Systematic Review for AI-based Language Learning Tools

• ICALL emerged in response to greater emphasis on individual learner differences and applies AI methods to CALL, especially NLP, user modeling, expert systems, and intelligent tutoring systems.Source: . Introduction Ⅰ · p. 2
• AI language-learning tools include tutoring systems, writing assistants, virtual reality environments, chatbots, and adaptive systems, designed to personalize learning and adapt content to users’ pace, preferences, and needs - -.Source: . Introduction Ⅰ · p. 2
• Concerns include privacy, insufficient information, and teacher preparation; evidence on pedagogical effects and learner perceptions is needed to help teachers use these tools effectively and appropriately.Source: . Introduction Ⅰ · p. 2
• Existing reviews focus on particular AI tool types or AI’s future impact, leaving a need for an in-depth overview of recently developed tools, their technologies, and targeted language skills -.Source: . Introduction Ⅰ · p. 2

2-1 Research Aim and Study Design

• The review aimed to identify development trends in AI-based language learning tools and provide detailed information, asking what tools target different language skills and how they affect learning.The stated aims addressed concerns about insufficient information and teacher preparation in existing reviews.Source: 2-1 Research Aim and Study Design · p. 2
• The review focused on peer-reviewed journal articles published from 2017 to 2020, a three-year period chosen because AI was increasingly entering mainstream education.The authors linked the period to rapid growth in AI development and implementation of new techniques, algorithms, approaches, and models.Source: 2-1 Research Aim and Study Design · pp. 2–3
• The review followed PRISMA guidelines to report its purposes, methods, and findings and to identify and screen studies, as summarized in Figure 1.Figure 1 is the PRISMA chart of the review phases.Source: 2-1 Research Aim and Study Design · p. 3, Figure 1 · p. 3
• The review searched ERIC, Scopus, and Web of Science using an initial AI-and-language-learning query, limiting results to English-language peer-reviewed academic journals published from 2017 to 2020.The databases were selected for their accessibility and wide coverage of academic journals related to education.Source: 2-1 Research Aim and Study Design · p. 3
• A follow-up query using terms including ICALL, natural language processing, machine learning, intelligent tutor, bot, and virtual assistant returned 415 results.The follow-up search retained the same limits: English-language, peer-reviewed academic journals published from 2017 to 2020.Source: 2-1 Research Aim and Study Design · p. 3
• The screening process yielded 53 articles for the mixed-methods systematic review, after six non-empirical studies and three studies with insufficient methods or findings were excluded from 62 reviewed results.Earlier screening also removed duplicates, inaccessible results, and studies whose titles or abstracts did not indicate AI use for language learning.Source: 2-1 Research Aim and Study Design · p. 3

• Researchers coded study and participant characteristics quantitatively and AI technologies and learning findings qualitatively, then independently reviewed batches of 10 articles and reconciled notes to refine categories and identify trends.The coded characteristics included publication year, study design, location, target language and skill, grade, and proficiency; qualitative descriptions covered technology, pedagogical effects, and learner perceptions.Source: 2-2 Data Analysis · p. 3
• Studies clustered in 2019–2020 (N = 35, 66%); most were quantitative or mixed (N = 50, 94%), and English-focused (N = 37, 69%).Source: 2-2 Data Analysis · p. 3
• Targeted skills included speaking and listening (N = 14, 26%), writing (N = 11, 21%), and pronunciation (N = 11, 21%).Source: 2-2 Data Analysis · p. 3
• University students accounted for N = 32, 60%; beginners were the largest proficiency group (N = 10, 19%), though proficiency levels were relatively evenly distributed.Source: 2-2 Data Analysis · p. 4

3-1 Types of AI Tools Developed for Language Skill Areas

• Fourteen studies (26%) identified tools targeting speaking and listening skills.Source: 3-1 Types of AI Tools Developed for… · p. 4
• Four studies examined intelligent personal assistants such as Alexa for comprehensibility, usability, language-skill improvement, and willingness to communicate -.The reported areas included listening comprehension and speaking proficiency.Source: 3-1 Types of AI Tools Developed for… · p. 4
• Speaking and listening tools included programmable robots for group conversations ]14] and an NN-based dialogue system for free conversation practice.Source: 3-1 Types of AI Tools Developed for… · p. 4
• An NN-based multimodal dialogue system assessed spoken-language delivery, content, vocabulary, and grammar.Source: 3-1 Types of AI Tools Developed for… · p. 4
• Eleven studies (21%) identified writing tools, including machine translators, free-form writing software, and a blended course with automated feedback on self-correcting tasks.Source: 3-1 Types of AI Tools Developed for… · pp. 4–5
• Other writing systems addressed citations and referencing or rhetoric-category sentence classification, using latent semantic analysis, random forests, SVM, and Naive Bayes classifiers.Source: 3-1 Types of AI Tools Developed for… · p. 5

3) Pronunciation

• Eleven studies (21%) targeted pronunciation; seven used deep learning [22], with attention and neural-network architectures supporting diagnosis, training, and evaluation -.Examples included a multimodal system illustrating speech features and an interactive tool generating personalized voice models.Source: 3) Pronunciation · p. 5

