Evaluation Framework for Highlight Explanations of Context Utilisation in Language Models
A controlled evaluation of whether highlight explanations identify if a model uses context, which context it uses, and which spans inform its predictions.
Paper
