Invited talks

Marie-Catherine de Marneffe

CENTAL – UCLouvain / FNRS

Diversity in NLP: why, what, where and how

Diversity has become an advocated property of NLP datasets and systems. However, its use remains largely ad hoc, with limited theoretical grounding and definitions and measures that vary across studies. In this talk, I will describe a recent survey with colleagues of 308 diversity-related papers from the ACL Anthology. We propose an NLP-specific framework organized around four perspectives: why diversity matters, what is measured, where it is measured, and how, drawing from insights from other scientific fields where the concept of diversity has been thoroughly conceptualized. I will also look at Universal Dependencies from these perspectives.

Stephen Mayhew

Duolingo

Universal NER: Standing on the Shoulders of Giants

The huge success of Universal Dependencies has been an inspiration for similar projects. One such descendant is Universal NER, a large-scale, multilingual, shared-schema named entity recognition project, in the works since 2020. This talk will cover a brief history of the project, the philosophy behind the data, the tagset, and the annotations, as well as details on the data itself, including the latest release, Universal NER v2, presented at LREC 2026.