Research On Unsupervised Named Entity Recognition Based On Cross Language Pretrained Model

With the methods of information technology and artificial intelligence,natural language processing technology is more and more used in our daily life.Now,the main way to deal with language data is to monitor the quality of language learning.There are more than 6000 languages in the world today,and less than 1% of them have enough corpus.In order to apply natural language processing technology to more languages,so that users of different languages can also enjoy the bonus of artificial intelligence,cross language migration technology has gradually attracted attention.This technology makes use of the similarity and sharing features of languages,makes use of the abundant source language corpus,and does not use or use a small amount of data of the target language as the training data training model.By learning the common features between languages,the model can be applied to the target language without corpus.In this paper,translation alignment technology and confrontation learning technology are used to study cross language named entity recognition.(1)By using the method of translation,the entity recognition corpus is named according to the source language to generate the raw materials of the target language.This paper explores the influence of translation alignment method to generate data on the cross language ability of the pre trained model.(2)This paper explores the effect of confrontation learning in cross language transfer learning.This paper proposes two methods of language confrontation learning: word level language confrontation learning,word order language confrontation learning.Finally,the data set of connl ner in three languages(German,Spanish,Dutch)has been improved.(3)The relationship between the ability of cross language transfer and the similarity of source language and target language is explored.After the above research,this project explores the influence of confrontation learning on cross language pre training model,and improves the effect of cross language named entity recognition.In addition,this paper explores the language distance between the source language and the target language,and its relationship with the transfer effect.