Graph-to-Graph Annotation Conversion Based on Pretrained Models

Anonymous

Abstract: Annotation conversion is an effective way to construct datasets under new annotation guidelines based on existing datasets with little human labour. Previous work has been limited in conversion between tree-structured datasets and mainly focused on feature-based models which are not easily applicable to new conversion. In this paper, we propose two pretrained model-based graph-to-graph annotation conversion approaches, namely Label Switching and Graph2Graph Linear Transfer, which are able to deal with conversion between graph-structured annotations and require no manually designed feature. We manually construct a graph-structured parallel annotated dataset and evaluate the proposed approaches on it as well as four existing parallel annotated datasets. Experimental results show that the proposed approaches outperform two strong baselines across all the datasets. Furthermore, the combination of the two models have a better effect.