LingoQ: Bridging the Gap between EFL Learning and Work through AI-Generated Work-Related Quizzes

TL;DR

LingoQ tackles the problem of disconnect between EFL learning and real work tasks by generating work-contextual quizzes from workers' AI-assisted language queries. The system combines LingoQuery (desktop chatbot), LingoQuiz (mobile quizzes), and backend pipelines to produce and curate TOEIC/TOEFL-style questions tailored to ongoing work tasks, enabling low-burden, just-in-time practice. A three-week deployment with 28 EFL workers shows strong engagement, high-quality questions validated by experts, and significant improvements in self-efficacy, with measurable gains for beginners and potential for other proficiency levels through richer interaction. The study highlights design considerations for AI-mediated language learning in the workplace, including privacy, boundary management, and adaptive content, illustrating a practical path to leveraging growing LLM reliance for proficiency and engagement gains.

Abstract

Non-native English speakers performing English-related tasks at work struggle to sustain EFL learning, despite their motivation. Often, study materials are disconnected from their work context. Our formative study revealed that reviewing work-related English becomes burdensome with current systems, especially after work. Although workers rely on LLM-based assistants to address their immediate needs, these interactions may not directly contribute to their English skills. We present LingoQ, an AI-mediated system that allows workers to practice English using quizzes generated from their LLM queries during work. LingoQ leverages these on-the-fly queries using AI to generate personalized quizzes that workers can review and practice on their smartphones. We conducted a three-week deployment study with 28 EFL workers to evaluate LingoQ. Participants valued the quality-assured, work-situated quizzes and constantly engaging with the app during the study. This active engagement improved self-efficacy and led to learning gains for beginners and, potentially, for intermediate learners. Drawing on these results, we discuss design implications for leveraging workers' growing reliance on LLMs to foster proficiency and engagement while respecting work boundaries and ethics.

Figures (10)

• Figure 1: Main window and the interface components of LingoQuery. Users can open new chat threads of their choice in the thread list or via the New Chat button (ⓐ). The AI's responses for three major types of query intents---Look up (ⓑ), Translate (ⓔ), and Proofread (ⓕ)---provide the UI components tailored to each query type. The new message panel (ⓒ) incorporates a query prompt panel at the top. Users can insert a template query prompt in their message via the three quick-access buttons. When a query is sent via a keyboard shortcut, the system analyzes a screenshot of the user's active window, and the user can review both the surrounding context of the copied sentence and the task at the time of the query ⓓ. By clicking the star icon below AI responses (ⓑ), users can mark the message to increase the likelihood that the corresponding question will be included in LingoQuiz.
• Figure 2: Conversational pipeline of LingoQuery. When the user sends a new message ⓐ, the intent classifier ⓑ identifies the query intent ⓒ, which is then passed to the response generator ⓔ together with the chat history ⓓ. The response generator produces an appropriate response ⓕ structured according to the query intent. Finally, LingoQuery renders this structured response accordingly ⓖ.
• Figure 3: Question generation pipeline of LingoQ. When a query–response pair arrives ⓐ, the language query filter ⓑ identifies the query intent ⓒ, which is then passed to the Question generatorⓕ together with work contexts ⓓ and exam samples ⓔ. The generator produces two candidate questions ⓖ, which are evaluated by the Question evaluatorⓗ on two criteria: answerability and proficiency. The Question refinerⓘ refines each question up to two iterations, and items that still fail are discarded. Accepted questions are stored in the question pool ⓙ.
• Figure 4: Main screens of LingoQuiz. In the Dashboard screen [boxparam, border-radius=0pt, padding-left=2pt, padding-right=2pt, height=5.5pt, border-width=0pt, background-color=darkgray]A, users can check their records and stats, along with the number of new questions add to their question pool today. When starting a quiz by pressing the Start Quiz button ⓐ, a quiz with 10 unique questions are provided sequentially [boxparam, border-radius=0pt, padding-left=2pt, padding-right=2pt, height=5.5pt, border-width=0pt, background-color=darkgray]B. The new question that appears to the user for the first time is indicated by the star icon (ⓑ). For questions generated from messages with context, the task description is provided (ⓒ). To solve the question, the user can select an option (ⓓ) and press the Submit button (ⓕ) to submit an answer. Then the result the question is shown immediately, with explanation (ⓖ), regardless of whether the user had selected a correct answer or not. After solving the ten questions, questions with wrong answers appears again, until all are answered correctly. The progress bar (ⓔ) indicates the current progress. In the Ending screen [boxparam, border-radius=0pt, padding-left=2pt, padding-right=2pt, height=5.5pt, border-width=0pt, background-color=darkgray]C, users can practice a new quiz or return to the Dashboard screen.
• Figure 5: Selected questions actually generated by LingoQ during the deployment study for three participants. Each question consists of a stem with a blank, context reminding the task at the time of query (red text), and three alternatives with correct answers underlined.