Multilingual NLP and Machine Translation (Master S3 Class)

Basic information

Description

This course explores the theoretical and practical dimensions of multilingual NLP and one of its most emblematic applications, machine translation. With nearly 7,000 languages spoken around the world, each with unique and highly diverse linguistic structures, the course examines the challenges of designing NLP systems that can accommodate this vast diversity. How can we develop technologies that are inclusive of resource-poor languages or those with minimal digital data? What are the implications of linguistic diversity for the development and use of multilingual NLP models?
Students will explore the principles and methods for training and using LLMs. The course will critically evaluate whether current models can truly capture and respect the diversity of linguistic phenomena documented by linguists, from syntax and semantics to phonetics. The course also provides an in-depth look at the challenges of building and evaluating machine translation systems, and introduces the scientific problems that remain unsolved in this dynamic field. Beyond practical applications, the course also explores how NLP can support the documentation, preservation and analysis of endangered languages. Students will learn how computational methods can help linguists to compare languages, discover relationships and improve our understanding of language evolution.

Prerequisites

  • Good knowledge of the basics of statistical learning and LLMs.
  • Good knowledge of Python and NLP libraries (in particular HuggingFace 🤗 and Pytorch)

Learning outcomes

On successful completion of this course, students should:

  • Train, deploy and evaluate a machine translation system
  • Understand the scientific issues and challenges involved in developing NLP methods capable of handling the full diversity of languages
  • Understand how multilingual models of text and speech work and know how to use them