ClaimFlow: Tracing the Evolution of Scientific Claims in NLP

The article introduces ClaimFlow, a claim-centric framework for analyzing scientific claims in NLP, derived from 1,617 ACL Anthology papers annotated with 5,689 claims and 4,871 cross-paper claim relations. It establishes a new task called Claim Relation Classification, achieving a baseline performance of 0.81 macro-F1 using neural and large language models, indicating its feasibility while highlighting areas for improvement. This framework enables practitioners to trace the evolution of claims in NLP research, revealing that a significant majority of claims remain unchallenged, thus providing insights into the dynamics of scientific discourse in the field.