Article: An exploration of a collaborative training framework for ESP vocabulary and dialogue using AR multimodal scenario data fusion and reinforcement learning-based reward mechanism optimisation Journal: International Journal of Reasoning-based Intelligent Systems (IJRIS) 2026 Vol.18 No.20 pp.1 - 15 Abstract: ESP teaching is commonly deficient in authentic contextual settings and instant feedback, resulting in poor training outcomes and weak transfer of professional lexical and situational expression competence. This paper constructs an AR-based multimodal context-aware reinforcement learning model (AR-MCSR-ESP) and conducts experiments on academic and occupational ESP with 186 participants split into experimental and control groups. The experimental group trained via the proposed model achieved prominent score increases: 20.04% (business English), 19.98% (medical English), 22.14% (legal English), 22.01% (GMAT) and 23.24% (GRE), while the control group saw gains below 10%. Independent-samples t-tests confirm significant intergroup differences (p < 0.001), and expert assessment yields large effect sizes (Cohen's d > 1.90) in dialogue quality, semantic precision and pragmatic appropriateness. This framework offers reliable situational ESP instruction for universities, vocational organisations and in-service continuing education to enhance learners' practical English proficiency. Inderscience Publishers - linking academia, business and industry through research

Title: An exploration of a collaborative training framework for ESP vocabulary and dialogue using AR multimodal scenario data fusion and reinforcement learning-based reward mechanism optimisation

Authors: Fenrong Cui; Pengfei Yang

Addresses: Department of Basic Courses, Shaanxi Vocational and Technical College, Xi'an, 710100, Shaanxi, China ' School of Computer Science and Technology, Xidian University, Xi'an, 710071, Shaanxi, China

Abstract: ESP teaching is commonly deficient in authentic contextual settings and instant feedback, resulting in poor training outcomes and weak transfer of professional lexical and situational expression competence. This paper constructs an AR-based multimodal context-aware reinforcement learning model (AR-MCSR-ESP) and conducts experiments on academic and occupational ESP with 186 participants split into experimental and control groups. The experimental group trained via the proposed model achieved prominent score increases: 20.04% (business English), 19.98% (medical English), 22.14% (legal English), 22.01% (GMAT) and 23.24% (GRE), while the control group saw gains below 10%. Independent-samples t-tests confirm significant intergroup differences (p < 0.001), and expert assessment yields large effect sizes (Cohen's d > 1.90) in dialogue quality, semantic precision and pragmatic appropriateness. This framework offers reliable situational ESP instruction for universities, vocational organisations and in-service continuing education to enhance learners' practical English proficiency.

Keywords: augmented reality; English for specific purposes; ESP instruction; multimodal scenario data; reinforcement learning reward mechanism; AR-MCSR-ESP; graduate record examination; GRE.

DOI: 10.1504/IJRIS.2026.155776

International Journal of Reasoning-based Intelligent Systems, 2026 Vol.18 No.20, pp.1 - 15

Received: 16 Jan 2026
Accepted: 10 Jun 2026
Published online: 13 Aug 2026 *

Title: An exploration of a collaborative training framework for ESP vocabulary and dialogue using AR multimodal scenario data fusion and reinforcement learning-based reward mechanism optimisation

Keep up-to-date