Negation Scope Conversion for Unifying Negation-Annotated Datasets

Negation scope resolution is a technique that identifies the part of a sentence affected by the negation cue. The three major corpora used for it, the BioScope corpus, the SFU review corpus, and the Sherlock dataset, have different annotation schemes for negation scope. Due to the different annotations, it is difficult to use the three corpora together in the study of negation scope resolution. To address this issue by merging the corpora into a unified dataset based on a common annotation scheme, we propose a method for automatically converting the scopes of BioScope and SFU to those of Sherlock. We conducted an experiment to evaluate the accuracy of our method using a dataset obtained by manually annotating the negation scopes to a tiny portion of BioScope and SFU, verifying that our method can convert the scopes with high accuracy. In addition, we conducted another experiment to verify the effectiveness of our method from a pragmatic perspective, where we fine-tuned PLM-based negation scope resolution models using the unified dataset obtained by our method. The results demonstrated that the performances of the models increase when fine-tuned on the unified dataset, unlike the simply combined one, which supports the effectiveness of our method.