Structure and Implementation of New Practical English Textbooks Driven by Artificial Intelligence

• Static digital textbooks retain printed materials’ fixed sequences [1], although learners need contextual practice, prompt correction, and task transfer.Source: §1 · p. 1
• Standalone AI tools do not resolve textbook-level integration of content, pedagogy, assessment, and teacher orchestration, and feedback can miss unit goals.Source: §1 · pp. 1–2
• The study proposes a multi-level structure linking curriculum knowledge, learner modeling, adaptive task sequencing, formative feedback, and teacher governance, evaluated with classroom indicators over eight weeks.The stated indicators include task accuracy, speaking performance, feedback latency, and teacher workload.Source: §1 · p. 2

2.1. AI-supported English learning and feedback

• Chatbots, automated writing evaluation, and generative AI support interaction, feedback, and flexible materials, but teachers must verify generated content.Source: §2.1 · p. 2
• Most prior work centers on tools or activities rather than textbooks, leaving goals, lesson-plan connections, and consistent learning structures insufficiently integrated.The paper describes needed textbook features as a stable knowledge map, reusable task templates, observable progress indicators, and consistent editorial policies.Source: §2.1 · p. 2

2.2. Gaps in practical English textbook implementation

• Implementation gaps include coarse content granularity, assessment evidence that does not alter later learning paths, and weak teacher management of generated activities.Source: §2.2 · p. 2
• The proposed design treats the textbook as an organized environment that preserves curriculum logic while adding diagnostic, adaptive, and analytical functions.The engineering question is how to make textbook content machine-readable, pedagogically sound, and operationally feasible.Source: §2.2 · pp. 2–3

3.1. Textbook knowledge map and learner profile

• Unit knowledge maps connect communicative functions, vocabulary, grammar, pragmatics, and target outcomes to tasks with clear instructional objectives.The framework figure depicts connections among curriculum nodes, learner profiles, AI services, teacher feedback, and adaptive textbook pages.Source: §3.1 · p. 3, Figure 1 · p. 3
• Learner profiles record mastery, error patterns, response time, preferred learning modes, and feedback history; readiness combines concept mastery, task accuracy, engagement stability, and feedback uptake.Teachers configure the readiness coefficients at course level to avoid hidden pedagogical priorities.Source: §3.1 · p. 3

3.2. Adaptive Task Generation and Feedback Loop

• Adaptive tasks use pre-approved templates with input materials, expected outputs, difficulty tags, feedback rules, and evidence tags, filling scenario variables from a controlled resource bank.The paper states this approach avoids factual drift and keeps activities aligned with the syllabus and communication purpose.Source: §3.2 · pp. 3–4
• After sessions, performance differences update adaptive model parameters, influencing subsequent task difficulty, skill coverage, and remediation priority without changing teacher-locked curriculum requirements.The feedback loop includes real-time error reports, delayed recurring-error analysis, and teacher correction of pragmatics and discourse.Source: §3.2 · p. 4

3.3. Teacher Governance and Textbook Delivery

• Teachers can lock tasks, approve or reject exercises, set difficulty limits, and export class data; the system provides instructions but does not replace teachers.Source: §3.3 · p. 4
• The web-based textbook and teacher dashboard support adaptive exercises, feedback, progress monitoring, homework, generated-work review, and feedback tracking.The efficiency index weights mastery gain, productive-language performance, unit completion, teacher workload reduction, and governance compliance; evaluation also includes instructional quality and learner participation.Source: §3.3 · pp. 4–5

4.1. Experimental setting and overall learning performance

• 84.9% vs 72.4% completion accuracy, 81.7% vs 70.2% listening-response accuracy, 79.4% vs 68.1% scenario-writing accuracy, and 76.8% vs 66.0% speaking-task accuracy favored AI.The four intact classes included 186 first-year non-English majors; groups studied the same eight units and were taught by the same teaching team.Source: Figure 3 · p. 5, §4.1 · p. 5

4.2. Feedback efficiency and learner progression

• Feedback latency for basic errors fell from 26.5 hours to 3.2 minutes, revision acceptance rose from 48.6% to 71.3%, and repeated pragmatic errors per 100 tasks fell from 18.9 to 10.7.Source: §4.2 · p. 6
• After eight weeks, mastery index rose from 0.43 to 0.82 for AI and from 0.42 to 0.68 for control, while teacher correction time fell 31.6% compared with control.Teacher correction time dropped from 7.9 hours in Week 1 to 4.2 hours by Week 8.Source: Figure 4 · p. 6, §4.2 · p. 6

4.3. Robustness, usability, and implementation risk analysis

• Low-readiness learners improved by 18.4 points, compared with 15.2 points for medium-readiness and 10.1 points for high-readiness learners.Source: §4.3 · p. 6
• Task clarity (4.31), feedback usefulness (4.26), and interface consistency (4.18) led learner usability ratings; feedback trust was the lowest implementation indicator at 79.8%.Teacher-side governance scored 4.07, lower than learner-side indicators because batch-editing functions were lacking.Source: §4.3 · p. 6, Figure 5 · p. 7
• 6.4% of generated tasks required manual correction before release, and implementation safeguards included validated scenarios, limited data collection, prompt deletion of speaking records, and teacher override rights.Source: §4.3 · p. 7

5. Conclusion

• The proposed five-layer architecture links knowledge mapping, learner profiling, adaptive task generation, feedback orchestration and teacher governance within a stable curriculum.The prototype was found to provide more diagnostic, adaptive and evidence-based textbook content for practical English learning.Source: §5 · p. 7
• The experiment covered one university and eight weeks, and the prototype addressed general practical English rather than specialized fields in detail.The authors caution that results may not generalize to extended curriculum application at other institutions.Source: §5 · p. 7
• Future work will test the prototype across institutions, over longer learning cycles and in more specialized practical English areas.The authors also identify multimodal processing, privacy-preserving learner modelling and teacher-facing analytical explanations for improvement.Source: §5 · p. 7
• The authors describe an excellent AI-powered textbook as transparent, manageable and pedagogically sound, rather than merely an intelligent addition to existing materials.Source: §5 · p. 7