЕНИЯ В КЛАССЕ АНГЛИЙСКОГО ЯЗЫКА КАК ИНОСТРАННОГО

Аннотация. Растущая доступность разговорных инструментов на основе больших языковых моделей (LLM) – прежде всего ChatGPT и его специализированных производных – открывает качественно новые возможности для педагогики английского языка как иностранного (EFL). Настоящая статья предлагает теоретически обоснованный и эмпирически подкреплённый синтез современных знаний об использовании ИИ-ассистентов в качестве партнёров для

диалоговой практики в условиях EFL-обучения. Опираясь на рецензируемые исследования, опубликованные преимущественно в 2022–2025 годах, статья рассматривает три измерения ИИопосредованной диалоговой практики: её теоретические основания в рамках социокультурной теории обучения и коммуникативного подхода к преподаванию языков; задокументированное влияние на иноязычную речевую тревожность, готовность к коммуникации, устную беглость и учебную автономию; а также критические педагогические ограничения, которые необходимо учитывать при ответственной интеграции. В статье утверждается, что ИИ-партнёры для диалога, применяемые целенаправленно в рамках педагогической структуры под руководством преподавателя, обеспечивают ценную дополнительную коммуникативную практику – особенно в таких контекстах, как Узбекистан и другие страны Центральной Азии, где учащиеся структурно ограничены в доступе к аутентичному общению на английском языке. Статья завершается основанными на доказательствах рекомендациями по интеграции ИИ-инструментов в обучение говорению на EFL без вытеснения необходимых человеческих измерений коммуникативного языкового развития. Ключевые слова: ИИ-разговорные агенты, диалоговая практика в EFL, иноязычная речевая тревожность, готовность к коммуникации, коммуникативная компетенция, ChatGPT, Зона ближайшего развития, усвоение языка.

SUN'IY INTELLEKT YORDAMCHILARI MULOQOT AMALIYOTI SHERIGI SIFATIDA: INGLIZ TILINI XORIJIY TIL SIFATIDA O'QITISH DARSIDA QO'LLASH TAJRIBASI

Annotatsiya. Yirik til modellari (LLM) asosidagi suhbat vositalarining – xususan, ChatGPT va uning ixtisoslashgan hosilalarining – tobora kengroq tarqalishi chet til sifatida ingliz tili (EFL) pedagogikasiga sifat jihatidan yangi imkoniyatlar olib kirmoqda. Ushbu maqola EFL muhitida sun'iy intellekt yordamchilaridan muloqot amaliyoti sherigi sifatida foydalanishga oid zamonaviy bilimlarning nazariy asoslangan va empirik ma'lumotlarga tayanuvchi sintezini taqdim etadi. Asosan 2022–2025 yillarda chop etilgan ilmiy tekshirilgan tadqiqotlarga tayangan holda, maqola sun'iy intellekt vositachiligidagi muloqot amaliyotining uch o'lchovini ko'rib chiqadi: ijtimoiy-madaniy o'qitish nazariyasi va kommunikativ til o'qitish yondashuvi doirasidagi nazariy asoslar; chet tilda nutq tashvishi, muloqotga tayyorlik, og'zaki ravonlik va o'quvchi mustaqilligiga ta'sirining hujjatlashtirilgan natijalari; hamda mas'uliyatli integratsiya uchun bartaraf etilishi zarur

bo'lgan tanqidiy pedagogik cheklovlar. Maqolada sun'iy intellekt muloqot sheriklari – o'qituvchi rahbarligidagi pedagogik tizim doirasida maqsadli qo'llanilganda – qimmatli qo'shimcha kommunikativ amaliyotni ta'minlashi ta'kidlanadi. Bu ayniqsa O'zbekiston va keng Markaziy Osiyo kabi haqiqiy ingliz tilida muloqot imkoniyatlari tarkibiy jihatdan cheklangan kontekstlar uchun ahamiyatlidir. Maqola kommunikativ til rivojlanishining insoniy o'lchamlarini siqib chiqarmagan holda sun'iy intellekt vositalarini EFL gapirishni o'rgatishga integratsiya qilishga oid dalillarga asoslangan tavsiyalar bilan yakunlanadi. Kalit so'zlar: sun'iy intellekt suhbat agentlari, EFL muloqot amaliyoti, chet tilda nutq tashvishi, muloqotga tayyorlik, kommunikativ kompetensiya, ChatGPT, Yaqin rivojlanish zonasi, til o'zlashtirish.

