Over the last decade, language technologies have entered everyday life. However, while they excel for some tasks and languages, NLP is still far from being “solved.” Current systems, such as automatic translators or chatbots, mostly serve speakers of languages for which large amounts of data are available, but performance rapidly degrades when data is scarce. Other factors, such as limited consideration of the users during system design, further constrain the usefulness of language technologies. 

In this talk, I will provide an overview of my group’s recent work in three areas where state-of-the-art NLP systems still struggle to be helpful. First, I will discuss research on NLP for language documentation. Second, I will present work on NLP for Indigenous languages of the Americas. Third, I will describe research on LLMs for German dialects. I will conclude the talk with a discussion of key challenges the field of NLP will need to address in the coming years.