This video is a talk I delivered at the 2021 Competition on Legal Information Extraction/Entailment (COLIEE), where I represented the Neuralmind team, which secured the first position in the competition. The COLIEE was organized in conjunction with the International Conference on Artificial Intelligence and Law (ICAIL) in 2021.

Remarkably, our submissions obtained superior scores, with a lead of over six percentage points over the next highest scoring team. These results substantiated a counter-intuitive revelation: models demonstrating minimal or no adaptation to the target task are potentially more robust to variations in data distribution than those fine-tuned on it, particularly when confronted with limited labeled data.

Paper

You can read the paper associated with this talk on arxiv: To Tune or Not To Tune? Zero-shot Models for Legal Case Entailment