Introduction. Authentic spoken interaction in the target language remains one of the most valuable – and least available – resources for learners of English as a Foreign Language. In contexts where English functions exclusively as a foreign rather than a second language, including much of Central Asia, learners depend almost entirely on the formal classroom setting for opportunities to produce extended spoken English. Yet the realities of institutional education – large class sizes, rigid timetabling, and the competing demands of grammar instruction and examination preparation – conspire to reduce per-learner speaking time to a matter of minutes per session (Richards, 2015; Hanzawa, 2021). The structural result is a persistent and welldocumented gap between classroom language learning and the spontaneous communicative performance that genuine proficiency requires. The rapid maturation of large language model (LLM) technology has opened a qualitatively new avenue for addressing this gap. Contemporary AI conversational tools – including OpenAI's ChatGPT, voice-enabled platforms such as CallAnnie and Hi Echo, and purposebuilt EFL applications such as Mondly and Liulishuo – are capable of sustaining coherent, contextually appropriate, and linguistically varied dialogue across an essentially unlimited range of topics and registers. Unlike earlier generations of computer-assisted language learning (CALL) software, which were constrained to pre-scripted patternmatching interactions, these tools can respond adaptively to learner input, generate implicit and explicit feedback, and sustain extended interactional sequences that approximate authentic communicative practice (Wan & Moorhouse, 2024; Hwang & Chen, 2023). A rapidly growing body of empirical research – including systematic reviews, metaanalyses, and quasi-experimental studies conducted across diverse EFL contexts – has begun to map the conditions under which AI-assisted

dialogue practice most effectively supports communicative development (Jeon, 2025; Wu & Li, 2024; Du & Daniel, 2024). Theoretical framework. Three theoretical traditions underpin the use of AI-assisted dialogue in EFL instruction: Vygotsky's (1978) sociocultural theory, communicative language teaching (CLT), and Long's (1996) interactional hypothesis. Vygotsky's Zone of Proximal Development (ZPD) – the gap between what a learner can do independently and what they can achieve with support – provides the core rationale. Generative AI can fulfil the role of the 'more knowledgeable other' (MKO): tailoring output to the learner's proficiency level, providing calibrated scaffolding, and sustaining the iterative dialogue through which language is internalised (Sidorkin, 2025; Lantolf et al., 2014). Importantly, the ZPD emerges from the quality of the interaction itself, not from the tool alone – making purposeful task design essential. CLT holds that language is acquired most effectively through meaningful, purposive interaction rather than form-focused drilling (Richards & Rodgers, 2001). AI dialogue partners align naturally with this principle, generating contextually situated exchanges that require learners to produce and negotiate language in real time. Long's (1996) interactional hypothesis reinforces this further: negotiation of meaning – the collaborative repair of communicative breakdowns – is a primary driver of acquisition, and well-designed AI systems can simulate this process through clarification requests, reformulations, and targeted prompts. Finally, Willingness to Communicate (WTC) – a learner's readiness to initiate target language interaction – is depressed by speaking anxiety, low self-confidence, and fear of evaluation (MacIntyre et al., 1998). AI dialogue partners structurally address all three barriers by creating a non-judgmental, consequence-free practice environment. Research situates this within Informal Digital Learning of English (IDLE) as a productive interaction activity (PIA) – a modality shown to predict WTC development more strongly than passive digital engagement (Lee & Drajati, 2019; Rahimi & Fathi, 2024). AI assistants as dialogue partners: empirical evidence. Foreign language speaking anxiety (FLSA) is among the most robustly documented barriers to EFL communicative development (MacIntyre et al., 1998). The non-judgmental, low-stakes environment of AI-mediated interaction allows anxious learners to take communicative risks that conventional classroom settings inhibit. Empirical support is consistent across contexts: Fathi et al. (2024) found significant FLSA reductions and WTC gains following a six-week AI speaking intervention; Wang et al. (2024) demonstrated that avatar-based AI chatbots reduced anxiety

