TLDR: Extreme Summarization of Scientific Documents
This addresses the challenge of quickly summarizing complex scientific documents for researchers, though it is incremental as it builds on existing summarization tasks with a new dataset and method.
The authors tackled the problem of generating extreme summaries (TLDRs) for scientific papers, which requires high compression and expert knowledge, by introducing SciTLDR, a dataset of 5.4K TLDRs from 3.2K papers, and CATTS, a learning strategy that improves performance over baselines in automated and human evaluations.
We introduce TLDR generation, a new form of extreme summarization, for scientific papers. TLDR generation involves high source compression and requires expert background knowledge and understanding of complex domain-specific language. To facilitate study on this task, we introduce SciTLDR, a new multi-target dataset of 5.4K TLDRs over 3.2K papers. SciTLDR contains both author-written and expert-derived TLDRs, where the latter are collected using a novel annotation protocol that produces high-quality summaries while minimizing annotation burden. We propose CATTS, a simple yet effective learning strategy for generating TLDRs that exploits titles as an auxiliary training signal. CATTS improves upon strong baselines under both automated metrics and human evaluations. Data and code are publicly available at https://github.com/allenai/scitldr.