4) Grammar

• Seven studies (13%) targeted grammar with games, applications, immersive environments, and intelligent systems using neural networks, machine learning, and natural language processing.Neural-network modeling predicted grammatical challenges from learners’ first languages to create customized study plans; NLP methods generated feedback -.Source: 3) Pronunciation · p. 5

5) Vocabulary

• Five studies (9%) targeted vocabulary with machine-learning and NLP-based systems, platforms, robots, games, and mobile applications.An ICALL platform used POS annotation and syntactic parsing to generate multiple-choice gaps that visually enhanced target vocabulary items; one adaptive app targeted both vocabulary and grammar.Source: 3) Pronunciation · p. 5

6) Reading

• Four studies (8%) targeted reading with machine-learning systems that diagnosed reading problems and pushed appropriate resources.A model also identified pedagogical factors distinguishing high- from low-achieving readers to improve ESL reading instruction.Source: 3) Pronunciation · p. 5

Summary

• Across the 53 studies, NLP appeared more often in writing, grammar, vocabulary, and reading tools, while neural networks appeared more often in speaking, listening, and pronunciation tools.NLP techniques included POS annotation, language modeling, and machine translation.Source: 3) Pronunciation · p. 5

Correcting Errors in Grammar

• The game-based group significantly outperformed the cloze exercise group (N = 15) and the cloze exercise group with corrections (N = 15) on writing tasks (p = 0.000).Source: 3-2 Impact of AI Tools on Language… · p. 6

Assessing and Evaluating Conversations

• AI tools assessed speaking, evaluated conversations, and provided responses; learners showed gains in speaking and listening and became more confident and less anxious, -, -.After a one-hour trial of Google Assistant, 29 participants rated improved listening comprehension at M = 4.28 and motivation to improve listening at M = 4.24; they also reported reduced stress and enjoyable practice.Source: 3-2 Impact of AI Tools on Language… · p. 6

Suggesting Words

• After three weeks using an ML-based emotion-synonym suggestion system, participants improved emotion-word use by 1.77 points (p < 0.01).Vocabulary tools also detected Japanese expressions for morphological analysis and example sentences, and learners gained semantic knowledge of phrasal verbs, -. Learners found the tools interesting and useful, and survey respondents reported satisfaction with serious-game features -.Source: 3-2 Impact of AI Tools on Language… · p. 6

4) Improving Writing

• Writing tools helped learners reduce plagiarism and correct rhetorical, lexical, and grammatical errors, while feedback improved essay organization, coherence, supporting ideas, and conclusions, -,.The tools also supported editing and revising and assessed writing abilities,, -.Source: 4) Improving Writing · p. 6
• Among 34 participants, writing scores improved significantly after a machine-translated, multiple-draft task.Scores rose from initial draft M = 3.76 to final draft M = 4.56 (p = 0.000); lexical errors fell from initial M = 5.97 to final M = 3.82, and grammatical errors from initial M = 15.67 to final M = 9.82 (p = 0.000).Source: 4) Improving Writing · p. 6
• Learners described writing tools as effective and easy to use, and said detailed comments helped them recognize strengths and weaknesses and build writing knowledge, -,,.Source: 4) Improving Writing · p. 6

5) Improving Fluency in Pronunciation

• Pronunciation tools detected mispronunciations and supported speech diagnosis and assessment; learners improved fluency, comprehensibility, tone, and pronunciation accuracy, -, -.Learners also found the tools interesting, easy to use, and helpful for fluency, intonation, and tone training,,.Source: 4) Improving Writing · p. 6

6) Personalized Tutoring

• Reading tools classified learners, assessed reading ability, and pushed tailored resources; an intelligent tutoring system improved academic reading skills in main idea, text structure, and inference, -.The adaptive system matched resources to learner characteristics, including reading abilities, cognitive styles, and learning objectives.Source: 4) Improving Writing · pp. 6–7

• AI tools target speaking, listening, writing, pronunciation, grammar, vocabulary, and reading, and learners have reported language gains and found the tools useful.Source: . Discussions Ⅳ · p. 7
• Teachers need familiarity with AI fundamentals to guide learners’ tool choices, while trying focused activities can build experience for student-centered implementation.Teachers can match tools to learners’ preferences and needs; trying a tool for a specific activity need not involve a long-term commitment.Source: . Discussions Ⅳ · p. 7
• Learners generally perceive the tools as effective, interesting, easy to use, and helpful; surveys and interviews could clarify which tools work best for particular learners.Source: . Discussions Ⅳ · p. 7
• The review was limited to English-language peer-reviewed journal articles from 2017–2020, three databases, and institutionally accessible studies, possibly skewing findings toward university English learners in China or Japan.Source: . Discussions Ⅳ · p. 7
• Future research should improve system effectiveness and recognition accuracy, incorporate pedagogical knowledge through teacher collaboration and classroom study, and assess diverse learners and long-term learning contributions.Source: . Discussions Ⅳ · p. 7
• Researchers plan to expand access with additional languages and proficiency-appropriate materials, and should monitor applications of large-scale models such as BERT and GPT-3 to language learning.Source: . Discussions Ⅳ · p. 7