Font Size: a A A

Research On Cross-Lingual Semantic Dependency Graphs

Posted on:2022-09-16Degree:MasterType:Thesis
Country:ChinaCandidate:Y Q JiFull Text:PDF
GTID:2518306572960169Subject:Software engineering
Abstract/Summary:PDF Full Text Request
With the development of natural language processing technology,the use of natural language processing is more widespread for example,in applications such as chatbots,intelligent search and intelligent recommendations,all of which make use of natural language processing technology.In addition,communication between countries is becoming more frequent and the need for deep semantic understanding of various languages is increasing.The semantic dependency graph analysis task is one of the tasks set to address this need.Semantic dependency graph analysis is a graphical way of ordering semantic information,defining pairs of semantic units through dependency arcs and labels,so that questions such as when,where,what,why and who can be answered directly.We have defined a new cross-lingual semantic dependency graph representation scheme that can annotate multiple languages under the same system,providing more precise semantic information for downstream tasks.We also contribute a highprecision dataset on Chinese and English under this scheme,with 500 sentences in Chinese and 1000 sentences in English..In addition,we propose graph transformation algorithms based on two strategies,a pseudo-data enhancement algorithm based on Label Switching and an algorithm based on graph neural attention networks to solve the costly and difficult annotation problem.With the help of the powerful representational of the pre-trained language model and the efficient modelling capability of these two strategies,we can quickly and accurately convert existing large-scale graph(tree)datasets into annotations under this scheme,thus enabling the rapid construction of multilingual corpora and laying the foundation for subsequent research work.
Keywords/Search Tags:Semantic dependency graphs, graph transformation algorithms, data augmentation, graph networks
PDF Full Text Request
Related items