🎓 This work is adapted from Isabel Sebire MSc dissertation, which was awarded the MSc AI Dissertation Prize and the Claire Jones Prize at the University of Edinburgh. (5/5)

- We show that our content selection method increases textual entailment by +26%, and improves summary quality by up to +5.56 ROUGE-1.
- We offer a detailed qualitative analysis of failure cases and hallucination sources, laying the groundwork for future improvements. (4/5)

🔍 Our contributions:
We introduce a content selection pipeline that identifies salient legal information prior to summarisation, improving both the faithfulness and quality of summaries. (3/5)

Our findings provide evidence that content selection is an important step in making abstractive summarisation reliable in high-stakes domains. (2/5)