- Spans are now at the token level, making the model more efficient.
tibert.run_testnow correctly supports all possible document splits configuration.- Fix a bug in mention pruning that would cause successive mention to always be discarded.
- Fix a bug when selecting antecedents that would greatly impact model's performance, specifically in terms of MUC Recall.