Closed ijyliu closed 9 months ago
average embeddings
isn't this similar to the initial bert trained task of sentence sequences?
in jurafsky
with our sentence to sentence and paragraph averaged embeddings, we do get a good sense of both local and global coherence
footnote: other structural global methods were considered, but not implemented due to the short length of the task responses at only two paragraphs