Questions asked in Mandarin, answers in both languages and a technical glossary generated on the fly: a multi-source RAG pipeline tested on 91 Piazza messages.
With the influx of international students into computer science courses, the language barrier is becoming an obstacle that teaching teams are struggling to overcome. Researchers propose a tutor based on an LLM capable of receiving questions in a language other than English, for the moment Mandarin, and answering bilingually.
The system is based on a Retrieved Augmented Generation (RAG) that draws on several types of course materials, with a dedicated mechanism to isolate the real question in an often long or confusing message. It also produces a glossary of technical terms used, along with explanations, so that the student learns the English vocabulary of the discipline instead of just bypassing it.
Evaluated on 91 real posts from the Piazza forum of a course, the tutor gave a fair and useful answer in 83.52% of the cases. The translations were judged to be of good quality for 76.92% of the questions, and a relevant glossary was produced for 64.47% of them. The authors see this as a support that is difficult to offer with only the human assistants of a course.
Read the full article on the original source
Go to the source →