while enhancing self-perceived communicative competence; and Du and Daniel's (2024) systematic review confirmed that AI chatbots consistently reduce anxiety and improve speaking outcomes across multiple EFL settings. Beyond anxiety, AI dialogue partners produce measurable gains in oral fluency and proficiency. Wu and Yu (2024) found that GPT-4- powered chatbot practice yielded greater fluency and content gains than worksheet-based activities, while Hwang et al. (2024) demonstrated that AI mobile applications produced improvements across vocabulary use, grammatical accuracy, and fluency in post-classroom independent practice. A further advantage is personalised, adaptive feedback. Unlike classroom teachers managing large groups, AI systems respond continuously to individual learner output. Huang et al. (2022) identify personalisation, user-friendliness, and timeliness as the defining features of high-impact AI language tools. Jeon et al. (2023) confirmed that AI systems providing targeted corrective feedback produced significant speaking gains, findings reinforced by Wu and Li's (2024) meta-analysis identifying personalised error correction as the strongest predictor of learning outcome improvement. Finally, AI partners can simulate an unlimited range of communicative scenarios – from informal exchanges to professional and institutional communication – exposing learners to the genre diversity that cross-contextual competence requires. This is particularly significant in contexts such as Uzbekistan, where authentic English interaction is structurally limited (Nizomova, 2025; Functional-Stylistic Features of Uzbek Legal Discourse, 2024). Rahimi and Fathi (2024) further found that AI dialogue practice enhanced learner autonomy – improving metacognitive awareness and strengthening self-regulated learning dispositions. Critical considerations and documented limitations. Despite the empirical support surveyed above, AI dialogue systems carry well-documented limitations that require direct engagement. The most theoretically significant concerns pragmatic and sociolinguistic authenticity. While LLMs produce linguistically fluent output, they do not reliably model sociolinguistic variation, cultural specificity, or the pragmatic conventions of authentic human interaction. Learners may develop grammatical accuracy without acquiring the register sensitivity and audience awareness that genuine communicative competence demands. As Nizomova (2025) underscores, communicative competence is not reducible to linguistic form – it depends on understanding how social roles and institutional contexts shape effective communication.

A second limitation involves phonetic feedback. Du and Daniel (2024) note that AI chatbots consistently struggle to capture nuances of sound production, stress, and intonation – features critical for EFL learners managing strong L1 phonological transfer. Human or specialist pronunciation support therefore remains necessary. Third, sustained AI dialogue practice risks fostering over-reliance, whereby learners passively accept AI-generated language rather than actively constructing their own communicative output. Generative AI tends to provide direct answers rather than encouraging knowledge coconstruction – the very process most conducive to genuine language development (Arxiv, 2025). Finally, equitable access cannot be assumed. Reliable connectivity, platform subscriptions, and digital literacy competencies are unevenly distributed across EFL learner populations. Technological availability must not be conflated with accessibility, and AI integration must be accompanied by systematic digital competence development (Zamista & Azmi, 2023). Pedagogical implications and recommendations. The evidence reviewed in this paper points to three core recommendations for EFL educators integrating AI dialogue partners into speaking instruction. First, AI practice should be task-structured rather than openended. Learning outcomes are maximised when AI interaction is embedded within clearly specified communicative tasks – simulating a job interview, presenting an argument, or narrating an experience. Abdallah (2025) proposes a three-phase model: pre-activity vocabulary activation, main-activity AI dialogue, and post-activity peer reflection – ensuring AI functions as part of a coherent pedagogical sequence rather than an isolated supplement. Second, AI dialogue should complement rather than replace human interaction. Motivation, identity, and authentic communicative purpose are irreducibly human dimensions that AI cannot replicate. Teachers remain indispensable as architects of the learning environment – setting objectives, scaffolding tasks, and ensuring AI serves as a genuine developmental tool rather than a cognitive shortcut. Third, tool selection must be guided by pedagogical criteria: contextual appropriateness, feedback quality, curriculum alignment, and data privacy compliance. Not all AI dialogue tools are pedagogically equivalent, and novelty or commercial popularity should not substitute for evidence-based evaluation (Abdallah, 2025). Conclusion. The integration of AI conversational assistants as dialogue practice partners represents one of the most practically significant developments in EFL pedagogy of the past several years. The convergent evidence reviewed in this article indicates that AI-mediated

dialogue practice can reduce foreign language speaking anxiety, improve oral fluency and self-perceived communicative competence, provide personalised and adaptive scaffolding, expand learners' communicative repertoire across genres and registers, and support the development of autonomous learning dispositions. These effects are most robust when AI practice is purposefully designed, theoretically grounded in ZPD and CLT principles, and positioned within a broader pedagogical framework that preserves the irreducibly human dimensions of communicative language development. The limitations documented in the literature – reduced pragmatic authenticity, weak phonetic feedback, over-reliance risk, and inequitable access – are real and consequential. They call not for rejection of AI dialogue practice but for principled, critically informed integration that addresses these constraints directly. Future research should prioritise longitudinal investigations examining sustained developmental trajectories, context-specific studies within Central Asian and Uzbek EFL settings where authentic English interaction is structurally limited, and systematic work on the task design principles that most effectively leverage the interactional affordances of contemporary AI systems for communicative language development. Ultimately, AI assistants extend rather than replace the fundamentally human dimensions of language learning. They make sustained communicative practice accessible to learners who would otherwise lack it – and in doing so, they offer a meaningful contribution to closing the gap between classroom instruction and the authentic communicative competence that English language education aspires to develop.