The naming-telling relationship is an important grammatical method for combining clauses into clause corhplexes, and it is also an important basis for Chinese text-level syntactic and semantic analysis. In this paper, by introducing the window sliding mechanism, the text and its component sharing relationship are transformed into the prediction problem of the text segment and the component sharing relationship within the segment. At the same time, in view of the problem of combining and selecting prediction results, this paper proposes a variety of candidate elimination strategies based on the grammatical limitation of the sharing relationship between words and phrases. The experimental results show that the method in this paper still achieves comparable experimental results with the traditional NTC-based methods in the absence of the boundary information of the clause complex, especially the recall rate at the to-be-predicted position where the shared components are indeed missing is improved by about 0.4 points.
小句复合体；边界；滑动窗口；预训练语言
罗智勇、韩瑞昉、张明明、韩玉蛟、赵志琳
展开 >
北京语言大学
小句复合体；边界；滑动窗口；预训练语言
Chinese national conference on computational linguistic
Nanchang(CN)
The 21st Chinese national conference on computational linguistic
NETL
