CAREER: Self-Directed Human-LLM Coordination for Language Learning and Information Seeking

University Of Maryland, College Park, College Park MD

Investigators

Abstract

Effective seeking of information is a crucial aspect of daily life that relies heavily on language skills. In societies with a diverse population, many people are not proficient in the majority language. People with less majority-language proficiency face challenges communicating their needs in daily activities such as medical consultations, discussing workplace benefits with employers, exploring housing, and others. A typical solutions pairs these individuals with professional interpreters or bilingual volunteers. Such support is often expensive and unsustainable. To fill the gap, this project will develop innovative tools based on AI Large Language Models (LLMs) to help people advance their language abilities for effective information seeking. Users without technical expertise in computers or AI will be guided to design their own personalized tutoring systems. The resulting digital tutor will assist its user in creating learning plans, practicing strategies, and tracking their progress in advancing their language proficiency. The study will generate rich data, metrics, and benchmarks for language learning for real-life information-seeking practices. The project will advance the science of human-computer interaction, language science and education, and natural language processing. The process of designing and using their own digital tutor will increase user's understanding of AI (AI literacy), and both the power and limitations of LLMs. The tutoring systems will assist users in navigating various information-seeking scenarios, ultimately improving the person's language proficiency and quality of life. This project provides a novel approach to tackling the challenges faced by people with less majority-language proficiency. They will go through a process wherein they a) design, utilize, and refine an AI-enabled tutoring system that guides their language use for situational purposes, and b) advance their language proficiency along a learning path tailored by themselves to their needs. The experience will stimulate the individual's critical thinking regarding actions to take at the different stages of their language learning. Using the system provides the individual with feedback to adjust their learning to improve outcomes. There are four sets of research activities involved in the lifecycle of this project. Activity 1 employs experience sampling to pinpoint individuals' current practices and challenges in language use for daily information seeking. It help pinpoint the user's essential needs in fine-grained manner. Activity 2 compiles a bank of system building blocks, setting the stage for each individual-as-designer's assembly and customization of their tutoring system. Activity 3 engages each individual-as-learner-and-designer in crafting, using, and refining their tutoring system. It will generate longitudinal data to track the progress of a person's language use in daily information seeking, as well as their coordination with the tutoring system. Activity 4 initiates an open data program to uncover value trade-offs in interactions between participants and large language models. It will enable the exploration of emerging ways to align model-generated content with individuals' core values. Together, these efforts will generate essential knowledge, novel methods and datasets, functioning tool-kits, and educational materials for enhancing language minorities' information seeking through effective language use. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.