Search papers, labs, and topics across Lattice.
Nantes Université, École Centrale Nantes, CNRS, LS
2
0
4
Augmenting hierarchical models with corpus-level prototypes boosts Rhetorical Role Labeling, especially for rare roles, by up to 4 Macro-F1.
LLMs judging medical QA are heavily biased by the answer's source, but a little fine-tuning can make smaller models surprisingly fair.
