Distant Supervision For Chinese Semantic Role Labeling

Semantic role labeling (SRL) is dedicated to recognizing the semantic predicate-argument structure of a sentence. Previous works on SRL are mostly based on public datasets. However, existing public SRL datasets have limited annotated data, which restricts the generalization performance of the model and their downstream applications. Manually labeling data is time-consuming and labor-intensive. Inspired by the task of distant supervision for relation extraction, this paper brings in external language knowledge: collocation table and semantic dependency data, and finally obtained 82,182 high-quality synthetic data containing 319,302 predicate-argument-role pairs based on chunk dependency parsing and syntactic rules, and divided semantic roles into 27 types, the accuracy of the synthetic data is 89.33% by manual judgement, which provides new data and ideas for Chinese semantic role labeling tasks. Based on this data, use bert [1] to build a semantic role labeling model to further examine data quality and generalization performance.