Research On Cross-lingual Sentence Summarization

At present,information overload is a common issue on the Internet,while auto-matic text summarization can effectively improve this trouble.However,current work on automatic text summarization is mainly focused on monolingualism,that is,the origin text and the generated summary belong to the same language.This method has some limitations,which greatly hinders people's ability to retrieve foreign lan-guage text.In response to this problem,this work focuses on cross-lingual sentence summarization,which generate summaries in foreign languages from the source text Implementing this technology is mainly through the following three methods?.The cross-lingual sentence summarization model based on "teacher-student"framework.Cross-lingual sentence summarization task is a zero-shot learning task.We build pseudo corpus by parallel corpus of machine translation,and parallel data of monolingual sentence summarization.This pseudo data is used to train the end-to-end cross-lingual sentence summarization model,which is defined as the "student”model.Using parallel data of monolingual sentence summarization to train monolingual sen-tence summarization model is defined as "teacher-summarization" model.The word probability distribution output by the "teacher-summarization" model can provide effective supervision information to the "student" model,so that the "student" model can overcome the defects brought by the pseudo corpus.?.The cross-lingual sentence summarization based on relay attention mechanism.In encoder-decoder architecture,attention alignment accuracy seriously affects the performance of the model.Pseudo corpus is formed by translating the parallel data of monolingual summarization using a machine translation model,so attention alignment information in the monolingual sentence summarization model is transferred to the machine translation model,thus forming cross-lingual attention alignment information and providing effective attention alignment monitoring information for the cross-lingual sentence summarization model.?.The cross-lingual sentence summarization based on contrastive attention mechanism.Attention mechanism plays an important role in the encoder-decoder structure.On this basis,we added a contrastive attention mechanism to get the ir-relevant information between the target end and the source end through opponent attention,while the original attention mechanism(positive attention)gets the most relevant information between the target end and the source end.The two attentions combine to distinguish relevant and irrelevant information in the source text and jointly generate the summary text.