Advancing Translation Learning with AI: Dataset, Educational Exploration, and Multi-Agent Chatbot Design

Abstract

This research presents the design, development, and evaluation of an AI-powered chatbot to support English–Arabic translation learning in higher education. It is motivated by three key factors: (1) the heavy instructional workload faced by faculty members at the College of Languages, Princess Nourah bint Abdulrahman University, particularly in supervising and assessing student translation work across various stages of the learning process; (2) the availability of the Saudi Learner Translator Corpus, a rich bilingual dataset offering authentic translation material; and (3) recent advancements in AI and natural language processing, which enable the creation of intelligent, context-aware learning tools. The study contributes to AI-assisted translation education by introducing a structured methodology for generating high-quality English–Arabic parallel sentences from SauLTC and leveraging them to build a modular chatbot system. The chatbot integrates machine learning, deep learning, and NLP techniques, and performs a range of translation tasks, including translation generation, contextual example retrieval, learner translation evaluation, and competence testing. These tasks are implemented through multi-agent architecture, in which each agent is responsible for a specific function to deliver personalized, real-time feedback. The system combines retrieval-based and generative models to provide real-time, personalized support. For instance, one agent handles translation generation, another retrieves relevant examples from a bilingual corpus, a third evaluates user-provided translation, and a fourth tests overall translation competence using similarity scoring. This architecture design enables flexibility, scalability, and pedagogical support, enhancing the learner’s experience and translation accuracy. The system was rigorously evaluated through a mixed-methods approach: corpus quality was assessed using cosine similarity and expert review, generative and retrieval models were evaluated via embedding-based similarity, and the overall system performance was tested in a user study where participants interacted with the chatbot and completed a structured survey. The findings demonstrate the chatbot’s effectiveness in facilitating translation learning while reducing instructor workload, highlighting its potential as an innovative educational tool.