Beyond Translation: Human-Centered NLP for Cross-Lingual Communication

How can we develop NLP technology to effectively support cross-lingual communication, especially given recent progress in machine translation and multilingual language models? In this talk, I will present two main threads of work that aim to broaden the scope of machine translation to more directly support people's needs.

In the first thread, I'll consider the difficulty people face when weighing the potential benefits of machine translation against the risks it may pose. This difficulty arises because users—who typically do not speak either the input or output language—often cannot assess translation quality. I will present results from a human study in medical settings, which highlights the strengths and weaknesses of state-of-the-art quality estimation techniques. Next, I'll discuss how even accurate translations can fail when users lack background knowledge that is implied in the source language. I will introduce techniques for automatically generating explicitations that explain missing context by considering cultural differences between source and target audiences. Throughout, I will discuss ongoing research directions aimed at developing human-centered NLP approaches for cross-lingual communication.

Organized by:

MaiNLP research lab
LMU Munich