Using Next Sentence Prediction for Chinese Frame Disambiguation

Chinese frame disambiguation is a task that identifies the correct frame evoked by the target word in a given sentence. This task can effectively address the issue of polysemy in Chinese frame semantic parsing. This task is generally regarded as a multi-class classification task, where frames are typically treated as discrete labels. However, due to the issue of class imbalance in Chinese frame disambiguation, it is not suitable to be treated as a multi-class classification task. By using the definitions of frames in Chinese FrameNet, we construct sentence pairs of “Target Word Context - Frame Definition”. We then reformulate the Chinese frame disambiguation task as a binary classification task by leveraging the Next Sentence Prediction (NSP). The experimental results of blocked cross-validation show that our method outperforms previous methods.