Fine-tuned T5 small model for use as a frame semantic parser in the Frame Semantic Transformer project. This model is trained on data from FrameNet.
Usage
This is meant to be used a part of Frame Semantic Transformer. See that project for usage instructions.
Tasks
This model is trained to perform 3 tasks related to semantic frame parsing:
- Identify frame trigger locations in the text
- Classify the frame given a trigger location
- Extract frame elements in the sentence
Performance
This model is trained and evaluated using the same train/dev/test splits from FrameNet 1.7 annotated corpora as used by Open Sesame.
| Task | F1 Score (Dev) | F1 Score (Test) |
|---|---|---|
| Trigger identification | 0.75 | 0.71 |
| Frame Classification | 0.87 | 0.86 |
| Argument Extraction | 0.76 | 0.73 |