Papers by Diego Rossini
Binary Token-Level Classification with DeBERTa for All-Type MWE Identification: A Lightweight Approach with Linguistic Enhancement (2026.findings-eacl)
Copied to clipboard
| Challenge: | Current approaches focus on specific MWE types, such as transformer-based models that incorporate linguistic features like dependency parsing for verbal discontinuous patterns. |
| Approach: | They propose a binary token-level classification approach that integrates linguistic feature integration and data augmentation to improve multiword expression (MWE) identification. |
| Outcome: | The proposed model outperforms the Qwen-72B model on the CoAM dataset by 12 points while using 165 times fewer parameters. |